THE AI TOOLKIT / SKILLS

Skills.
Useful instructions. Inspected in context.

Give your AI a repeatable way to work. Explore selected skills with original review notes, practical installation guidance and the source material behind each choice.

Selected for a reason. Reviewed within a stated scope. Every entry includes our own findings and limitations. Source review is not a runtime test, a security certificate or proof of a universal “best.”

01 / START WITH THE JOB

What are you working on?

Explore a category, then refine your shortlist.
All categories are available in the filters.

PLAN THE HANDOFF

Put the resources to work.

Explore 88 workflows

Task briefs, working sheets and acceptance checks. Suggested plans; each resource needs its own setup.

BUILT BY UNDOMINATED

Evidence tools you can inspect.

Original source, MIT licence and complete downloads.
Each review names the checks performed and their limits.

02 / THE SELECTION

Find a fit. Read the reasoning.

Alphabetical by default. Every match is shown.
Compare up to three before opening a full review.

Runtime, access and execution

Labels summarise reviewed source wording. A local process can still send data to external services. Read each review's prerequisites and permissions; these filters do not certify compatibility or enforce access.

759 of 759 selected skills

Languages and tooling · GitHub community

.NET Timezone

Resolves .NET timezone work through the right library first — TimeZoneConverter for cross-platform, NodaTime for recurring schedules and strict DST arithmetic — and ships both Windows and IANA identifiers on every answer, seven copy-paste code patterns, and the pitfall list from DateTimeKind.Unspecified to DST transitions that skip or repeat local times.

timezonedatetimedotnet

Why we selected itRoutes every request through the right library before any code is written — TimeZoneConverter for cross-platform work, NodaTime when recurring schedules or strict DST rules are in play — instead of letting agents reach for TimeZoneInfo on the wrong platform.

Weigh up.NET-specific; the discipline transfers to other stacks but the APIs and libraries do not.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Data and analytics · dbt Labs

Adding a dbt unit test

Writes a dbt unit test that mocks upstream inputs instead of querying production.

dbttestingwarehouses

Why we selected itThe description requires mocked upstream inputs and expected outputs.

Weigh upWarehouse-specific type rules live in the skill references and were not executed here.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Development workflow · addyosmani

Adversarial Decision Review

Submit a bounded artifact and its contract to an independent reviewer before accepting a consequential decision.

development-workflowaddyosmani

Why we selected itThe definition separates CLAIM from reviewer input, requires artifact-plus-contract only and warns against recursive persona orchestration.

Weigh upAn adversarial prompt does not establish independence or expertise. The skill requires host delegation; self-questioning is explicitly a degraded fallback.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Testing · Trail of Bits

AFL++

Runs AFL++ as a multi-core campaign and forbids afl-system-config on a production or development machine because that script needs root and turns OS protections off.

fuzzingaflc

Why we selected itafl-system-config and afl-persistent-config are called out as root scripts that disable OS security features.

Weigh upThe security warning says not to fuzz on production systems or the development environment.

Source reviewed

2 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Agent development · Anthropic

Agent Development for Claude Code Plugins

Explains how to write a Claude Code plugin agent file (frontmatter, trigger description, system prompt) and ships a Bash script that checks one agent file.

claude-codeplugin-devagents

Why we selected itIt sets the agent frontmatter rules: a name of 3-50 lowercase letters, numbers and hyphens, a required description, model (inherit, sonnet, opus or haiku) and color, and an optional tools list it says to keep to the minimum needed.

Weigh upLast commit touching plugins/plugin-dev/skills/agent-development at revision 41f32e40: 2026-04-28 (ce721c1f).

Source reviewed

6 sources · 2026-10-21

Read the full review Upstream setup linkedApache-2.0

Model evaluation · Google

Agent Platform Eval Flywheel

Evaluates and iterates on Google Cloud Agent Platform models and agents, with a confirmation card before every billed evaluation run.

google-cloudevaluationllm-as-judge

Why we selected itEvaluation, inference and scenario-generation calls need their own Yes/No confirmation for every run, asked in one turn and executed only after the reply.

Weigh upSetup tells the agent not to create a virtual environment and to pip install missing packages into the current environment.

Source reviewed

4 sources · 2026-10-13

Read the full review Install command in reviewApache-2.0

AI application development · Google

Agent Platform GenAI Inference

Calls Gemini and third-party OpenMaaS models on Google Cloud Agent Platform, asking for model, project and region instead of defaulting.

google-cloudinferenceopenai-compatible

Why we selected itThe file says never to guess, volunteer or default to a model, and to ask for the project ID and region rather than assume them.

Weigh upRegion availability is checked with a live generateContent call to the exact model and region, which the file requires before answering region-specific requests.

Source reviewed

4 sources · 2026-10-13

Read the full review Install command in reviewApache-2.0

Kubernetes model serving · Microsoft

AI Runway AKS setup

Pins Microsoft's AI Runway AKS setup skill, which requires confirmation before install, deployment, or GPU provisioning.

microsoftaksai-runway

Why we selected itNo selected skills record uses this source path.

Weigh upThis review did not execute the resource.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Cloud operations · Microsoft

AKS Automatic Readiness

Pins Microsoft's AKS Automatic readiness skill, which assesses a cluster or local manifests without changing the cluster and writes files only after an accepted diff.

azureakskubernetes

Why we selected itThe first guardrail says the assessment never modifies cluster state and names kubectl apply and az aks update as forbidden.

Weigh upThe resource was not executed, and no Azure API, CLI, or MCP tool was called.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

AKS GPU diagnosis · Microsoft

AKS GPU inference

Pins Microsoft's AKS GPU and KAITO diagnosis skill, which stays on read-only evidence unless scaling, drain, deletion, or add-on changes are separately authorized.

aksgpukaito

Why we selected itThe file requires the host to authorize shell, kubectl, and az against the bound target before any referenced pipeline, and it forbids routing kubectl through Azure MCP or around host policy.

Weigh upThe resource was not executed.

Source reviewed

6 sources · 2026-10-15

Read the full review Upstream setup linkedMIT

AKS incident matching · Microsoft

AKS known issues

Pins Microsoft's AKS known-issue skill, which refuses a match unless every catalog qualifier is present and requires approval before any mutation.

aksknown-issuesapproval

Why we selected itThe file says not to use the skill unless every qualifier for a catalog row is present, and it says codes are not matched as substrings.

Weigh upThe resource was not executed.

Source reviewed

6 sources · 2026-10-15

Read the full review Upstream setup linkedMIT

AKS packet capture · Microsoft

AKS network capture

Pins Microsoft's AKS packet-capture skill, which will not claim a capture ran unless the host gate passes and which separates capture intent from debug-container consent.

akspacket-captureauthorization

Why we selected itThe file says a governed Azure CLI tool does not authorize shell, kubectl, artifact storage, or packet capture, and it forbids routing kubectl through Azure MCP or adding an unapproved execution path.

Weigh upThe resource was not executed.

Source reviewed

6 sources · 2026-10-15

Read the full review Upstream setup linkedMIT

AKS incident diagnosis · Microsoft

AKS troubleshooting

Pins Microsoft's live AKS troubleshooting skill, which does not restart, delete, cordon, drain, scale, upgrade, or reconfigure a resource unless remediation is explicitly requested.

akstroubleshootingread-only

Why we selected itThe file forbids restart, delete, cordon, drain, scale, upgrade, and reconfigure unless the user explicitly asks, and it says to propose a fix without applying it uninvited.

Weigh upThe resource was not executed.

Source reviewed

6 sources · 2026-10-15

Read the full review Upstream setup linkedMIT

Security review · Trail of Bits

Algorand vulnerability scanner

Pins Trail of Bits' Algorand scanner, which treats a short or empty coverage table as an unfinished scan and does not let a clean Tealer run stand in for patterns it never examined.

algorandtealpyteal

Why we selected itA coverage table with a missing row or an empty verdict must be reported as an incomplete scan, and not-applicable requires a reason rather than silence.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Creative production · Anthropic

Algorithmic Art

Creates original generative art with p5.js through a two-step method: author an algorithmic philosophy as a named movement with a manifesto, then express it as seeded-randomness sketches with an interactive parameter viewer.

p5jsgenerative-artcreative-coding

Why we selected itSeparates the aesthetic idea from its expression: an algorithmic philosophy is authored first as a named movement with a manifesto, and only then expressed as p5.js sketches that are 90% algorithm, 10% parameters.

Weigh upAesthetics and process guidance, not a rendering-performance reference for large particle counts or shader work.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

AI application development · Amazon Web Services

Amazon Bedrock

Bedrock model invocation, Knowledge Bases, Guardrails and AgentCore guidance, with quota, prompt-caching and cost-tracking workflows.

bedrockawsmodel-invocation

Why we selected itIt requires setting maxTokens explicitly on every Converse or InvokeModel call, because an unset value defaults to the model maximum and reserves far more quota, a cause of ThrottlingException named in the file.

Weigh upMost procedures live in the references directory, which was not opened; the file says to read the relevant reference before responding.

Source reviewed

5 sources · 2026-10-13

Read the full review Install command in reviewApache-2.0

Data engineering · dbt-labs

Answer Business Questions with dbt

Answer warehouse questions through existing semantic metrics before resorting to carefully scoped SQL.

data-engineeringdbt-labs

Why we selected itIt answers a business question from the semantic layer before it writes SQL, and it stops suggesting changes at the semantic model.

Weigh upThe skill is not for building, testing, or running dbt models.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

AI application development · Anthropic

Anthropic Claude API

Routes Claude API work to language-specific SDK guidance and distinguishes self-hosted tool loops from managed agent execution.

anthropicclaudeapi

Why we selected itExplains the operational differences among a plain API call, a tool runner, managed agents and the separate Agent SDK.

Weigh upThis is Anthropic-specific and its instructions can strongly influence model, SDK and architecture choices.

Source reviewed

8 sources · 2026-10-14

Read the full review Install command in reviewApache-2.0

Research and documentation · Anthropic

Anthropic Documentation Coauthoring

Guides a human and agent through audience discovery, section-by-section drafting and a separate-reader review of a substantial document.

documentationrfcspecifications

Why we selected itStarts with the audience, intended impact and missing context, then lets the author curate ideas before the agent drafts each section.

Weigh upReader-model agreement is a spot check, not factual verification or a human usability study; the author still needs to verify facts and links.

Source reviewed

3 sources · 2026-10-15

Read the full review Upstream setup linkedNot established

Design and interfaces · Anthropic

Anthropic Frontend Design

Builds a visual direction from the product’s audience and subject, then reviews typography, layout and interface copy against that brief.

designtypographyfrontend

Why we selected itRequires a compact design plan and a critique of generic choices before implementation.

Weigh upThis is an opinionated design process, not a component library, usability study or automated accessibility audit.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Documents · Anthropic

Anthropic PDF Processing

Separates PDF extraction, actual form-field filling and image-coordinate text annotations, with explicit field maps and rendered-output review.

pdfformsocr

Why we selected itField discovery precedes filling, with field IDs, one-based page numbers and type-specific allowed values carried into the handoff.

Weigh upcheck_bounding_boxes.py reads authored JSON, not the PDF. It prints FAILURE messages without a failure exit status, caps diagnostics and does not validate page boundaries, text width or accessibility.

Source reviewed

10 sources · 2026-10-15

Read the full review Install command in reviewAnthropic custom terms

Builds editable decks or adapts a template, separating content checks, structural validation and full-resolution visual review.

powerpointpptxslides

Why we selected itSeparates editable native charts, source content and template structure rather than treating a slide image as an editable deck.

Weigh upLibreOffice font substitution can disagree with PowerPoint. Neither an XML pass nor a thumbnail grid proves readable full-size output or native PowerPoint fidelity.

Source reviewed

8 sources · 2026-10-15

Read the full review Install command in reviewAnthropic custom terms

Agent development · Anthropic

Anthropic Skill Creator

Develops skills from realistic task fixtures and baseline comparisons, with separate trigger-description experiments and human output review.

skillsevaluationtriggering

Why we selected itSeparates activation experiments from task-output quality, preserving both a baseline and human feedback.

Weigh upFull comparisons require supported subagents and Claude CLI; reduced Claude.ai paths explicitly omit baseline benchmarking. A source review supplies neither benchmark results nor installation proof.

Source reviewed

9 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Documents · Anthropic

Anthropic Spreadsheets

Produces editable, formula-driven workbooks from specified inputs, with separate recalculation receipts and business-logic checks.

excelxlsxformulas

Why we selected itDistinguishes formula expressions from cached calculated values, so a workbook handoff can include both rather than silently replacing formulas.

Weigh upRecalculation rewrites the workbook in place. Retain the original: --force can accept destructive external-link loss, and link detection is not exhaustive.

Source reviewed

7 sources · 2026-10-15

Read the full review Install command in reviewAnthropic custom terms

Browser testing · Anthropic

Anthropic Webapp Testing

Guides local browser checks with Playwright and includes a wrapper that starts development servers around an automation command.

playwrightbrowserlocal-development

Why we selected itUses rendered-page reconnaissance before choosing selectors for dynamic applications.

Weigh upThe server helper checks only whether a TCP port accepts a connection, not whether the expected application is healthy.

Source reviewed

8 sources · 2026-10-14

Read the full review Install command in reviewApache-2.0

Documents · Anthropic

Anthropic Word Documents

Separates new Word-document generation from editing existing OOXML, with explicit checks for tables, comments and tracked changes.

worddocxtracked-changes

Why we selected itDocuments the different create, edit and read paths instead of treating Word files as plain text.

Weigh upRestricted Anthropic terms apply; this is a source-available reference, not an open-source document package.

Source reviewed

8 sources · 2026-10-15

Read the full review Install command in reviewAnthropic custom terms

Development workflow · Addy Osmani

API and Interface Design

Designs public interfaces so the right call is easy and a breaking change is explicit.

apicontractsdesign

Why we selected itThe skill treats every observable behaviour as a commitment and tells the author to define the contract before the implementation.

Weigh upError strategies must not be mixed across endpoints.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

API inventory · Microsoft

API Center management SDK

Pins Microsoft's API Center Python skill, which requires DefaultAzureCredential and rejects connection strings and API keys for the management client.

api-centerazuredefault-azure-credential

Why we selected itThe file says to prefer DefaultAzureCredential and to avoid connection strings, account keys, and API keys because they bypass Entra audit and rotation.

Weigh upThe resource was not executed.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

API management · Microsoft

API Management SDK

Pins Microsoft's API Management Python skill, which requires DefaultAzureCredential and rejects connection strings and API keys for the management client.

api-managementazuredefault-azure-credential

Why we selected itThe file says to prefer DefaultAzureCredential for the management client and to avoid connection strings, account keys, and API keys.

Weigh upThe resource was not executed.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Model evaluation · Arize AI

Arize Evaluator

Creates and runs Arize LLM-as-judge and code evaluators through the ax CLI, with a rule against fabricating scores.

arizellm-as-judgeevaluation

Why we selected itIf an evaluation task fails, is cancelled or produces no scores, the file says to report the failure and not run a manual evaluation, invent scores or estimate percentages.

Weigh upThe file tells the agent to run ax commands directly without checking versions, environment variables or profiles first. The four reference files it links were not opened.

Source reviewed

5 sources · 2026-10-13

Read the full review Install command in reviewMIT

Observability · Arize AI

Arize Instrumentation Health

Read-only audit of exported Arize spans for nine instrumentation defects, including missing token counts behind empty cost dashboards.

arizeopeninferencetracing

Why we selected itThe audit is read-only by default and treats exported span content as untrusted data, never following instructions found inside span attributes.

Weigh upSpan export is delegated to the arize-trace skill, which was not reviewed; this skill does not export spans on its own.

Source reviewed

6 sources · 2026-10-13

Read the full review Install command in reviewMIT

Languages and tooling · Astral

Astral ruff

Astral's skill for linting, fixing and formatting Python with ruff: when to use it, how to invoke it, which changes to avoid, and how Black, Flake8 and isort commands map onto ruff.

astralpythonruff

Why we selected itIt is Astral's own skill for its tool, shipped in Astral's claude-code-plugins marketplace, and it says which form to run: uv run ruff when ruff is a project dependency, so the pinned version is used, uvx ruff for one-off checks, or a global ruff.

Weigh upLast commit touching plugins/astral/skills/ruff at revision f3ce88a7: 2026-02-27 (f3ce88a).

Source reviewed

5 sources · 2026-10-21

Read the full review Upstream setup linkedMIT OR Apache-2.0

Languages and tooling · Astral

Astral ty

Astral's skill for type-checking Python with ty: invocation through uv, ty check options, [tool.ty] configuration with per-file overrides, mypy and Pyright command mappings, and rule-specific ignores.

astralpythontyping

Why we selected itIt is Astral's own skill for ty, from astral-sh/claude-code-plugins, covering ty check on a directory or file, rule severity flags (--error, --warn, --ignore), Python version and platform targeting, and [tool.ty] configuration with [[tool.ty.overrides]] for tests.

Weigh upLast commit touching plugins/astral/skills/ty at revision f3ce88a7: 2026-02-27 (f3ce88a).

Source reviewed

5 sources · 2026-10-21

Read the full review Upstream setup linkedMIT OR Apache-2.0

Package Management · Astral

Astral uv Python Workflow

Establishes disciplined uv workflows across standalone scripts, lockfile-managed projects, and ephemeral CLI tool execution.

astraluvpython

Why we selected itThe skill picks one uv workflow from scripts, projects, tools, or the pip interface instead of mixing them.

Weigh upThe skill says not to use uv in projects managed by other tools, not to call pip inside a uv project, and not to run python directly there.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewMIT

Testing · Trail of Bits

Atheris

Sets up a Python fuzz harness, including C extensions, instead of treating a unit test as a fuzzer.

fuzzingpythonatheris

Why we selected itThe description names TestOneInput and AddressSanitizer.

Weigh upCC-BY-SA-4.0 applies to an adaptation you redistribute.

Source reviewed

2 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Security review · Trail of Bits

Audit Augmentation

Overlays external findings — SARIF static-analysis results, weAudit annotation files and Trailmark 0.4.x binary-analysis exports — onto a Trailmark code graph by file and line overlap and builds severity-based subgraphs.

sarifstatic-analysisgraph

Why we selected itMaps SARIF results, weAudit annotations and v0.4 binary-analysis exports onto Trailmark graph nodes by file and line overlap and builds severity-based subgraphs.

Weigh upBinary-graph import (engine.augment_binary) needs Trailmark 0.4.0+; SARIF and weAudit import are v0.2-safe.

Source reviewed

5 sources · 2026-10-22

Read the full review Upstream setup linkedCC-BY-SA-4.0

Development tooling · trailofbits

Authenticated GitHub CLI Workflow

Prefer authenticated GitHub CLI reads and scoped clones when inspecting GitHub resources.

development-toolingtrailofbits

Why we selected itThe hook denies an entire batched fetch when one GitHub URL matches; the shim blocks contents API patterns and some unscoped temporary clones, while cleanup validates the session identifier.

Weigh upInstalling the skill directory alone does not activate the separate Claude plugin hooks or PATH shim. Hook matching covers selected tools and URL forms; it is not a complete network policy. Session cleanup recursively removes its clone directory.

Source reviewed

9 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Infrastructure development · HashiCorp

AWS AMI builder

Builds an AMI with Packer's amazon-ebs builder, filters the source AMI instead of hardcoding one, and warns that the build spends AWS money.

packeramiaws

Why we selected itThe template pins the hashicorp/amazon plugin and selects the newest Canonical Ubuntu AMI with an owner filter rather than an AMI id.

Weigh upThe skill does not build an image for a cloud other than AWS.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewMPL-2.0

Content moderation · Microsoft

Azure AI Content Safety for Python

Pins Microsoft's Python Content Safety skill, which keeps new code on Entra credentials and can stop analysis when a blocklist matches.

content-safetymoderationblocklist

Why we selected itNew code is told to use DefaultAzureCredential, and AzureKeyCredential is limited to an existing keyed deployment that has not been migrated to Entra ID.

Weigh upThis review did not execute the resource.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Microsoft's TypeScript guide to the @azure-rest/ai-content-safety REST client: scoring text and images for hate, sexual, violence and self-harm severity, managing custom blocklists, and folding the results into one allow-or-block decision.

content-safetymoderationblocklist

Why we selected itIt warns that ContentSafetyClient is a function, not a class, and checks isUnexpected() after each call whose response body it reads.

Weigh upBoth install lines Microsoft prints for this skill, `npx skills add microsoft/skills --skill azure-sdk-typescript` and `/plugin install azure-sdk-typescript@skills`, install the whole azure-sdk-typescript plugin rather than the Content Safety skill alone.

Source reviewed

6 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Content moderation · Microsoft

Azure AI Content Safety SDK for Java

Microsoft's Java guide to com.azure:azure-ai-contentsafety at 1.1.0-beta.1: text and image harm analysis with ContentSafetyClient, and blocklist management with BlocklistClient.

content-safetymoderationblocklist

Why we selected itIt pairs ContentSafetyClient for analysis with BlocklistClient for blocklist create or update, list, get, item add, item list, item remove and delete.

Weigh upBoth install lines Microsoft prints for this skill, `npx skills add microsoft/skills --skill azure-sdk-java` and `/plugin install azure-sdk-java@skills`, install the whole azure-sdk-java plugin rather than the Content Safety skill alone.

Source reviewed

6 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Document translation · Microsoft

Azure AI Document Translation for Python

Pins Microsoft's Python document-translation skill, which uses Entra for new clients and limits blob SAS permissions on source and target containers.

document-translationblob-storageentra-id

Why we selected itThe file says new code should use DefaultAzureCredential, avoid API keys, and set AZURE_TOKEN_CREDENTIALS in production to constrain the credential chain.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Language understanding · Microsoft

Azure AI Language Conversations for Python

Pins Microsoft's Python conversation-analysis skill, which tells new code to use token credentials and reserves API keys for unmigrated keyed deployments.

conversational-language-understandingentra-idazure-ai

Why we selected itThe file says new code should use DefaultAzureCredential, and AzureKeyCredential only for an existing keyed deployment that has not been migrated to Entra ID.

Weigh upThis review did not execute the resource.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Cloud development · microsoft

Azure AI Search for Python

Create vector-capable indexes and combine keyword, vector and semantic queries.

cloud-developmentmicrosoft

Why we selected itNew code is told to use DefaultAzureCredential. An API key is only for a deployment that has not moved to Entra yet.

Weigh upsetup_vector_index.py creates or updates an index in the named service. It does not upload documents.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Text analytics · Microsoft

Azure AI Text Analytics for Python

Pins Microsoft's Python text-analytics skill, which keeps new code on DefaultAzureCredential and reserves API keys for unmigrated keyed deployments.

text-analyticsentra-idpython

Why we selected itThe file says to prefer DefaultAzureCredential and to avoid connection strings and API keys because they bypass Entra audit and rotation.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Text translation · Microsoft

Azure AI Text Translation for Python

Pins Microsoft's Python text-translation skill, which requires a custom subdomain for token auth and keeps the global endpoint on key plus region.

text-translationentra-idpython

Why we selected itThe file says token-credential auth requires a custom subdomain, and that a global endpoint must stay on the key-based path until a custom subdomain exists.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Speech transcription · Microsoft

Azure AI Transcription for Python

Pins Microsoft's Python transcription skill, which prefers DefaultAzureCredential in production and says never to hardcode credentials.

speech-to-textentra-idpython

Why we selected itThe file says two auth modes are supported, and that production code should prefer DefaultAzureCredential and never hardcode credentials.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Pins Microsoft's Python image-analysis skill, which keeps new code on DefaultAzureCredential and states image size and dimension limits.

image-analysisentra-idpython

Why we selected itThe file says new code should use DefaultAzureCredential, avoid API keys, and constrain the production credential chain with AZURE_TOKEN_CREDENTIALS.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Realtime voice · Microsoft

Azure AI Voice Live for Python

Pins Microsoft's Python Voice Live skill, which uses async Entra credentials and says not to pair this SDK with sync clients.

voicewebsocketentra-id

Why we selected itThe file says the SDK is async-only and not to pair it with sync clients from other Azure SDKs in the same call path.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Azure SDKs · microsoft

Azure App Configuration for Python

Review key-value, label, feature-flag and snapshot operations for Azure App Configuration.

azure-sdksmicrosoft

Why we selected itThe retained reference scopes a snapshot by key and label and closes both async credential and client. Provider-based refresh and feature evaluation are separate application dependencies.

Weigh upThe SDK stores flag configuration; it does not by itself evaluate targeting rules or refresh every application. azure-identity is an additional dependency, and require_envvar=True requires the credential-selection variable even in local development.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Azure deployment · Microsoft

Azure app onboard deploy

Pins the app-onboard deploy phase, which requires a what-if preview and an approval gate before it changes Azure, then says to stop.

azuredeploymentwhat-if

Why we selected itPreflight requires authentication, a mandatory what-if preview, RBAC, and the resource group before the deploy approval gate.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Deployment readiness · Microsoft

Azure app onboard prereq

Pins Microsoft's Azure app-onboard prerequisite skill, which forbids package-manager install, build, and test commands except a later per-command consent gate.

microsoftazurereadiness

Why we selected itNo selected skills record uses this source path.

Weigh upThis review did not execute the resource.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Cloud development · microsoft

Azure Blob Storage for Python

Upload, stream and manage blobs using explicit overwrite, metadata and delegated-access choices.

cloud-developmentmicrosoft

Why we selected itShared access signatures in this skill are user-delegation keys tied to Entra, not account keys.

Weigh upThe samples call the storage account URL when executed. The skill does not create the storage account.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Cloud migration · Microsoft

Azure cloud migrate

Pins Microsoft's cross-cloud migration skill, which requires an assessment before code changes and forbids editing the source directory.

azuremigrationassessment

Why we selected itThe rules say to generate an assessment before any code migration and not to skip phases.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Azure compliance · Microsoft

Azure compliance and security audit

Pins Microsoft's Azure compliance skill, which requires az login before an azqr or Key Vault expiration review and separates the report from remediation.

azurecompliancekey-vault

Why we selected itThe prerequisites require az login and permission to read resource configuration and Key Vault metadata before the scan.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Review repository, tag and manifest operations with explicit deletion boundaries.

azure-sdksmicrosoft

Why we selected itThe cleanup reference checks last-updated time and empty tag lists before deleting by digest, but has no deployment-reference or rollback-artifact check.

Weigh upThe source download_manifest example uses an API renamed to get_manifest in the SDK; verify the installed version before copying it. An old untagged manifest may still be used by digest, so age/tag filters do not prove safe deletion.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Databases · microsoft

Azure Cosmos DB for Python

Use Cosmos DB NoSQL clients for partition-aware document operations and parameterized queries.

databasesmicrosoft

Why we selected itPoint reads are specified as needing both the item id and the partition key, and queries are supposed to be parameterized.

Weigh upThe skill links scripts/setup_cosmos_container.py. That path returned HTTP 404 at revision fc164ba1d7a0c10377dbcc1c9ac07253b9329fdb, so the advertised CLI is not in the tree.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Azure cost analysis · Microsoft

Azure cost analysis

Pins Microsoft's Azure cost-analysis skill, which refuses forecasts, pricing estimates, rightsizing, commitments, and budgets, and will not reshape cost results in a shell.

azurecost-analysisaks

Why we selected itThe description refuses forecasts, pricing estimates, rightsizing, commitments, and budgets, and the workflow says never to transform MCP results in a shell or interpreter.

Weigh upThe resource was not executed.

Source reviewed

6 sources · 2026-10-15

Read the full review Upstream setup linkedMIT

Azure cost forecasting · Microsoft

Azure Cost Estimation

Pins Microsoft's Azure cost-estimation skill, which separates forecasts from hypothetical prices and refuses shell transforms and mixed currencies.

azure-costforecastpricing

Why we selected itThe workflow refuses to reshape cost-tool results in a shell or interpreter and refuses to combine more than one currency.

Weigh upThis review did not execute the resource.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Azure cost optimization · Microsoft

Azure Cost Optimization

Pins Microsoft's cost-optimization skill, which may recommend a change and forbids deleting, resizing, purchasing, or deploying resources.

azure-costrightsizingcommitments

Why we selected itThe workflow limits the skill to recommendations and explicitly forbids delete, resize, purchase, and deploy.

Weigh upThis review did not execute the resource.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Cloud operations · Microsoft

Azure Deploy

Pins Microsoft's Azure deploy skill, which stops unless a separate validation skill has already marked the deployment plan validated.

azuredeploymentazd

Why we selected itThe file forbids the agent from marking a plan Validated and says to stop when that status or the Validation Proof section is missing.

Weigh upThe resource was not executed, and no Azure API, CLI, or MCP tool was called.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Cloud operations · Microsoft

Azure Enterprise Infra Planner

Pins Microsoft's enterprise Azure infrastructure skill, which holds IaC and deployment until the user approves the plan and acknowledges deploy risk.

azurebicepterraform

Why we selected itPhase 5 approval is required before IaC, and a plan that is not approved stops both generation and deployment.

Weigh upThe resource was not executed, and no Azure API, CLI, or MCP tool was called.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Publish CloudEvents or Event Grid events with a correctly scoped endpoint and client.

azure-sdksmicrosoft

Why we selected itThe namespace reference supplies namespace_topic explicitly and uses CloudEvent; the async example closes credential and publisher independently.

Weigh upPublishing acceptance does not prove delivery or exactly-once handling. The namespace example publishes events and does not implement receive/acknowledge/reject consumers. Credential-chain options depend on azure-identity version.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Cloud development · microsoft

Azure Event Hubs for Python

Publish partitioned events and consume streams with explicit checkpoint ownership.

cloud-developmentmicrosoft

Why we selected itProduction checkpointing is a separate package plus a blob container, not a feature of the producer client alone.

Weigh upSending with the helper publishes events to Azure. The skill does not create the namespace.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Cloud development · microsoft

Azure Identity for Python

Select and diagnose Entra credential chains across development, hosted workloads and delegated access.

cloud-developmentmicrosoft

Why we selected itThe package floor is Python 3.9, and the certificate class the file tells you to import is CertificateCredential, not ClientCertificateCredential.

Weigh upGetting a token calls the Entra authority for that cloud. The skill does not call the storage or compute API itself.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Cloud development · Microsoft

Azure Identity library for .NET

Microsoft's .NET guide to choosing an Azure.Identity credential: DefaultAzureCredential while developing, a specific managed identity in production, plus service principal, developer and sovereign-cloud options.

identityentra-iddefault-azure-credential

Why we selected itPractice 1 keeps DefaultAzureCredential for development and builds a ManagedIdentityCredential from a user-assigned client ID for production, and the environment-based sample switches between the two on builder.Environment.IsProduction().

Weigh upBoth install lines Microsoft prints, `npx skills add microsoft/skills --skill azure-sdk-dotnet` and `/plugin install azure-sdk-dotnet@skills`, install the whole azure-sdk-dotnet plugin, not this skill alone.

Source reviewed

6 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Cloud development · Microsoft

Azure Identity library for Rust

Microsoft's Rust guide to azure_identity, which has no DefaultAzureCredential: DeveloperToolsCredential for local work, ManagedIdentityCredential for Azure-hosted code and ClientSecretCredential for service principals.

identityentra-idauthentication

Why we selected itIt states that the Rust SDK has no DefaultAzureCredential and marks `use azure_identity::DefaultAzureCredential;` as incorrect, pointing to DeveloperToolsCredential for local development and ManagedIdentityCredential for production.

Weigh upBoth install lines Microsoft prints for this skill, `npx skills add microsoft/skills --skill azure-sdk-rust` and `/plugin install azure-sdk-rust@skills`, install the whole azure-sdk-rust plugin rather than the identity skill alone.

Source reviewed

6 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Cloud development · Microsoft

Azure Identity library for TypeScript

Microsoft's TypeScript guide to @azure/identity: DefaultAzureCredential gated on AZURE_TOKEN_CREDENTIALS, managed and workload identity, service principals, browser and device-code sign-in, and bearer-token providers.

identityentra-iddefault-azure-credential

Why we selected itPractice 1 puts DefaultAzureCredential on local development and ManagedIdentityCredential or WorkloadIdentityCredential on production, and both DefaultAzureCredential samples pass requiredEnvVars: ["AZURE_TOKEN_CREDENTIALS"].

Weigh upBoth install lines Microsoft prints for this skill, `npx skills add microsoft/skills --skill azure-sdk-typescript` and `/plugin install azure-sdk-typescript@skills`, install the whole azure-sdk-typescript plugin rather than the identity skill alone.

Source reviewed

6 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Cloud development · Microsoft

Azure Key Vault Certificates library for Rust

Microsoft's Rust guide to azure_security_keyvault_certificates: creating a self-signed X.509 certificate as a long-running operation, updating, listing and deleting certificates, and signing with a certificate's key through the Key Vault keys crate.

key-vaultcertificatesx509

Why we selected itIt shows begin_create_certificate returning a Poller that is awaited directly, with a certificate policy for subject CN=example.com and issuer Self.

Weigh upBoth install lines Microsoft prints for this skill, `npx skills add microsoft/skills --skill azure-sdk-rust` and `/plugin install azure-sdk-rust@skills`, install the whole azure-sdk-rust plugin rather than the certificates skill alone.

Source reviewed

6 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Cloud development · microsoft

Azure Key Vault for Python

Manage versioned secrets, keys and certificates while separating retrieval, cryptography and lifecycle operations.

cloud-developmentmicrosoft

Why we selected itSecret listing in the sample returns properties, and the comment says that is names only, not values.

Weigh upThe clients talk to the vault URL when the code runs. The skill does not create the vault or grant RBAC.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Cloud development · Microsoft

Azure Key Vault Keys (Java)

Microsoft's Java guide to com.azure:azure-security-keyvault-keys: creating RSA, EC and Managed HSM symmetric keys, importing key material, rotating by policy, and encrypting, signing and wrapping through CryptographyClient.

key-vaultcryptographyazure

Why we selected itIt builds sync, async and cryptography clients and creates RSA keys of 2048 and 4096 bits, EC keys on P-256, P-384 and P-521, HSM-backed variants, and a symmetric key marked Managed HSM only.

Weigh upBoth install lines Microsoft prints for this skill, `npx skills add microsoft/skills --skill azure-sdk-java` and `/plugin install azure-sdk-java@skills`, install the whole azure-sdk-java plugin rather than the Key Vault keys skill alone.

Source reviewed

6 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Cloud development · Microsoft

Azure Key Vault Keys SDK for .NET

Microsoft's .NET guide to Azure.Security.KeyVault.Keys: creating RSA, EC and symmetric keys in Key Vault or Managed HSM, rotating them by policy, and running encrypt, wrap and sign operations through CryptographyClient.

key-vaultcryptographydotnet

Why we selected itIt walks the key lifecycle in one file: create RSA, EC and Oct keys, list versions, update properties, delete and purge, back up and restore, rotate, and set a policy that rotates 30 days before a 90-day expiry.

Weigh upBoth install lines Microsoft prints for this skill, `npx skills add microsoft/skills --skill azure-sdk-dotnet` and `/plugin install azure-sdk-dotnet@skills`, install the whole azure-sdk-dotnet plugin rather than the Key Vault keys skill alone.

Source reviewed

6 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Cloud development · Microsoft

Azure Key Vault Keys SDK for TypeScript

Microsoft's TypeScript guide to @azure/keyvault-keys for creating, rotating and deleting Key Vault keys and for encrypt, sign and wrap operations through CryptographyClient; the same file also carries secret operations.

key-vaultcryptographytypescript

Why we selected itIt creates keys four ways (generic RSA, RSA with keySize 2048, EC on P-256, and with keyOps, tags and an expiry), rotates a key on demand and sets a policy that rotates P30D before a P90D expiry.

Weigh upBoth install lines Microsoft prints for this skill, `npx skills add microsoft/skills --skill azure-sdk-typescript` and `/plugin install azure-sdk-typescript@skills`, install the whole azure-sdk-typescript plugin rather than the keys skill alone.

Source reviewed

6 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Cloud development · Microsoft

Azure Key Vault Secrets library for Rust

Microsoft's Rust guide to azure_security_keyvault_secrets: setting, reading, tagging, deleting and listing Key Vault secrets, with a structured error for a missing secret.

key-vaultsecretsrust

Why we selected itIt covers set, get, update properties (content type and tags), delete, and paged listing of secret properties with SecretClient.

Weigh upBoth install lines Microsoft prints for this skill, `npx skills add microsoft/skills --skill azure-sdk-rust` and `/plugin install azure-sdk-rust@skills`, install the whole azure-sdk-rust plugin rather than the secrets skill alone.

Source reviewed

6 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Cloud development · Microsoft

Azure Key Vault Secrets SDK for TypeScript

Microsoft's TypeScript guide to @azure/keyvault-secrets for setting, reading, versioning, deleting, recovering and backing up Key Vault secrets; the same file also carries key and cryptography operations.

key-vaultsecretstypescript

Why we selected itIt walks a secret's lifecycle: set it with enabled, expiresOn, contentType and tags, read the latest or a given version, list names and versions, soft-delete, purge and recover.

Weigh upBoth install lines Microsoft prints for this skill, `npx skills add microsoft/skills --skill azure-sdk-typescript` and `/plugin install azure-sdk-typescript@skills`, install the whole azure-sdk-typescript plugin rather than the secrets skill alone.

Source reviewed

6 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Cloud operations · Microsoft

Azure Kubernetes

Pins Microsoft's AKS cluster skill, which refuses to print secrets or subscription identifiers and sends application deployment to a different skill.

azureakskubernetes

Why we selected itThe guardrails forbid requesting or printing secrets and forbid asking the user to paste subscription IDs.

Weigh upThe resource was not executed, and no Azure API, CLI, or MCP tool was called.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Azure portal navigation · Microsoft

Azure portal search links

Pins Microsoft's portal-link skill, which signs in with Connect-AzAccount or device code and only builds links for resource types listed as enabled.

azure-portalauthenticationdeep-link

Why we selected itThe file names authentication as Connect-AzAccount or device code with -UseDeviceAuthentication, and a 401 tells the operator to sign in again with the tenant that can access the subscription.

Weigh upThe resource was not executed.

Source reviewed

6 sources · 2026-10-15

Read the full review Upstream setup linkedMIT

Azure SDKs · microsoft

Azure Queue Storage for Python

Review queue send, receive, visibility and deletion operations for Python workers.

azure-sdksmicrosoft

Why we selected itThe reference distinguishes peek from receive, exposes dequeue_count/visibility behavior and shows clear_messages as a destructive operation; current service docs require the latest valid pop receipt.

Weigh upVisibility is not an exclusive-processing guarantee. update_message returns a new pop receipt; the repeated-update example must retain the latest result. Approximate queue length is not an exact backlog count.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Cloud inventory · Microsoft

Azure Resource Lookup

Pins Microsoft's Azure resource lookup skill, which refuses deployment and change requests and treats Resource Graph as read-only.

azureresource-graphinventory

Why we selected itThe description says not to use the skill to deploy or change resources, and the constraints say never to mutate through Resource Graph.

Weigh upThis review did not execute the resource.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Cloud architecture diagrams · Microsoft

Azure Resource Visualizer

Pins Microsoft's Azure resource visualizer, which diagrams one resource group, keeps secret values out of the drawing, and does not modify or delete resources.

azuremermaidarchitecture

Why we selected itThe never-do list says not to modify or delete Azure resources and calls the analysis read-only.

Weigh upThis review did not execute the resource.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Cloud development · microsoft

Azure Service Bus for Python

Implement queue and topic consumers with explicit settlement, sessions and dead-letter handling.

cloud-developmentmicrosoft

Why we selected itThe skill text prefers DefaultAzureCredential and says connection strings bypass Entra audit, but the admin script uses SERVICEBUS_CONNECTION_STRING first when that variable is set.

Weigh upThe script can create and delete Service Bus entities through the management API. It does not process the application messages themselves.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Azure SDKs · microsoft

Azure Tables for Python

Review partition-aware entity operations and transactions for Azure Tables.

azure-sdksmicrosoft

Why we selected itThe reference submits one transaction for a shared partition and handles TableTransactionError; the definition distinguishes service-level table operations from entity-level client operations.

Weigh upTokenCredential support differs by endpoint: the current SDK documentation supports it for Storage Tables, not universally for Cosmos Table endpoints. Upsert alone does not protect concurrent writers or make arbitrary workflows idempotent.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Azure upgrades · Microsoft

Azure upgrade

Pins Microsoft's in-Azure upgrade skill, which requires an assessment and an explicit confirmation before the original app is deleted or stopped.

azureupgradefunctions

Why we selected itThe catalogue has other Azure skills from this repository and not this plan, SKU, Java SDK, and Redis upgrade procedure.

Weigh upThis review did not execute the resource.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Azure deployment checks · Microsoft

Azure validate

Pins Microsoft's pre-deploy Azure check, which stops without an approved plan and will not run azd deploy itself.

azurevalidationdeployment

Why we selected itThe catalogue does not already have this gate between azure-prepare and azure-deploy.

Weigh upThis review did not execute the resource.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Infrastructure · hashicorp

Azure Verified Module Review

Review Terraform module structure against the pinned AVM requirements guide.

infrastructurehashicorp

Why we selected itThe guide separates mandatory and advisory requirements, requires registry version pins for AVM references and constrains resource for_each identities.

Weigh upThis skill is guidance, not AVM certification. Provider ranges and requirement identifiers must be checked against current AVM specifications before submission.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMPL-2.0

Azure Web PubSub · Microsoft

Azure Web PubSub Service Python

Pins the Web PubSub service Python skill, which prefers DefaultAzureCredential and says to limit client tokens with roles.

azureweb-pubsubentra-id

Why we selected itThe authentication block prefers DefaultAzureCredential and says connection strings and account or API keys bypass Entra audit and rotation.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Documentation and knowledge · Sentry

Blog writing guide

Applies one publisher's voice rules, including a banned-language list, and does not pretend that voice is universal.

writingvoiceediting

Why we selected itThe voice rules are publisher-specific, including a banned-phrase list and a named-author requirement.

Weigh upThe opening must state the problem or the conclusion, not company history.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

AI evaluation · Braintrust

Braintrust Build Eval Dataset

Draft or audit an evaluation dataset's population, label provenance, transform invariants and split lifecycle. This is a review procedure, not an executable dataset validator.

datasetsvalidationprovenance

Why we selected itAdds dataset construction and audit before the catalogue's existing sample-size and scorer-validation stages; it requires population and label provenance rather than treating more rows as better evidence.

Weigh upInstructions do not mechanically validate data, enforce test-touch counts or prove a corpus uncontaminated. Sample-size calculations and scorer implementations are explicitly outside this skill.

Source reviewed

8 sources · 2026-10-15

Read the full review Upstream setup linkedApache-2.0

AI evaluation · Braintrust

Braintrust Define Eval Release Gate

Translate completed, intact evaluation evidence into owned ship-or-hold rules. Separates improvement targets from non-negotiable guardrails; does not implement or activate CI.

evaluationrelease-gatesreliability

Why we selected itConsumes a scorer-validation verdict and confirmatory results to produce an explicit decision contract, a different job from constructing data or computing sample size.

Weigh upNo statistical engine or CI implementation is shipped in the selected directory. A written gate is not an enforced deployment block.

Source reviewed

7 sources · 2026-10-15

Read the full review Upstream setup linkedApache-2.0

AI evaluation · Braintrust

Braintrust Design Eval Experiment

Prepare a comparison manifest before inference: freeze the serving path, tool permissions, pairing, treatment implementation and stopping rules. It does not run an experiment.

evaluationreproducibilitybatch-inference

Why we selected itMakes batching, provider, region, cache state and decoding part of the controlled comparison instead of treating a model name as a reproducible configuration.

Weigh upNo batch submission, retry implementation, quota enforcement or provider integration is supplied. The manifest is a plan, not a run receipt.

Source reviewed

7 sources · 2026-10-15

Read the full review Upstream setup linkedApache-2.0

Browser verification · addyosmani

Browser Testing with DevTools

Inspect real browser behavior, network activity and rendering through Chrome DevTools MCP.

browser-verificationaddyosmani

Why we selected itThe source explicitly distinguishes isolated/dedicated profiles from attaching to a running profile and treats DOM, console and network output as untrusted data.

Weigh upAccessibility-tree inspection is not a screen-reader test. Browser content is untrusted; attaching to an existing authenticated profile expands access to unrelated sessions.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Developer Tools · Buf Technologies

Buf Protocol Buffers Design and Toolchain

Official Buf skill for Protocol Buffers (.proto) design, buf.yaml toolchain configuration, Connect/gRPC service authoring, and breaking change prevention.

bufprotobufgrpc

Why we selected itProtovalidate constraints are required on fields, not left as a later cleanup.

Weigh upThe skill says not to invent a style when the repo has none; ask which style to copy.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Testing and reliability · GitHub community

Bug Receipt

Creates an evidence-bearing defect or incident handoff that distinguishes verified, partial and blocked outcomes.

debuggingverificationincident-handoff

Why we selected itRequires the observed baseline, root-cause evidence, changed artifacts and relevant checks to accompany a completion status.

Weigh upThe validator checks supplied fields and status consistency; it cannot prove that a command actually ran or that a claimed observation is correct.

Source reviewed

11 sources · 2026-10-15

Read the full review Install command in reviewMIT

Development workflow · GitHub community (awesome-copilot)

Bug Reproduction Brief

Reduce an uncertain bug report to an evidence-backed failing fixture before repair.

development-workflowgithub

Why we selected itThe workflow stops at a reproduction, labels second-hand evidence unverified and restores the last removed condition when minimization loses the failure.

Weigh upTwo successful reproductions do not prove determinism; production writes and sensitive records should not be used simply to increase confidence.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Model evaluation · MLflow

Build a Scorer

Turns user-defined failure criteria into a small scorer implementation and validation plan, then hands registration and evaluation runs to a separate workflow.

mlflowllm-as-judgescorers

Why we selected itRoutes each atomic criterion to code, a matching built-in or an LLM judge, with an explicit why-not-cheaper justification and at least one user-outcome check.

Weigh upThis is scorer design, not an end-to-end evaluation runner. Phase 8 assigns dataset creation, registration, runs and result interpretation to agent-evaluation, with tracing gaps handed to instrumenting-with-mlflow-tracing.

Source reviewed

7 sources · 2026-10-14

Read the full review Install command in reviewApache-2.0

AI application development · Anthropic

Build an MCP App (Interactive UI Widgets)

Adds interactive UI widgets (forms, pickers, confirm dialogs, charts) to an MCP server: the tool declares a ui:// resource, the server serves the widget HTML, and the host renders it in a sandboxed iframe in the chat.

mcpformscsp

Why we selected itIt says when a widget is worth building (structured input Claude cannot infer, picking from a list, confirming a destructive or billable action, visual output, live progress) and when elicitation covers the need with no UI code.

Weigh upLast commit touching plugins/mcp-server-dev/skills/build-mcp-app at revision 41f32e40: 2026-04-28 (f4b5494f).

Source reviewed

4 sources · 2026-10-21

Read the full review Upstream setup linkedApache-2.0

AI application development · Anthropic

Build an MCP Server

Interviews the developer about an MCP server before any code, recommends one deployment model and a tool-design pattern, then scaffolds a remote or local stdio server or hands off to the MCP app or MCPB skill.

mcptoolsoauth

Why we selected itBefore any scaffolding it asks five questions (what the server connects to, who will use it, how many actions it exposes, whether a tool needs mid-call input or rich UI, and the upstream auth) and then recommends one of four paths: remote streamable HTTP, MCP app, MCPB or local stdio.

Weigh upLast commit touching plugins/mcp-server-dev/skills/build-mcp-server at revision 41f32e40: 2026-04-17 (b00abee2).

Source reviewed

7 sources · 2026-10-22

Read the full review Upstream setup linkedApache-2.0

AI application development · Anthropic

Build an MCPB (Bundled Local MCP Server)

Packages a local stdio MCP server with its runtime and dependencies into a .mcpb zip archive with a manifest.json, so a user can install it without setting up Node or Python.

mcppackagingbundles

Why we selected itIt defines an MCPB in the source's own terms: a zip archive holding manifest.json, the server code with its bundled dependencies and an icon, which the host launches as a stdio MCP server, built with npx @anthropic-ai/mcpb pack.

Weigh upLast commit touching plugins/mcp-server-dev/skills/build-mcpb at revision 41f32e40: 2026-04-17 (b00abee2).

Source reviewed

4 sources · 2026-10-21

Read the full review Upstream setup linkedApache-2.0

Research and evidence · GitHub community

Build Evidence Map

Builds a structured map of claims, evidence and unresolved gaps, with a local validator for the map’s structure.

evidenceresearchdecision-records

Why we selected itPreserves supporting, contradicting, qualifying and missing evidence as separate relationships instead of flattening a contested choice into confident prose.

Weigh upThe bundled validator checks structure and declared excerpts; it does not establish that a source is true or that the excerpt supports the claim.

Source reviewed

11 sources · 2026-10-15

Read the full review Install command in reviewMIT

Data engineering · dbt-labs

Build the dbt Semantic Layer

Define semantic entities, dimensions and metrics using the specification supported by the project.

data-engineeringdbt-labs

Why we selected itIt builds dbt semantic models and metrics, and it chooses the latest or legacy spec from the project that is already there.

Weigh upThe skill does not finish when only one of the two validations has passed.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Security · trailofbits

Burp Project Parser

Extract and search captured Burp traffic while distinguishing empty results from parser startup failures.

securitytrailofbits

Why we selected itParsing is delegated to a local Burp Suite Professional process. The skill text says it does not parse .burp files itself.

Weigh upThe shell script launches the local burpsuite_pro.jar headless. It does not ship Burp and it does not open a new scan against a host.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Security · trailofbits

C and C++ Security Review

Coordinate scoped native C/C++ security review and assemble findings with an explicit coverage ledger.

securitytrailofbits

Why we selected itThe declared tools are Workflow, AskUserQuestion, Bash, and Read, and the assemble step is a local Python script run with uv.

Weigh upBash can still write files even though Write is not listed. The skill does not itself compile or execute the program under review as a test.

Source reviewed

4 sources · 2026-10-14

Read the full review Upstream setup linkedCC-BY-SA-4.0

Development workflow · GitHub community (awesome-copilot)

C# / TypeScript API Contract Drift

Compare C# controller and DTO contracts with TypeScript or JavaScript consumers in both directions.

development-workflowgithub

Why we selected itIt compares C# controller and DTO changes with TypeScript or JavaScript callers and writes a report instead of a patch.

Weigh upThe skill does not modify code.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Testing · Trail of Bits

cargo-fuzz

Requires the nightly toolchain and a fuzz_target harness before a Rust fuzz run counts.

fuzzingrustcargo

Why we selected itThe description names the nightly toolchain requirement.

Weigh upCC-BY-SA-4.0 applies to an adaptation you redistribute.

Source reviewed

2 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Browser extension connectivity · Trail of Bits

Chrome MCP troubleshooting

Pins the macOS Chrome skill that refuses to leave both Claude native hosts active and says a fix does not apply until Chrome and Claude Code restart.

chromemcpmacos

Why we selected itThe file says the two native hosts cannot be used at the same time and tells the operator to disable the one that is not in use.

Weigh upThe resource was not executed.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Delivery automation · addyosmani

CI and Delivery Automation

Design quality gates, previews and staged deployment around a reproducible CI workflow.

delivery-automationaddyosmani

Why we selected itThe guide separates feature deployment from flag release, includes flag cleanup and treats CI failure output as a bounded feedback loop for agents.

Weigh upExample action/runtime versions require refresh; pipeline success is limited to its checks. Production migrations and external side effects may not be reversible by code rollback.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Research and documentation · Anthropic

Claude Academy guide

Answers a Claude how-to first, then may name at most two Academy items copied from a fetched catalog.

learningclaudecatalog

Why we selected itIt answers a Claude how-to first, then may name at most two Academy items copied from a fetched catalog.

Weigh upThe skill embeds no course list; a recommendation depends on fetching https://academy.claude.com/assets/data/catalog.json.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableApache-2.0

Claude Code automation · Anthropic

Claude automation recommender

Pins the Claude Code setup skill that recommends hooks, skills, and MCP servers and says this skill must not create or modify files.

claude-codeautomationshooks

Why we selected itThe skill says it is read-only, does not create or modify files, and tells the user to implement recommendations separately.

Weigh upThis review did not execute the resource.

Source reviewed

6 sources · 2026-10-15

Read the full review Upstream setup linkedApache-2.0

Security review · Anthropic

Claude Security

Runs Anthropic's Claude Security scans inside a Claude Code session: a whole-repository or scoped scan, a scan of a branch, pull request or commit, and suggested fixes as patch files the user applies.

security-reviewvulnerability-researchpatches

Why we selected itThe README calls it the in-your-session version of Anthropic's hosted Claude Security product: one skill that drives three jobs, scan codebase, scan changes and suggest patches, through the plugin's own scan workflow and agents.

Weigh upProprietary licence. The LICENSE grants a limited, revocable licence "to install, run, and modify the Plugin for your internal use, solely with Claude Code or other Anthropic products and services", and, "Except as the Agreement expressly permits", forbids using "the Plugin or any part of it with, or to develop, any non-Anthropic product or service, including any competing product", as well as distributing or publishing it to any third party.

Source reviewed

21 sources · 2026-10-22

Read the full review Install command in reviewProprietary (Anthropic, PBC)

CLAUDE.md maintenance · Anthropic

CLAUDE.md improver

Pins the CLAUDE.md skill that must print a quality report and wait for approval before it edits those files.

claude-mdauditconfirmation

Why we selected itThe skill says it can write CLAUDE.md files only after it presents a quality report and the user approves.

Weigh upThis review did not execute the resource.

Source reviewed

6 sources · 2026-10-15

Read the full review Upstream setup linkedApache-2.0

Agent development · Cloudflare

Cloudflare Agents SDK

Guides stateful Cloudflare agent applications through current SDK documentation, persistence, scheduling and approval-aware tool flows.

cloudflareagentsdurable-state

Why we selected itTreats installed SDK versions and current docs as the basis for implementation.

Weigh upThis is specific to Cloudflare’s Agents SDK and runtime, not a general multi-provider agent framework.

Source reviewed

8 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Analytics pipelines · Cloudflare

Cloudflare Basin

Pins Cloudflare's Basin skill, which names the SQL auth token and says to check the installed Wrangler version before running the new basin commands.

cloudflarebasinwrangler

Why we selected itBasin SQL is tied to the WRANGLER_BASIN_SQL_AUTH_TOKEN variable and the REST path /basin-sql/query/{BUCKET}.

Weigh upThe file names the SQL auth token and a pre-command version check. It does not show a delete, a dry-run, or a refusal to start a pipeline.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Cloud operations · Cloudflare

Cloudflare cf CLI

Pins Cloudflare's cf CLI skill, which tells an agent to search the local command index and to keep Wrangler when a Wrangler config is already present.

cloudflarecliwrangler

Why we selected itThe catalogue already has Wrangler, Workers, and Turnstile skills from this repository and not the cf CLI skill.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Cloud development · Cloudflare

Cloudflare Durable Objects

Designs per-entity state and coordination with explicit persistence, concurrency and runtime-test guidance.

cloudflaredurable-objectssqlite

Why we selected itDefines where Durable Objects fit and where stateless Workers or independent requests fit better.

Weigh upA single coordination object can become a bottleneck; object boundaries must fit the workload.

Source reviewed

8 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Cloud development · Cloudflare

Cloudflare Email Service

Implements Cloudflare Email Sending and Email Routing: Workers send_email bindings, the REST API, inbound email() handlers, wrangler setup, and deliverability (SPF/DKIM/DMARC, bounces, suppression lists) — with the field-naming asymmetries between binding and REST spelled out.

emailtransactional-emailcloudflare-workers

Why we selected itTakes a retrieval-over-pre-training stance with named retrieval sources, the right posture for a 2025-launched product whose API surface is still moving.

Weigh upCloudflare-platform-only: the binding, wrangler commands and REST endpoints are Cloudflare’s, not general SMTP or templating guidance.

Source reviewed

7 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Web Frameworks · Cloudflare

Cloudflare Migrate to vinext

Migrates Next.js applications to Cloudflare Workers via vinext (Vite-based Next.js reimplementation), handling ESM conversions and worker deployment.

cloudflarevinextnextjs

Why we selected itMigration is defined as a package and config change, not an application rewrite.

Weigh upThe file says not to rewrite next/* imports and not to copy webpack or Turbopack config into Vite.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewMIT

Plan or troubleshoot Access, Gateway, device clients, private networks, DLP and CASB from the actual traffic path.

cloud-securitycloudflare

Why we selected itIt is an architecture and troubleshooting map for Cloudflare One that refuses guessed API identifiers.

Weigh upThe skill retrieves current docs and does not embed the live API schema.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Network migration · Cloudflare

Cloudflare One migrations

Pins Cloudflare's One migration skill, which stages rules disabled and refuses placeholder DLP and unapproved deletion of stale rules.

cloudflarezero-trustmigration

Why we selected itThe catalogue has the Cloudflare One product skill from this commit and not this migration assessment file.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Product selection · Cloudflare

Cloudflare product discovery

Pins Cloudflare's product-discovery skill, which blocks a new Wrangler setup, runs cf migrate only when asked, and says a version rollback does not restore resource data.

cloudflarewranglerworkers

Why we selected itA project that is not already on Wrangler must stay on cf: the file says not to install Wrangler or create a Wrangler configuration file, and to run cf migrate only when the user asks.

Weigh upThe product table still names the wrangler skill for C3. The CLI section says that where the table names a Wrangler command or file, the cf equivalent is used instead of translating from memory.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Sandbox development · Cloudflare

Cloudflare Sandbox next

Pins Cloudflare's sandbox-next skill, which stops on the stable package, refuses a mixed preview image, and keeps live credentials in the Worker.

cloudflaresandboxcontainers

Why we selected itThe file says to stop when the app uses the default @cloudflare/sandbox package, and not to mix an @next Worker package with a stable container image.

Weigh upSeveral different tasks in the retrieval table share the URL https://developers.cloudflare.com/sandbox/index.md, so the table does not by itself distinguish those pages.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Cloud operations · Cloudflare

Cloudflare Sandbox preview migration

Pins Cloudflare's stable-to-preview Sandbox migration, which refuses an unagreed production cutover and forbids invented APIs and live secrets in the sandbox environment.

cloudflaresandboxworkers

Why we selected itThe catalogue has sandbox-stable from this commit and not the migration off that stable package.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Cloud development · cloudflare

Cloudflare Sandbox Stable SDK

Maintain existing stable Sandbox SDK applications without mixing preview package, image, command or terminal APIs.

cloud-developmentcloudflare

Why we selected itIt keeps stable @cloudflare/sandbox code on the stable command APIs and stops a half-applied move to @next.

Weigh upThe skill is not a full SDK manual and does not vendor the command signatures.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Security · Cloudflare

Cloudflare Turnstile Spin

Integrate a Turnstile widget with an existing frontend and server-side Siteverify, including secret-safe helpers and replay validation.

turnstilebot-managementcaptcha-migration

Why we selected itIt wires Cloudflare Turnstile into an existing frontend and backend, with server-side siteverify and a token that is not pasted into chat.

Weigh upA pure-static site with no backend is out of scope, because siteverify needs a server.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Web performance · Cloudflare

Cloudflare Web Performance Audit

Builds a performance review from browser traces, request details and relevant source instead of generic optimization advice.

core-web-vitalsdevtoolslighthouse

Why we selected itStarts by discovering available measurement tools and records measurements it could not collect.

Weigh upA lab trace does not establish real-user Core Web Vitals or prove a resource is unused in every interaction.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Cloud development · Cloudflare

Cloudflare Workers Best Practices

Reviews Workers code against its configured runtime, focusing on request lifetime, state isolation, streaming and binding contracts.

workerscloudflarestreaming

Why we selected itUses installed types and compatibility settings instead of assuming the newest package defines the current runtime.

Weigh upWorkers-specific guidance is not interchangeable with a Node server checklist.

Source reviewed

7 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Capture CPU or heap profiles for a specific deployed Worker or Durable Object and turn pprof evidence into a focused optimization proposal.

performancecloudflare

Why we selected itIt profiles deployed Workers and Durable Objects by capturing a pprof file and mapping hotspots back to source.

Weigh upThe skill does not install cf and does not start a profile without the account, Worker, and deployed version.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Cloud operations · Cloudflare

Cloudflare Wrangler Workflow

Runs Wrangler work in projects that already have a Wrangler configuration file, against the installed Wrangler version and an explicitly identified target.

wranglercloudflaredeployment

Why we selected itDistinguishes local simulation from remote bindings so development commands are not assumed to be harmless.

Weigh upThe skill authorizes no cloud action by itself; commands can deploy or mutate real resources.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Databases · Cockroach Labs

CockroachDB SQL

Cockroach Labs skill that turns a natural-language request into CockroachDB SQL, applies the relevant rule files with 00-fundamental-principles.md first, and tells the agent to run EXPLAIN on every generated query when a cluster connection is available.

sqlschemaexplain

Why we selected itSKILL.md marks plan validation as mandatory: when connected, it tells the agent to run EXPLAIN on every generated query, fix and re-run it after a parsing or syntax error, and include the EXPLAIN output in the response.

Weigh upThe skill directory skills/cockroachdb-query-and-schema-design/cockroachdb-sql last changed in a commit dated 2026-07-22, which is also the date of the pinned revision.

Source reviewed

9 sources · 2026-10-21

Read the full review Upstream setup linkedApache-2.0

Autonomous review loop · Trail of Bits

Code improver

Pins the generic review loop that will not choose a reviewer, will not start on an unconfirmed scope, and will not commit.

reviewworkflowscope

Why we selected itThe user must name the reviewer, there is no default, and a missing name is a question rather than a guess.

Weigh upThe resource was not executed.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Security review · Trail of Bits

CodeQL

Drives a CodeQL scan and warns that each Bash call is a fresh shell, so state must be written to files.

codeqlstatic-analysissecurity

Why we selected itA heading states that each Bash call is a fresh shell.

Weigh upCC-BY-SA-4.0 applies to adaptations you redistribute.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Documentation · getsentry

Collaborative Document Drafting

Build a substantial document through context gathering, section refinement and reader testing.

documentationgetsentry

Why we selected itThe workflow offers a freeform alternative, separates context from refinement and uses a reader without the original conversation to find hidden assumptions.

Weigh upA fresh-model reading test detects ambiguities but cannot verify factual claims or stakeholder agreement. The fixed image-access assumption depends on the actual host.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Agent development · Anthropic

Command Development for Claude Code

Explains how to write Claude Code slash-command files: frontmatter fields, $ARGUMENTS and positional arguments, @file references, inline Bash and plugin command patterns.

claude-codeplugin-devprompts

Why we selected itIt covers the five command frontmatter fields it names (description, allowed-tools, model, argument-hint, disable-model-invocation) together with $ARGUMENTS, $1/$2 positional arguments, @file references and inline Bash.

Weigh upLast commit touching plugins/plugin-dev/skills/command-development at revision 41f32e40: 2026-03-17 (6b70f99f).

Source reviewed

5 sources · 2026-10-21

Read the full review Upstream setup linkedApache-2.0

Model evaluation · Replicate

Compare Replicate Models

Shortlists Replicate models and compares them on measured cost, prediction time, output quality and input schema.

replicatemodel-selectioninference-cost

Why we selected itIt tells the agent to run a few predictions and read the metrics field for actual cost and predict_time instead of assuming them.

Weigh upTest predictions are billed, and the file asks for no cost confirmation before running them.

Source reviewed

4 sources · 2026-10-13

Read the full review Install command in reviewApache-2.0

Security review · Trail of Bits

Constant-time analysis

Routes a timing review by language and says when not to use the analyzer.

cryptographytimingside-channels

Why we selected itThe description names secret-dependent branches and division on secrets.

Weigh upCC-BY-SA-4.0 applies to an adaptation you redistribute.

Source reviewed

2 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Testing · trailofbits

Constant-Time Runtime Testing

Design statistical and dynamic tests for secret-dependent timing in cryptographic implementations.

testingtrailofbits

Why we selected itThe file is a handbook for measuring a running cryptographic implementation, and it sends static compiler inspection to a different plugin.

Weigh upThere is no allowed-tools list and no bundled runner. Links to papers and tools are references. The file does not vendor dudect or Timecop.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Development workflow · Addy Osmani

Constraint-driven development

Writes the quality bar down and watches the diff for a weakened gate.

qualityconstraintsratchet

Why we selected itThe quality bar is a CONSTRAINTS.md file with a tool behind each number, not a verbal standard.

Weigh upThe interview must not run in CI, /loop, or other non-interactive contexts.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Agent development · Addy Osmani

Context engineering

Orders agent context as a hierarchy instead of dumping the repository into the prompt.

contextagentsinstructions

Why we selected itIt orders context in five levels, from rules files to conversation history, and says what to load at each level and when.

Weigh upIts thresholds (trim at 75% of the window, about 2,000 lines of focused context per task) are the author's guidance, not measurements this review checked.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Planning and delivery · GitHub community

Conventional Branch

Creates and validates Git branches against the Conventional Branch specification — feature/, bugfix/, hotfix/, release/, chore/ — with a validation ruleset, an invalid-example table, trunk detection and a mapping to Conventional Commits.

gitbranchingbranch-naming

Why we selected itTurns branch naming from a style argument into a validation ruleset — lowercase only, no consecutive hyphens, dots only in release/ version descriptions — with an invalid-example table that shows each rule being broken.

Weigh upThe Conventional Branch specification is a community convention, not something git enforces. The skill ships no linter or hook, so names are checked only while the agent runs.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Planning and delivery · Matt Pocock

Conversation to Specification

Turns an already-discussed change into a repository-aware specification for the configured issue tracker.

specificationplanninggithub

Why we selected itUses existing project context, glossary and architecture decisions before synthesizing requirements.

Weigh upThis is synthesis after discussion, not a discovery interview or a check that every stakeholder requirement is present.

Source reviewed

16 sources · 2026-10-15

Read the full review Install command in reviewMIT

Web performance · Addy Osmani

Core Web Vitals

Diagnoses LCP, INP and CLS from field evidence first — CrUX p75 with a labelled origin fallback, then a browser trace — and routes to per-metric references, with explicit rules against claiming field improvement from a lab run.

core-web-vitalslcpinp

Why we selected itMeasurement discipline is the spine: CrUX p75 first with a labelled fallback when page-level data is missing, trace analysis of only the failing metric's insights, and re-measurement after the fix with field verification explicitly deferred until new user data exists.

Weigh upCross-links the sibling performance skill's references (../performance/references/MEASUREMENT.md and RUM.md); installing core-web-vitals alone leaves those two links dangling.

Source reviewed

7 sources · 2026-10-15

Read the full review Install command in reviewMIT

Azure cost governance · Microsoft

Cost Governance

Pins Microsoft's cost-governance skill, which requires a confirmed scope and amount before a budget write and keeps the existing budget on conflict.

azure-costbudgetconfirmation

Why we selected itBudget setup is a write, and the file requires confirmed scope, amount, thresholds, and recipients before that write.

Weigh upThis review did not execute the resource.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Testing · Trail of Bits

Coverage analysis

Turns an uncovered region into the next harness or seed change, instead of treating a long run as done.

fuzzingcoveragellvm

Why we selected itThe description says to use it when a fuzzer plateaus.

Weigh upCC-BY-SA-4.0 applies to an adaptation you redistribute.

Source reviewed

2 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Web performance · Microsoft

CPU Profile Analysis

Walks an agent through analyzing V8 CPU profiles (.cpuprofile) and Chrome DevTools trace files: parsing large files safely, reconstructing call trees from sampling data, measuring self versus total time, and comparing code paths.

profilingcpu-profileflamegraph

Why we selected itSets out the sampling-profiler basics before any tooling: self versus total time, idle/program/GC samples, and that a function missing from the samples is expected rather than suspicious.

Weigh upSampling data is statistical, not exact: the skill itself warns that per-function attribution is approximate and marker-based region detection tolerates gaps.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Security review · Trail of Bits

Crypto Protocol Diagram

Extracts a cryptographic protocol's message flow from source, an RFC, a paper, pseudocode, a ProVerif model or a Tamarin model, writes it as an annotated Mermaid sequence diagram in a Markdown file and prints an ASCII version.

cryptographyprotocolmermaid

Why we selected itProduces a Mermaid sequenceDiagram in a Markdown file, plus an ASCII diagram printed inline, from protocol source code or a spec: RFC, paper, pseudocode, prose, ProVerif (.pv) or Tamarin (.spthy).

Weigh upIt diagrams protocol flow; for formal verification its SKILL.md points to mermaid-to-proverif.

Source reviewed

5 sources · 2026-10-22

Read the full review Upstream setup linkedCC-BY-SA-4.0

Data and analytics · Cube

Cube build model

Cube's skill for authoring and changing a Cube semantic model in YAML on a dev-mode branch, through the Cube MCP tools or the Cube CLI, with validation, commit, and publishing only on an explicit request.

analyticsdata-modelsmetrics

Why we selected itSpells out the write loop for both paths: seven Cube MCP tool steps (startDataModelEdit, read, writeDataModelFile with its recompile and validationError, getDataModelChanges, searchDataModel and runQuery on the same branchName, commitDataModelChanges, then mergeToDefaultBranch only on an explicit publish request) and an eight-step CLI dev-mode block that ends with build-status --wait and exit-dev-mode.

Weigh upThe skill directory skills/cube-build-model last changed in a commit dated 2026-09-23 (git log at the pin), the same date as the pinned revision 5e5ef13c.

Source reviewed

5 sources · 2026-10-21

Read the full review Upstream setup linkedApache-2.0

AI application development · Datadog Labs

Datadog LLM Observability Instrumentation

Instruments a Python or Node.js/Next.js backend with Datadog LLM Observability, auditing session-ID plumbing and persisting the dependency before reporting.

llm-observabilityinstrumentationddtrace

Why we selected itRuns detection before editing: runtime, LLM framework, app framework, existing instrumentation, and the manifest to persist into are all identified in Phase 1.

Weigh upEdits source and dependency manifests, so every change needs review and a test run before it ships.

Source reviewed

8 sources · 2026-10-15

Read the full review Install command in reviewMIT

Model evaluation · Datadog Labs

Datadog Trace RCA

Root-causes production AI failures from LLM traces through phased span analysis, then recommends prompt, tool, routing or retrieval changes quoted from the trace evidence.

evalroot-cause-analysisllm-traces

Why we selected itStructures the analysis as explicit phases from input resolution through open and axial coding to recommendations, instead of offering generic debugging advice.

Weigh upRequires either the Datadog LLMO MCP toolset or the pup CLI with usable authentication and existing LLM Observability traces.

Source reviewed

7 sources · 2026-10-15

Read the full review Install command in reviewMIT

Data engineering · dbt-labs

dbt Analytics Engineering

Build and revise transformations with source profiling, explicit lineage and targeted validation.

data-engineeringdbt-labs

Why we selected itIt builds and changes dbt models with a shell limited to dbt and jq, and it requires dbt show before the work is called done.

Weigh upThe allowed shell is dbt and jq, not an unrestricted Bash.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Analytics engineering · dbt Labs

dbt Core to v2 migration triage

Pins dbt Labs' v2 migration triage, which asks before a credential check and forbids edits until autofix changes have been reviewed.

dbtmigrationcredentials

Why we selected itThe catalogue has other dbt skills from this repository, including upgrading-dbt, and not this Core-to-v2 triage procedure.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Databases · dbt Labs

dbt Documentation Maintenance

Audits descriptions in a dbt manifest and drafts missing model or column documentation in the project’s existing style.

dbtdocumentationsql

Why we selected itUses dbt’s resolved manifest rather than guessing YAML layout, and clearly limits column coverage to columns declared in that manifest.

Weigh upThe audit measures nonempty descriptions, not their correctness or every physical warehouse column; imported model nodes can also appear in a manifest.

Source reviewed

9 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

MCP configuration · dbt Labs

dbt MCP server configuration

Pins dbt Labs' MCP setup skill, which keeps tokens out of committed config and leaves SQL and codegen off until their flags are changed.

dbtmcptoken

Why we selected itConfiguration that may be committed must use environment-variable references, and the file says never to log, display, or echo token values.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Data engineering · dbt-labs

dbt State Reuse

Configure and investigate server-backed reuse decisions separately from manifest selectors and deferral.

data-engineeringdbt-labs

Why we selected itIt explains dbt State as a paid server that can skip or clone a node, and it separates that from state:modified.

Weigh updbt State is not local and is not free.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Infrastructure development · Pulumi

Debug a Failed Pulumi Operation

Reads the diagnostics Pulumi already recorded for a failed update or preview and places the fix in the program, state or environment.

pulumiinfrastructure-as-codedebugging

Why we selected itIt reads the recorded engine events with pulumi api instead of re-running the failed operation, and confirms the stack and version with the user before reading further.

Weigh upThe file only describes records read from Pulumi Cloud through pulumi api, and its commands need jq.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Model evaluation · Confident AI

DeepEval

Confident AI's skill for adding a committed pytest eval suite to an AI agent, RAG pipeline, chatbot or LLM app: it reuses the user's dataset or else generates goldens with deepeval generate, runs deepeval test run and changes the app between rounds based on the failures.

evaluationsevaltesting

Why we selected itConfident AI publishes it in confident-ai/deepeval at skills/deepeval, with frontmatter license Apache-2.0, author Confident AI and version 1.0.0, ten reference files and four templates: single-turn tracing, single-turn no tracing, multi-turn end to end, and shared metric lists.

Weigh upThe skills/deepeval directory last changed in a commit dated 2026-05-21; the pinned revision 59a4f00a is dated 2026-10-16.

Source reviewed

11 sources · 2026-10-21

Read the full review Upstream setup linkedApache-2.0

Development workflow · Deno Land Inc.

Deno

Deno's own skill for Deno 2.9+ projects that keep package.json, commit deno.lock, and do not run npm lifecycle scripts unless they are approved.

denotypescriptjavascript

Why we selected itThe catalogue has no Deno skill.

Weigh upThis review did not execute the skill or run Deno.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Security · GitHub community

Dependabot

Configures and manages Dependabot end to end: ecosystem detection across 25+ package managers, monorepo glob directories, grouped and multi-ecosystem updates, cooldown periods, ignore and allow rules, and PR command etiquette including the January 2026 deprecations.

dependabotdependenciessupply-chain

Why we selected itCovers the configuration surface teams actually trip on: plural directories with globs for monorepos versus the glob-less singular directory, grouped updates by type and name pattern, and cooldown periods that delay adopting freshly released versions.

Weigh upGitHub-product-specific: Renovate, GitLab dependency bots and other updaters share concepts but not this configuration surface.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Cloud operations · Microsoft

Deploy to AKS

Pins Microsoft's skill for deploying an existing web app to a running AKS cluster, which requires az login and refuses to provision the cluster.

azureakskubernetes

Why we selected itThe workflow line requires an existing AKS cluster, az login, and a configured kubectl before deployment.

Weigh upThe resource was not executed, and no Azure API, CLI, or MCP tool was called.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Software architecture · addyosmani

Deprecation and Migration Planning

Plan incremental retirement of APIs, systems and database schema with consumer migration evidence.

software-architectureaddyosmani

Why we selected itThe guide distinguishes advisory from compulsory deprecation and assigns migration responsibility to infrastructure owners rather than merely announcing a deadline.

Weigh upA rollback deployment cannot undo arbitrary data loss; zero observed traffic does not prove no offline or external consumer remains.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Data pipelines · Confluent

Developing Kafka Python client

Confluent skill that scaffolds a confluent-kafka-python producer, consumer or both, or adds Schema Registry to existing Python code, for Confluent Cloud, local Docker or WarpStream, after a confirmation step its instructions require.

streamingpythonconfluent

Why we selected itSKILL.md tells the agent that Schema Registry is non-negotiable in every project, to register schemas in an explicit register_schema() step, and to configure serializers with auto.register.schemas=False and use.latest.version=True.

Weigh upThe skill directory skills/developing-kafka-python-client last changed in a commit dated 2026-08-24; the pinned revision is dated 2026-10-06. SKILL.md metadata says version 1.0.0.

Source reviewed

9 sources · 2026-10-21

Read the full review Upstream setup linkedApache-2.0

Testing and reliability · Matt Pocock

Diagnosing Bugs

A debugging procedure that establishes a reproducible symptom, tests competing explanations and preserves a regression check.

debuggingreproductionregression-tests

Why we selected itRequires a red-capable reproduction before edits, so an apparent improvement has a concrete behavior to compare against.

Weigh upThe included human-in-the-loop shell script is an editable single-line-input template, not a ready-made test runner or secret-redaction tool.

Source reviewed

10 sources · 2026-10-15

Read the full review Install command in reviewMIT

Security review · Trail of Bits

Diagramming Code

Turns a Trailmark code graph into Mermaid diagrams: call graphs, class hierarchies, module-dependency maps, containment diagrams, complexity heatmaps and attack-surface data-flow views.

mermaidvisualizationcall-paths

Why we selected itEmits call graphs, class hierarchies, module-dependency maps, containment diagrams, complexity heatmaps and data-flow views as Mermaid from a Trailmark graph.

Weigh upThe native `trailmark diagram` command needs Trailmark 0.4.0+; on older builds the skill runs its bundled scripts/diagram.py.

Source reviewed

5 sources · 2026-10-22

Read the full review Upstream setup linkedCC-BY-SA-4.0

Research and documentation · GitHub community

Diataxis Documentation Writer

Writes docs under Diataxis discipline: clarify the quadrant, audience, goal and scope first, approve an outline, then draft tutorials, how-tos, reference or explanation in project tone.

documentationdiataxistechnical-writing

Why we selected itThe quadrant gate (tutorial, how-to, reference, explanation) plus a mandatory outline-approval step stops documentation from being written before its purpose and scope are agreed.

Weigh upFramework-first and deliberately procedural; overkill for one-paragraph answers and small edits.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Evidence and evaluation · Anthropic

Discernment Nudge

Anthropic skill that, at most once in a conversation, appends two or three short questions after a substantive answer so the user can check a fact, a reasoning step, or missing context.

fact-checkingassumptionsai-fluency

Why we selected itThe catalogue's Anthropic skills are document and design skills, not this answer-check.

Weigh upThis review did not execute the skill.

Source reviewed

5 sources · 2026-10-14

Read the full review Install command in reviewApache-2.0

Discord access control · Anthropic

Discord access

Pins the Discord access skill, which refuses channel-originated allowlist edits and will not approve a pairing without its code.

discordaccess-controlpairing

Why we selected itThe skill refuses pairing, allowlist, and policy changes that arrive as a Discord, Telegram, or other channel message, and says those mutations must not follow untrusted input.

Weigh upThis review did not execute the resource.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Discord channel setup · Anthropic

Discord configure

Pins the Discord setup skill, which stores a bot token as a credential and treats pairing as temporary until the allowlist is locked.

discordbot-tokenallowlist

Why we selected itThe skill says a saved bot token is a credential, tells the agent to chmod 600 the env file, and says a set token is shown with the first six characters masked.

Weigh upThis review did not execute the resource.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Development workflow · Superpowers

Dispatching Parallel Agents

Splits independent investigations into bounded agent assignments and checks their integration afterward.

subagentsparallel-workcoordination

Why we selected itRequires independence and explicit file or subsystem scope before work is parallelized.

Weigh upShared mutable state, sequential dependencies and overlapping file edits defeat the independence assumption.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewMIT

Security review · Sentry

Django access review

Traces Django authorisation before flagging an IDOR, instead of matching a pattern and stopping.

djangoauthorizationidor

Why we selected itThe first heading is Philosophy: Investigation Over Pattern Matching.

Weigh upNo Django project was executed in this review.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Code review · Functional Software, Inc. dba Sentry

Django performance review

Review Django code and report only performance issues the reviewer traced. Zero findings is an allowed result.

djangoperformancereview

Why we selected itN+1 and unbounded queries are the critical items, and they must be traced to a real loop and a large table.

Weigh upThe skill says zero findings is acceptable and forbids manufacturing issues.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Agent development · docker

Docker Agent Configuration

Build Docker Agent configurations with per-agent toolsets, provider credentials, delegated teams and bounded autonomy.

agent-developmentdocker

Why we selected itIt writes a Docker Agent agent.yaml, including models, toolsets, and sub-agents, and it does not cover running or serving the agent.

Weigh upThe skill does not cover docker agent run, serve, or share.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Agent deployment · Docker

Docker Agent deploy

Pins Docker's agent serve, share, and eval skill, which refuses a non-loopback listener that has no auth token.

dockeragentmcp

Why we selected itThe catalogue has docker-agent-run and docker-agent-config from this repository and not the serve, share, and eval skill.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Agent development · Docker

Docker Agent run

Pins Docker's skill for docker agent run, including the safety modes and the exception that a local stdio MCP server stays outside the sandbox VM.

dockeragentssandbox

Why we selected itThe catalogue has Docker's agent-config and sandbox-credential skills and not this run skill.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Containers · Docker

Docker Build Strategies

Teaches Docker BuildKit multi-stage builds, target execution, cache mount typing, and minimal distroless runtime patterns.

dockerbuildkitdockerfile

Why we selected itPrescribes explicit BuildKit cache mounts (--mount=type=cache,target=/root/.cache) and discrete build targets instead of invalidating entire layers on unrelated changes.

Weigh upThe build helper was inspected but not executed; the multi-stage and layer-caching reference guides were not re-reviewed.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Containers · Docker

Docker Compose Patterns

Establishes production patterns for compose.yaml: strict healthcheck dependencies, curl sidecars for scratch images, and secret masking.

dockerdocker-composeorchestration

Why we selected itEnforces condition: service_healthy on depends_on stanzas to prevent premature service crashes on unready databases and backends.

Weigh upCompose-managed volume deletion requires scoped confirmation; external volumes follow the standalone guardrails.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

States what a destructive Docker command will remove and waits for an explicit yes, except one narrow container cleanup.

dockerclisafety

Why we selected itPrune, kill, image, network, builder, context, and standalone volume deletes stay on the confirmation rule, including when the request is only to clean up.

Weigh upDocker stop interrupts a process but does not remove its container; the skill treats it separately from removal tiers.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Containers · docker

Docker Project Foundations

Create the first Dockerfile, Compose stack and dockerignore with local-port and build-secret boundaries.

containersdocker

Why we selected itIt creates a first Dockerfile, compose.yaml, and .dockerignore, and it mounts npm credentials instead of copying them into the image.

Weigh upThe skill says not to use it for a mature Docker setup that only needs a small edit.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Sandbox configuration · Docker

Docker Sandboxes environment file

Pins Docker's sbxenv skill, which treats sbx env plan as a no-change preview and will not default to auto-approving host lifecycle commands.

dockersandboxessbxenv

Why we selected itsbx env plan is described as printing the apply set and changing nothing, while create and run require approval before they do any work. The file says never to default to -y for a file or kit that has not been reviewed.

Weigh upA failing preRemove is only a warning. Removal can still stop afterwards when the destroy plan no longer matches. Those are different gates.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Sandbox kits · Docker

Docker Sandboxes kits

Pins Docker's kit-spec skill, which forbids secret kit arguments and says a schema check does not prove composition or that a host is blocked.

dockersandboxeskits

Why we selected itThe file forbids passing a secret as a kit argument, keeps OAuth off mixin kits, and calls token passthrough a security downgrade.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Container sandboxes · Docker

Docker Sandboxes lifecycle

Pins Docker's local sbx lifecycle skill, which requires a dry-run before prune and forbids mounting a secrets file into the sandbox.

dockersandboxessbx

Why we selected itThe catalogue already has Docker's destructive-guardrails and network-credentials skills and not this lifecycle file.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Configure standalone sbx egress policy and service credentials while distinguishing proxy sentinels, OAuth passthrough and host-side secret resolution.

sandbox-securitydocker

Why we selected itIt separates Docker Sandboxes egress policy from credential injection, and storing a secret does not open the network.

Weigh upEgress policy does not protect a credential that has already left the sandbox.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Documentation and knowledge · Addy Osmani

Documentation and ADRs

Records an architecture decision with a status, instead of leaving the reason in chat.

adrdocumentationdecisions

Why we selected itBefore creating an ADR it asks the agent to find and match the repository's existing convention (location, numbering, headings) and to surface conflicting conventions.

Weigh upAn ADR is not evidence the decision was correct.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Documentation · GitHub community (awesome-copilot)

Documentation Drift Audit

Compare documentation claims with repository code and configuration using a bounded evidence report.

documentationgithub

Why we selected itdocs_drift.py checks command names, filesystem links and regex-derived env references; the definition requires manual confirmation and separates inferred gaps from confirmed drift.

Weigh upThe helper is heuristic: it merges package scripts across a repository, approximates imports and returns zero despite findings. It reads complete local env-file text even though only variable names are reported.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Databases · DuckDB

DuckDB read file

DuckDB's read-file skill: one DuckDB CLI command that picks a reader by file extension and returns a file's schema, row count and first 20 rows, for local paths or HTTPS, S3, GCS and Azure URLs.

duckdbparquetsql

Why we selected itDuckDB publishes it in duckdb/duckdb-skills as one skill of the duckdb-skills Claude Code plugin (plugin.json version 0.2.4, license MIT); LICENSE is the MIT text with "Copyright 2018-2025 Stichting DuckDB Foundation".

Weigh upThe skills/read-file directory last changed in a commit dated 2026-03-30; the pinned revision 7feda8e0 is dated 2026-04-14.

Source reviewed

5 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Security · trailofbits

DWARF Debug Information

Inspect DWARF symbols, types, line tables and address mappings using tool-specific commands.

securitytrailofbits

Why we selected itThe skill reads DWARF with local dumpers and, when the standard matters, tells the agent to look up dwarfstd.org rather than answer from memory.

Weigh upallowed-tools includes WebSearch and WebFetch, so a standards lookup is a network call. The file does not attach a debugger or decompile the whole binary.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Release engineering · Expo

EAS Update

Publishes an Expo OTA update only after the channel, environment, runtime and native-build boundary are explicit.

expootaeas

Why we selected itPublishing is treated as a remote mutation that needs an explicit project, channel, environment, platform, runtime, and message.

Weigh upThe file says EAS Update has plan allowances and that publishing can affect installed apps.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Migration · Effectful Technologies

Effect v3 to v4

Migrates Effect v3 to v4 by searching the generated reference one symbol at a time, and forbids a compatibility shim.

effecttypescriptmigration

Why we selected itpackage.json is fixed before the type-check, merged packages are removed, and every remaining Effect package shares one version.

Weigh upThe two clones are unpinned main and the v3 branch. An existing .repos/effect must be the canonical Effect-TS/effect repo at version 4.x, or it is deleted. This review did not clone either repo.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Observability · elastic

Elasticsearch Cluster Health

Triage unassigned Elasticsearch shards through allocation evidence.

observabilityelastic

Why we selected itThe process separates reassignment triggers such as NODE_LEFT from the decider currently blocking allocation and treats missing valid primary copies explicitly.

Weigh upNot applicable to Serverless cluster APIs. Green allocation status does not establish application health, and the most likely cause remains a diagnosis to validate.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Observability · Elastic

Elasticsearch ES|QL

Pins Elastic's ES|QL skill, which routes queries through the elastic CLI and refuses to treat a Serverless build number as a feature gate.

elasticsearchesqlobservability

Why we selected itThe catalogue has Elastic's LLM observability skill and not this ES|QL skill from the same repository.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Data pipelines · elastic

Elasticsearch File Ingest

Load CSV or JSON into Elasticsearch with explicit mapping and bulk-result checks.

data-pipelineselastic

Why we selected itThe definition distinguishes CSV rows, JSON arrays and NDJSON and forbids sending raw arrays as one bulk document; index-to-index migration is outside this skill.

Weigh upA successful bulk HTTP response can contain failed items. Document counts alone cannot prove content equivalence, especially when IDs overwrite existing documents.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Search and retrieval · elastic

Elasticsearch Index Mapping Design

Design index field types and storage settings from concrete access patterns.

search-and-retrievalelastic

Why we selected itThe workflow separates full-text, exact-match and retrieve-only fields and calls out flattened fields as an alternative to unbounded dynamic mappings.

Weigh upChanging existing field types requires a new index/reindex strategy; Serverless manages shard and replica settings. Type choices need actual precision and retention requirements.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Search and retrieval · elastic

Elasticsearch Query DSL Optimization

Use query profiles and mappings to investigate slow Elasticsearch searches.

search-and-retrievalelastic

Why we selected itThe workflow profiles before recommending infrastructure changes and distinguishes query-clause cost from aggregation cost.

Weigh upProfile timings add overhead and omit parts of end-to-end latency. Removing a leading wildcard changes matching unless the intended contract permits it; wildcard-field behavior needs version-specific evidence.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Data pipelines · elastic

Elasticsearch Reindex Planning

Prepare destination mappings and verify asynchronous local or remote reindex operations.

data-pipelineselastic

Why we selected itThe definition separates local slicing from remote query partitioning and requires explicit destination preparation and post-copy settings restoration.

Weigh upReindex does not copy mappings or continuously synchronize later writes. Disabling replicas or refresh changes availability/visibility; counts need concurrent-write and ID-collision context.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Search and retrieval · elastic

Elasticsearch Search Relevance

Choose deterministic query rules or organic ranking changes from mappings and example searches.

search-and-retrievalelastic

Why we selected itThe decision table separates pinned/excluded documents from multi_match and analyzer tuning and warns against simulating deterministic pinning with extreme boosts.

Weigh upTop-hit inspection for one query is not a relevance benchmark. Query-rule availability and rule-query wiring depend on version and deployment.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Security review · Trail of Bits

Entry-point analyzer

Lists state-changing smart-contract entry points and their access class, and excludes view functions, exploit writing, and non-contract code.

smart-contractsauditentry-points

Why we selected itWhen NOT to Use sends vulnerability hunting, exploit proofs, gas work, and non-contract code elsewhere.

Weigh upThe skill does not detect vulnerabilities and tells the reader to use a different audit skill for that.

Source reviewed

2 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

API design · JetBrains

Error Model Validation Architect

Turns ad-hoc exception handling into a deliberate public error contract: an explicit taxonomy across eight failure categories, status-code mapping rules, machine-readable codes separated from human messages, framework exception coverage, and a logging rule that keeps internal detail out of client payloads.

error-handlingerror-taxonomyhttp-status-codes

Why we selected itForces an explicit error taxonomy before any code — at minimum eight categories from malformed request to downstream dependency failure — and forbids collapsing every non-success outcome into one generic envelope.

Weigh upThe worked context is Kotlin plus Spring (Bean Validation targets, @ControllerAdvice, security entry points); the taxonomy method transfers, the framework specifics do not.

Source reviewed

6 sources · 2026-10-13

Read the full review Install command in reviewMIT (upstream Kotlin backend agent skills)

Model evaluation · LangChain

Eval Engineering

Turns an agent repository and traces into reviewed task specs, verifiers and Harbor runs, with a cost confirmation before scored runs.

agent-evalsbenchmark-designharbor

Why we selected itBefore a scored model run, the file requires confirming the model, trial count, judge, timeout and maximum expected cost with the user, unless that run plan was already authorized.

Weigh upThe workflow routes to twelve reference documents, multi-turn simulation code and two scripts in the skill directory. Only SKILL.md was read for this review.

Source reviewed

3 sources · 2026-10-13

Read the full review Install command in reviewMIT

Development workflow · getsentry

Evidence-Based Skill Writing

Choose a skill structure, preserve source adaptation decisions and iterate against working and holdout examples.

development-workflowgetsentry

Why we selected itSKILL.md is a router: each bundled reference is loaded only for the step that needs it, not all at once.

Weigh upquick_validate.py needs Python 3.12 and pyyaml, and it is invoked with uv. A passing run does not mean the new skill does the task.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Mobile development · expo

Expo Data Fetching

Design Expo network flows with distinct loading, error, empty and cached-content states.

mobile-developmentexpo

Why we selected itThe definition explicitly distinguishes resolved empty data from missing data and keeps cached content visible when background refresh fails.

Weigh upClient-public environment values are bundled into the app. Retry examples require idempotency review, and offline-paused queries can have neither data nor an active loading flag.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Mobile development · Expo

Expo Design System

Builds a token theme (color, spacing, typography, radius, shadow, motion) plus variant/size/state component conventions inside an Expo app, with grep-based drift audits and a named gallery of AI-generated-looking anti-patterns.

design-systemdesign-tokensreact-native

Why we selected itAdopt-before-you-build rule: extend an existing styling library in its own idiom and never introduce a second token system beside it, so adoption cannot become drift.

Weigh upWhat looks native (palettes, HIG rules, controls) is delegated to the sibling expo-native-ui skill; this skill governs where values live, not which values to pick.

Source reviewed

7 sources · 2026-10-15

Read the full review Install command in reviewMIT

Mobile development · expo

Expo Development Clients

Build a native-capable Expo development client for device and simulator testing.

mobile-developmentexpo

Why we selected itThe source separates custom native dependencies from Expo Go use and distinguishes a development client from production store submission.

Weigh upEAS/TestFlight availability and billing depend on account and plan; the source cloud-build price generalization is not a current quote. iOS device archives and simulator artifacts are different build targets.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Mobile development · expo

Expo DOM Components

Use isolated DOM components to bring selected web UI into native Expo apps.

mobile-developmentexpo

Why we selected itThe definition separates serializable props from async native-action callbacks and documents different web and native rendering paths.

Weigh upA webview adds memory and bridge overhead; native routing hooks that require synchronous state must be evaluated in the parent. Exposed native callbacks require input validation.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Native modules · Expo

Expo Module Authoring

Scaffolds an Expo native module with an explicit slug, platform and feature set, then replaces the generated sample.

exposwiftkotlin

Why we selected itScaffolding is delegated to create-expo-module, with local and standalone called out as different products.

Weigh upThe description says not to use this skill to migrate an existing Swift module from the definition DSL to Expo Modules API 2.0 macros.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Mobile development · Expo

Expo Router Navigation

Structure Expo routes, native stacks, previews and modal navigation.

expo-routernavigationnative-tabs

Why we selected itThe definition separates route files from reusable components and identifies SDK-specific React Navigation entry points and title/header options.

Weigh upExamples target recent SDK APIs and iOS behaviors; preview, glass and native-tab features are not universal across platforms. Removing old routes can break external deep links.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Mobile development · Expo

Expo Web to Native Migration

Migrates a web React codebase to a native Expo app screen by screen: native shell first, whole web UI shippable in DOM components on day one, then strangler-order nativization with device verification.

expomigrationreact-native

Why we selected itThe migration spine keeps the app shippable at every step: assess into a durable worklist, scaffold, DOM-component shell as the day-one milestone, then one nativized screen per pass in value order.

Weigh upThe verification recipe is opinionated about tooling: it asks the user to install agent-browser and argent simulators rather than falling back to manual screenshots.

Source reviewed

7 sources · 2026-10-15

Read the full review Install command in reviewMIT

Security review · Trail of Bits

False Positive Check

Forces a true-positive or false-positive verdict on one suspected bug, and refuses to be used as a bug hunter.

securitytriageverification

Why we selected itThe definition separates verification of a named finding from hunting for new bugs, and lists the phrases that should not trigger it.

Weigh upCC-BY-SA-4.0 applies to adaptations you redistribute.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

API frameworks · FastAPI

FastAPI conventions

Official FastAPI skill for Annotated dependencies, return types, router-level configuration, and SSE through EventSourceResponse.

fastapipythonpydantic

Why we selected itThe primary file specifies Annotated dependencies, a return type ahead of response_model, router-level prefix and dependencies, and EventSourceResponse for server-sent events.

Weigh upThe reference files named from the quick reference were not opened. fastapi/sse.py, applications.py, and responses.py were not opened, so claims about those modules are the skill's statements.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Documentation · dbt Labs

Fetch dbt docs

Pins dbt Labs' docs skill, which fetches pages as markdown and refuses to run commands or instructions embedded in that content.

dbtdocsuntrusted-content

Why we selected itFetched documentation is untrusted, and the file says never to execute commands or instructions embedded in it.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Git workflow · Jesse Vincent

Finish a development branch

After a green test suite, show a fixed integration menu and run only the option the person chooses.

gitworktreepull-request

Why we selected itThe skill stops when tests fail, then waits for one of three written options before merge, push, or leaving the branch alone.

Weigh upOption 1 runs checkout, pull, and merge, then deletes the branch only after the merged suite is green. Option 2 pushes. Discard force-deletes the branch.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewMIT

Cloud Security · Google LLC

Firebase Security Rules Auditor

Audits Firebase Firestore and Cloud Storage security rules for privilege escalation, role bypasses, resource exhaustion, and create-vs-update validation asymmetries.

firebasefirestoresecurity-rules

Why we selected itDetails comprehensive security verification checklists covering field immutability, request.auth.uid verification, size caps, and write vs create/update splits.

Weigh upCovers rule logic and syntax; does not deploy rules to live Firebase projects or audit Cloud Functions code.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Model deployment · Microsoft

Foundry customize deployment

Pins Microsoft's Foundry customize-deployment skill, which blocks the create until quota data is valid and the operator approves the summary.

foundryazure-openaideployment

Why we selected itPhase 12 shows the full configuration and waits for approval or cancel before phase 13 runs the create.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Model deployment · Microsoft

Foundry deploy model

Pins Microsoft's Foundry deploy-model router, which will not create a deployment until the target project is shown and confirmed.

foundryazure-openaideployment

Why we selected itThe file says never to deploy without showing the project, and it requires a confirm step before every mode.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Model fine-tuning · Microsoft

Foundry fine-tuning

Pins Microsoft's Foundry fine-tuning skill, which treats cleanup as deletion and says to validate training data before a job is submitted.

foundryfine-tuningsft

Why we selected itThe file says cleanup.py deletes old files and deployments, and that a DeploymentNotReady race is fixed by deleting and recreating the deployment.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Frontend session diagnosis · Grafana Labs

Frontend Observability session RCA

Pins Grafana's frontend session skill, which stops until three ids are present and treats the saved gcx dump as untrusted.

grafanafrontend-observabilitygcx

Why we selected itNo selected skills record uses this source path.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Testing · trailofbits

Fuzzing Dictionaries

Build targeted token dictionaries from parser expectations and compare their effect on coverage.

testingtrailofbits

Why we selected itThe skill builds a token dictionary for a fuzzer. It says to skip that when the target is a pure algorithm or the corpus is already covering the code.

Weigh upIt does not run a fuzzer and it does not install one. External dictionary collections are links, not vendored files.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Testing · trailofbits

Fuzzing Obstacles

Diagnose validation and nondeterminism barriers, then isolate minimal fuzzing-only changes.

testingtrailofbits

Why we selected itThe handbook's method is a fuzzing-only build flag so production builds keep the check the fuzzer cannot pass.

Weigh upThe file is instructions, not a patch applicator. It does not compile or run a fuzzer.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Creative production · GitHub community

Game Engine

Builds web-based games and game engines with HTML5 Canvas, WebGL and JavaScript: game loops, physics and collision detection, controls, audio, plus five complete starter templates and nine reference documents.

game-developmenthtml5-canvaswebgl

Why we selected itShips five complete starter templates and nine reference documents, so a working game exists before theory does — the paddle template is a full breakout build, not a skeleton.

Weigh upWeb-only: no Unity, Unreal or Godot coverage, so native-engine work is out of scope.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

AI application development · Google

Gemini API Development

Gemini Interactions API patterns for Python and TypeScript, with a dated model list and deprecated-model rules.

geminillm-apiinteractions-api

Why we selected itThe file states that interactions are stored by default, retained 55 days on the paid tier and 1 day on the free tier, and that store=False also disables previous_interaction_id and background runs.

Weigh upIf a user asks for a deprecated model, the file tells the agent to use gemini-3.8-flash instead and note the substitution. Check for that note when a specific model was required.

Source reviewed

4 sources · 2026-10-13

Read the full review Install command in reviewApache-2.0

AI application development · Google

Gemini Live API Development

Gemini Live API guidance for real-time audio, video and text over WebSockets, with stated session limits.

geminirealtime-audiowebsockets

Why we selected itIts Limitations section states session ceilings without compression (15 minutes audio-only, 2 minutes audio plus video, about 10 minutes per connection) and points to context compression and session resumption.

Weigh upThe quick-start authentication examples pass a literal API key placeholder to the client, including in JavaScript, which the file's own best practices advise against for browsers. Use ephemeral tokens in client code.

Source reviewed

5 sources · 2026-10-13

Read the full review Install command in reviewApache-2.0

Testing and reliability · Trail of Bits

Genotoxic

Builds a Trailmark graph, runs a mutation-testing framework and, where supported, necessist, and uses survived mutants, unnecessary test statements and call-graph data to sort findings into false positives, missing unit tests and fuzzing targets.

mutation-testingtriagefuzzing

Why we selected itBuilds a Trailmark graph, runs mutation testing and, where the language allows, necessist, then classifies survived mutants and unnecessary test statements as false positives, missing unit tests or fuzzing targets.

Weigh upIt installs missing tools itself: `uv tool install trailmark`, a mutation framework per references/mutation-frameworks.md (sudo apt-get or dnf, brew, cargo install or go install …@latest), and `cargo install necessist`.

Source reviewed

6 sources · 2026-10-22

Read the full review Upstream setup linkedCC-BY-SA-4.0

CI and delivery · GitHub community (awesome-copilot)

GitHub Actions Efficiency Review

Find evidenced CI waste while preserving required validation and measuring actual impact.

ci-and-deliverygithub

Why we selected itThe reporting reference separates expected from measured savings, runner time from critical-path latency and cache-warmup or matrix changes from optimization effects.

Weigh upStatic workflow inspection cannot prove cancellation, gating or savings. Undocumented matrix legs may still protect real users, so confirm compatibility commitments before removal.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

CI and delivery · GitHub community (awesome-copilot)

GitHub Actions Runtime Upgrades

Upgrade deprecated action runtimes using compatible releases and immutable commit pins.

ci-and-deliverygithub

Why we selected itThe definition requires full-SHA uses references, preserved workflow behavior and verification of artifact/signing outputs and post-upgrade logs.

Weigh upLatest major does not imply compatibility with self-hosted runners or existing inputs. Local build success does not prove the upgraded action executes correctly on GitHub.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Security review · Sentry

GitHub Actions Security Review

Reviews GitHub Actions for an external attacker, and withholds a finding that has no entry point, payload and execution path.

github-actionscisupply-chain

Why we selected itThe attacker is someone without write access. workflow_dispatch input injection and push-only expression injection on protected branches are out of scope.

Weigh upThe file treats ${{ }} in if:, with: and job or step env: as runtime evaluation rather than a shell sink. A later unquoted use of that variable can still be ordinary shell injection.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Planning and delivery · GitHub community

GitHub Release Workflow

Cuts a library release end to end: tag-derived versioning, code-diff-first SemVer analysis against commit intent, Keep a Changelog authoring, and a gated release-branch plus PR handoff.

releasessemverchangelog

Why we selected itVersion analysis prefers the code diff as primary signal with commit messages secondary, and states the reconciliation rule outright: on conflict the diff wins, with worked examples.

Weigh upSingle-package GitHub repositories are assumed; monorepos and non-SemVer tag schemes fall outside the versioning logic.

Source reviewed

7 sources · 2026-10-15

Read the full review Install command in reviewMIT

Development workflow · GitHub community (awesome-copilot)

GitHub Secret Scanning Operations

Configure GitHub secret detection and triage exposure with credential rotation and explicit bypass decisions.

development-workflowgithub

Why we selected itIt configures GitHub secret scanning and push protection, and it treats rotation as the first remediation step.

Weigh upThe skill does not scan the working tree itself; pre-commit scanning is a separate plugin and MCP tool.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Design and interfaces · Google Labs / Stitch

Google Stitch Design-System Extraction

Reads frontend source and theme files to draft a structured DESIGN.md for design-system documentation or later Stitch use.

design-systemsdesign-tokenscss

Why we selected itIt writes a DESIGN.md by reading stylesheets, theme files, and component source, and it says no build or runtime is required.

Weigh upThe skill does not build or render the application, and it does not call Stitch unless the user wants the document pushed.

Source reviewed

10 sources · 2026-10-15

Read the full review Upstream setup linkedApache-2.0

Development workflow · Hugging Face

Gradio App Builder

A Gradio reference and example collection for building Python interfaces, event-driven layouts and chat demos.

gradiopythondemos

Why we selected itConnects component signatures and event listeners to complete examples covering forms, streaming and chat interfaces.

Weigh upThe static signatures are a source snapshot, not a version compatibility guarantee; check current Gradio documentation and your installed package.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewApache-2.0

Spaces · Hugging Face

Gradio LoRA Space Builder

Pins Hugging Face's LoRA Space skill, which tries a public read before asking for a token and waits for one approval before a private publish.

huggingfaceloragradio

Why we selected itAn unauthenticated read comes first. A token is requested only when no valid cached token can read a private or gated repo.

Weigh upThis review did not execute the resource.

Source reviewed

3 sources · 2026-10-14

Read the full review Upstream setup linkedApache-2.0

Grafana alerting · Grafana Labs

Grafana alerting and IRM

Pins Grafana's alerting and IRM skill, which requires a bearer token, checks the receiver before live routing, and silences notifications without stopping evaluation.

grafanaalertingsilences

Why we selected itNo selected skills record uses this source path.

Weigh upThe resource was not executed.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Observability · Grafana Labs

Grafana Alloy

Write a Grafana Alloy pipeline, validate it before running it, and treat an unhealthy component on the local UI as the failure.

alloyopentelemetrylogs

Why we selected itThe file runs alloy fmt and alloy validate before alloy run, then reads the local component API for any health state other than healthy.

Weigh upThe config-syntax reference was inspected; components.md and collection-patterns.md remain outside this review. alloy fmt, alloy validate, and alloy run were not executed.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Observability · grafana

Grafana App Admission Control

Implement validators and mutators for Grafana app resources before persistence.

observabilitygrafana

Why we selected itIt separates grafana-app-sdk validation, which cannot change the object, from mutation, which can.

Weigh upA validator cannot modify the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Observability · grafana

Grafana App Reconciler Logic

Implement scoped resource reconciliation and status updates with explicit event and finalizer behavior.

observabilitygrafana

Why we selected itIt implements a grafana-app-sdk TypedReconciler that skips an unchanged generation and writes status with UpdateObject.

Weigh upThe skill does not use client.Update for status.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Grafana app scaffolding · grafana

Grafana App SDK concepts

Pins Grafana's app-sdk concepts skill, which asks before adding an operator and deletes unkept files in the in-repo scaffold.

grafanaapp-sdkscaffold

Why we selected itNo selected skills record uses this source path.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Metrics cost control · grafana

Grafana Cloud Adaptive Metrics

Pins Grafana's Adaptive Metrics skill, which separates a write bearer key from query basic auth and deletes a rule to roll back.

grafanaadaptive-metricscardinality

Why we selected itNo selected skills record uses this source path.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Grafana Cloud administration · grafana

Grafana Cloud Admin

Pins Grafana's cloud admin skill, which labels stack deletion as destructive and checks a new token on the org endpoint.

grafanaadminrbac

Why we selected itNo selected skills record uses this source path.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Application observability · grafana

Grafana Cloud Application Observability

Pins Grafana's app observability skill, which requires publisher scopes and treats Faro 401 and 404 as different failures.

grafanaapmfaro

Why we selected itNo selected skills record uses this source path.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Grafana Cloud cost control · grafana

Grafana Cloud Cost Management

Pins Grafana's cost-management skill, which says to preview Adaptive Metrics rules before apply and to disable or delete them to roll back.

grafanacostadaptive-metrics

Why we selected itNo selected skills record uses this source path.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Prometheus cost analysis · grafana

Grafana Cloud dpm-finder

Pins Grafana's dpm-finder skill, which checks a metrics-read key before the scan and skips metrics that return HTTP 422.

grafanaprometheusdpm

Why we selected itNo selected skills record uses this source path.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Cloud provider integrations · grafana

Grafana Cloud Integrations

Pins Grafana's cloud integrations skill, which scopes provider credentials and refuses in-place edits of provisioned dashboards.

grafanaawsazure

Why we selected itNo selected skills record uses this source path.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Grafana Cloud k6 · Grafana Labs

Grafana Cloud k6 investigation

Pins Grafana's Cloud k6 investigation skill, which reads with authenticated gcx and forbids using the script-update command as a way to discover URLs.

grafanak6gcx

Why we selected itNo selected skills record uses this source path.

Weigh upThe resource was not executed.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Grafana MCP access · grafana

Grafana Cloud MCP setup

Pins Grafana's MCP setup skill, which defaults agents to --disable-write and a Viewer token.

grafanamcpdisable-write

Why we selected itNo selected skills record uses this source path.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Private connectivity · Grafana Labs

Grafana Cloud private connectivity

Pins Grafana's private connectivity skill, which limits AWS PrivateLink to one region unless peering exists and rejects a public DNS answer.

grafanaprivatelinkdns

Why we selected itNo selected skills record uses this source path.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Synthetic and load testing · Grafana Labs

Grafana Cloud testing

Pins Grafana's Cloud testing skill, which runs a local k6 smoke test before a cloud launch and pairs check creation with a rollback.

grafanasynthetic-monitoringk6

Why we selected itNo selected skills record uses this source path.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Observability · grafana

Grafana CUE Kind Definitions

Define versioned resource schemas and generate matching app types from CUE.

observabilitygrafana

Why we selected itIt scaffolds a grafana-app-sdk kind in CUE and regenerates clients instead of editing pkg/generated.

Weigh upGenerated Go and TypeScript files are overwritten and must not be hand-edited.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Grafana dashboards · Grafana Labs

Grafana dashboard authoring

Pins Grafana's dashboard JSON skill, which checks the payload locally before POST and sends it with a dashboards write token and overwrite set.

grafanadashboardsoverwrite

Why we selected itNo selected skills record uses this source path.

Weigh upThe resource was not executed.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Grafana datasource provisioning · Grafana Labs

Grafana datasource provisioning

Pins Grafana's datasource provisioning skill, which asks before generating a file, refuses inline secrets and pasted tokens, and defaults query access to the server proxy.

grafanadatasourcesprovisioning

Why we selected itNo selected skills record uses this source path.

Weigh upThe resource was not executed.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Observability · Grafana

Grafana Fleet Management

Pins Grafana's Alloy fleet skill, which validates a pipeline locally before the Fleet Management API and keeps the cloud API key in an environment variable.

grafanaalloyfleet-management

Why we selected itThe catalogue has Grafana Alloy, Loki, and Prometheus skills and not this Fleet Management remote-config skill.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Observability · Grafana Labs

Grafana Loki

Pins Grafana's Loki skill for LogQL stream selectors, parsers, and the push and read paths it names.

lokilogqlobservability

Why we selected itThe catalogue has Grafana Alloy, PromQL, and Mimir skills and not this Loki skill.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Observability · Grafana Labs

Grafana Mimir

Stand up Grafana Mimir for long-term Prometheus metrics, with the demo's single replica called out and the tenant header required once multitenancy is on.

mimirprometheusmetrics

Why we selected itThe monolithic demo sets replication_factor to 1, and the file says filesystem storage is only for development.

Weigh upThe Docker example uses grafana/mimir:latest, which is not a pinned image. The Helm example names mimir-distributed 6.0.6; that chart version was not checked against the chart repository.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

On-call routing · Grafana Labs

Grafana OnCall and IRM

Pins Grafana's OnCall and IRM skill, which previews a routing template before a live alert and authenticates schedule calls with a token.

grafanaoncallirm

Why we selected itNo selected skills record uses this source path.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Grafana OSS administration · Grafana Labs

Grafana OSS configuration

Pins Grafana's OSS skill, which authenticates API checks with a bearer token, separates 401 from 403, and shows a service-account token with no expiry.

grafanaossbearer-token

Why we selected itNo selected skills record uses this source path.

Weigh upThe resource was not executed.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Observability · Grafana Labs

Grafana Pyroscope

Select SDK or Alloy profiling and inspect service-specific CPU, allocation and contention profiles.

pyroscopeprofilingobservability

Why we selected itIt instruments an application for Grafana Pyroscope, either against a local server on port 4040 or Grafana Cloud with a numeric user and API key.

Weigh upThe skill does not grant Cloud access; it requires a numeric Pyroscope user and an API key.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Observability · Grafana Labs

Grafana Tempo

Local and Helm setup notes plus TraceQL for a Tempo tracing backend.

tempotraceqltracing

Why we selected itThe file separates a localhost Docker Compose check from a Helm install and names the object-storage bucket as configuration, not as a secret in the command.

Weigh upThe Docker example clones github.com/grafana/tempo with no revision, and the Helm example pins chart version 1.61.3. Neither command was run.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Security review · Trail of Bits

Graph Evolution

Diffs Trailmark code graphs built at two source snapshots (commits, tags or directories) to surface new attack paths, blast-radius growth, taint changes and privilege-boundary shifts that a text diff does not show.

diff-reviewblast-radiusgraph

Why we selected itDiffs Trailmark graphs at two snapshots (commits, tags or directories) to surface new attack paths, blast-radius growth, taint and privilege-boundary changes.

Weigh upAn empty `trailmark diff` can mean the wrong language rather than no change, because --language defaults to python; the skill says to pass it explicitly and re-run.

Source reviewed

5 sources · 2026-10-22

Read the full review Upstream setup linkedCC-BY-SA-4.0

Testing · Trail of Bits

Harness writing

Designs a libFuzzer or cargo-fuzz harness that accepts every input size, stays deterministic, and does not put unrelated formats in one target.

fuzzingharnessrust

Why we selected itThe harness must accept empty, tiny, huge, and malformed inputs; crashes in the system under test are the signal, crashes in the harness are not.

Weigh upThe arbitrary crate is described as having no reverse serialization, so hand-built byte strings for a specific struct are a poor fit.

Source reviewed

2 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Web performance · Microsoft

Heap Snapshot Analysis

Investigates memory leaks from V8 heap snapshots: parsing multi-hundred-megabyte .heapsnapshot files, comparing before/after object deltas, and tracing retainer paths while skipping the weak edges and debugger artifacts that produce false suspects.

memory-leaksheap-snapshotallocation-tracing

Why we selected itBakes anti-bias discipline into the workflow: the analyst is told to start fresh and ignore prior investigation notes so stale suspects cannot steer the current analysis.

Weigh upThe retention-pattern catalogue (RowCache, HoverService, ObjectMutationLog) is VS Code-specific; the method transfers, the named suspects do not.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Model development · Hugging Face

hf-mem Memory Estimate

Guides metadata-based estimates for model weight memory and optional KV-cache scenarios without first downloading the weight tensors.

vramggufsafetensors

Why we selected itKeeps an exact GGUF choice, context length, batch size and cache dtype visible in the estimation workflow.

Weigh uphf-mem describes itself as experimental; its estimate is not a measured peak allocation or a guarantee that a serving workload fits.

Source reviewed

5 sources · 2026-10-14

Read the full review Attributed definition availableApache-2.0

Web frameworks · Yusuke Wada

Hono

Inline Hono reference that chains routes for the RPC client, generates Cloudflare binding types, and snapshots a route before changing it.

honoroutingtypescript

Why we selected itRPC types require chained routes. Binding types are generated (npm run typecheck in cf CLI projects, cf-typegen or wrangler types in wrangler projects), and the file says not to install @cloudflare/workers-types.

Weigh upThe CLI install is @hono/cli@next, with a TODO to switch at the 1.0 release; the README Requirements line names @hono/cli without @next. That package was not installed and npx hono --help was not run. Live pages on hono.dev, which the file tells the agent to fetch with curl, were not fetched.

Source reviewed

5 sources · 2026-10-13

Read the full review Install command in reviewMIT

Agent development · Anthropic

Hook Development for Claude Code Plugins

Explains how to write Claude Code plugin hooks, prompt-based and command-based, with their event list, input and output JSON and exit codes, and ships Bash utilities to validate, lint and test them.

claude-codeplugin-devhooks

Why we selected itIt separates prompt hooks, which it recommends and lists for Stop, SubagentStop, UserPromptSubmit and PreToolUse, from command hooks for deterministic checks, and gives the hook input JSON, output JSON and exit codes (0 success, 2 blocking error fed back to Claude).

Weigh upLast commit touching plugins/plugin-dev/skills/hook-development at revision 41f32e40: 2026-02-05 (2438937b).

Source reviewed

10 sources · 2026-10-22

Read the full review Upstream setup linkedApache-2.0

Machine learning · huggingface

Hugging Face API Tool Builder

Create reusable CLI scripts that inspect Hub API shapes and compose metadata requests into JSON/NDJSON pipelines.

machine-learninghuggingface

Why we selected itIt builds a reusable script against the Hugging Face HTTP API and tells that script to send HF_TOKEN as a bearer header.

Weigh upThe skill creates a script; it is not itself an API client that this review ran.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewApache-2.0

Cloud access preflight · Hugging Face

Hugging Face AWS Context Discovery

Establish the selected AWS profile, region and caller before a cloud workflow. Useful input to credential-scoped execution, but not an authorization check or a sandbox.

awscredentialsidentity

Why we selected itCaptures which account and region a future inference or evaluation job would actually use instead of assuming a default from the model or task name.

Weigh upThe source's SSO warnings are heuristics, not permission evidence. An ARN pattern cannot prove that IAM writes are forbidden or allowed; inspect the applicable policies separately.

Source reviewed

6 sources · 2026-10-14

Read the full review Upstream setup linkedApache-2.0

Data and research · Hugging Face

Hugging Face Dataset Workflows

Discovers a Hub dataset's subsets, splits, rows and parquet shards, while separating read-only exploration from optional repository creation and uploads.

datasetshugging-faceparquet

Why we selected itRequires resolving config and split before row work, making the sampling population explicit.

Weigh upThe opening read-only scope does not cover later repository creation and upload examples. A private destination is still an external disclosure of the uploaded contents.

Source reviewed

7 sources · 2026-10-14

Read the full review Install command in reviewApache-2.0

Machine learning · huggingface

Hugging Face Hub CLI

Navigate the hf command surface for Hub repositories, local cache, datasets, jobs, buckets and endpoints.

machine-learninghuggingface

Why we selected itIt is a command map for the hf CLI, including auth, jobs, spaces, and a secrets list that does not return secret values.

Weigh upThe skill does not mount repositories; that is hf-mount, a different project.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewApache-2.0

Model development · Hugging Face

Hugging Face LLM Trainer

Prepares and submits managed training jobs with dataset checks, monitoring and explicit persistence of trained artifacts.

fine-tuningtrlhugging-face

Why we selected itTreats dataset shape, credentials and model persistence as prerequisites rather than leaving them until a job fails.

Weigh upThe skill directs immediate job submission for training requests; make the target, hardware and spending scope explicit before invocation.

Source reviewed

8 sources · 2026-10-13

Read the full review Install command in reviewApache-2.0

Model evaluation · Hugging Face

Hugging Face Local Model Evaluations

Chooses a local evaluation framework and inference backend, starting with a small smoke run before scaling evaluation work.

evaluationsinspect-ailighteval

Why we selected itSeparates local GPU inference from a locally launched client that calls remote Inference Providers.

Weigh upThe provider-backed helper still sends inference requests over the network; local script execution does not mean local inference.

Source reviewed

7 sources · 2026-10-13

Read the full review Install command in reviewApache-2.0

Research publishing · Hugging Face

Hugging Face paper publisher

Pins Hugging Face's paper skill, which requires a write token, separates a paper check from indexing, and says linking keeps existing README content.

huggingfacepapersarxiv

Why we selected itThe prerequisites require a write-access Hugging Face token, and a permission-denied result is explained as missing write access.

Weigh upThis review did not execute the resource.

Source reviewed

3 sources · 2026-10-14

Read the full review Upstream setup linkedApache-2.0

Data and research · Hugging Face

Hugging Face Paper Reading

Retrieves AI research-paper text and structured metadata, with fallbacks to the original arXiv page or PDF when Hub text is incomplete.

papersarxivliterature-research

Why we selected itSeparates paper text from metadata such as authors and linked code, models and datasets, helping a research task collect the right evidence.

Weigh upHub indexing and HTML availability vary; confirm that the fetched text is the actual paper before summarizing methods or results.

Source reviewed

6 sources · 2026-10-13

Read the full review Install command in reviewApache-2.0

Development tooling · huggingface

Hugging Face Python Environment Setup

Prepare an isolated Python environment for Hugging Face cloud tooling.

development-toolinghuggingface

Why we selected itsetup_env.py chooses uv or venv and checks the interpreter version; check_versions.py prints NOT INSTALLED without making that a failing exit status.

Weigh upThe setup helper deletes an existing environment directory when interpreter versions differ and upgrades requirements on reuse. The version-report helper returns success even for missing packages.

Source reviewed

6 sources · 2026-10-14

Read the full review Install command in reviewApache-2.0

Cloud operations · huggingface

Hugging Face SageMaker Deployment Planning

Select a SageMaker deployment path from workload, traffic and model constraints.

cloud-operationshuggingface

Why we selected itThe workflow discovers serving requirements before choosing infrastructure and hands off environment, IAM, image selection and production configuration explicitly.

Weigh upLatency and wake-up examples are planning assumptions, not measured guarantees. Sibling setup, IAM and deployment skills are separate dependencies and may not accompany a single-directory installation.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewApache-2.0

Cloud operations · huggingface

Hugging Face SageMaker Deployment Templates

Review endpoint creation, scaling and alarm templates as one deployment plan.

cloud-operationshuggingface

Why we selected it_common.py waits for InService but does not test inference; deploy.py creates scaling/alarms after the endpoint, while teardown.py ignores several deletion results.

Weigh upExisting-model reuse is based on an exception message without comparing configuration. Deployment failures can leave billable resources; teardown can print completion despite failed deletions, and alarms without SNS do not notify.

Source reviewed

6 sources · 2026-10-14

Read the full review Install command in reviewApache-2.0

Model training · Hugging Face

Hugging Face Sentence Transformer Training

Routes embedding and reranker training to model-specific templates, dataset/loss checks and a before-and-after evaluation.

embeddingsrerankingretrieval

Why we selected itUses task-specific losses and evaluators instead of one generic fine-tuning recipe.

Weigh upThe default full training template attempts a public Hub upload and uses Trackio reporting. Review publication, telemetry and data rights before running it with credentials.

Source reviewed

6 sources · 2026-10-14

Read the full review Install command in reviewApache-2.0

Deployment · Hugging Face

Hugging Face Spaces Operations

Guides creating, building, debugging, and maintaining ML applications on Hugging Face Spaces across Gradio, Docker, and Static SDKs.

huggingfacespaceszerogpu

Why we selected itDetails exact hardware tiers, ZeroGPU allocation constraints (@spaces.GPU), and quota limits (free vs Pro vs Enterprise orgs) without guessing.

Weigh upZeroGPU is restricted to Gradio Spaces; Docker Spaces cannot use ZeroGPU dynamically.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewApache-2.0

Model development · Hugging Face

Hugging Face Trackio

Adds training metrics and diagnostic alerts, then retrieves run data through Trackio’s CLI or a local or hosted dashboard.

experimentstraining-metricsalerts

Why we selected itSeparates logging, alerting and retrieval into focused references, including a route from an alert to metrics around the relevant training step.

Weigh upThe private flag only affects newly created Spaces; inspect an existing Space’s visibility before uploading metrics. Webhooks and remote synchronization send data outside the machine.

Source reviewed

9 sources · 2026-10-13

Read the full review Install command in reviewApache-2.0

Machine Learning · Hugging Face

Hugging Face Transformers.js

Runs state-of-the-art machine learning models directly in JavaScript and TypeScript with WebGPU and WebAssembly.

huggingfacetransformers-jswebgpu

Why we selected itCovers client-side and Node.js inference patterns with hardware acceleration (WebGPU) and broad fallback (WASM), eliminating Python server dependencies.

Weigh upWebGPU acceleration requires compatible browser versions and operating system GPU drivers; falls back to slower WASM when unsupported.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewApache-2.0

Machine learning · huggingface

Hugging Face Vision Training

Prepare dataset-validated detection, classification or segmentation training jobs with explicit Hub persistence and job-secret handling.

machine-learninghuggingface

Why we selected itIt submits object-detection, classification, or SAM training to Hugging Face Jobs and saves results on the Hub.

Weigh upThe skill says not to poll a running job automatically.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewApache-2.0

Machine learning · huggingface

Hugging Face ZeroGPU

Design Gradio GPU handlers around process isolation, serialized state, dynamic duration and concurrent requests.

machine-learninghuggingface

Why we selected itIt constrains Gradio Spaces on ZeroGPU: the real GPU exists only inside spaces.GPU, and CUDA tensors must not cross that boundary.

Weigh upDocker, Static, and Streamlit Spaces cannot schedule onto ZeroGPU.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewApache-2.0

iMessage access control · Anthropic

iMessage access

Pins the iMessage access skill, which refuses channel-originated edits and defaults a missing policy file to allowlist because pairing would auto-reply.

imessageaccess-controlallowlist

Why we selected itThe skill refuses access changes that arrive as an iMessage, Telegram, or Discord notification and will not approve a pairing unless the user supplies the code.

Weigh upThis review did not execute the resource.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

iMessage channel setup · Anthropic

iMessage configure

Pins the iMessage setup skill, which only reports status, checks Full Disk Access, and pushes pairing back toward an allowlist.

imessagefull-disk-accessallowlist

Why we selected itIf listing chat.db fails with Operation not permitted, the skill says Full Disk Access is not granted and the server cannot read that database.

Weigh upThis review did not execute the resource.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Planning and delivery · GitHub community

Incident Post-Mortem

Walks a team from incident metadata through a reconstructed timeline, blameless 5-Whys root cause, quantified impact and owned action items, into a complete postmortem document with a common-mistakes table.

postmortemincidentblameless

Why we selected itThe blameless principle is operational, not aspirational: a common-mistakes table rewrites 'Bob forgot to check the config' into 'the deploy checklist did not include config validation' and treats human error as a symptom, never a root cause.

Weigh upSingle self-contained SKILL.md with no references; severity scales and SLA definitions must come from the team's own incident process.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Identity and access · Infisical Inc.

Infisical Access Control and Governance Guide

Infisical skill for deciding what a user or machine identity may do inside Infisical: org and project roles, granular secret permissions, ABAC, temporary access, approval policies and audit log streams.

infisicalaccess-controlpermissions

Why we selected itIt separates describeSecret (seeing that a secret exists and its metadata) from readValue (seeing the value), and warns that the legacy read action grants both.

Weigh upApproval workflows and access requests are Enterprise tier on Infisical Cloud and need an enterprise license when self-hosted, per references/approvals-and-audit.md.

Source reviewed

7 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Security · Infisical Inc.

Infisical Agent Guide

Infisical skill for configuring the Infisical Agent, a client daemon that authenticates, renews tokens and renders secrets into files through Go templates.

infisicalsecretssidecar

Why we selected itIt covers the agent's YAML config, all 6 agent auth methods, sinks, polling and the template functions listSecrets, listSecretsByProjectSlug, getSecretByName and dynamicSecret.

Weigh updynamicSecret needs its trailing principals argument for SSH dynamic secrets, which the skill marks as required.

Source reviewed

6 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Security · Infisical Inc.

Infisical API Skill

Infisical skill for calling the Infisical REST API directly: Universal Auth login, /api/v4/secrets CRUD and batch calls, projects and identities, pagination and rate limits.

infisicalapisecrets

Why we selected itIt states that /api/v4/secrets is not paginated and ignores offset and limit, while identities, memberships and certificates paginate and return { <resource>: [...], totalCount: n }.

Weigh upIt covers the core secrets, projects and identities API; KMS, certificate, sync, rotation and App Connection endpoints are left to their own skills.

Source reviewed

6 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Security · Infisical Inc.

Infisical App Connections Guide

Infisical skill for creating App Connections, the stored credentials Infisical uses to authenticate out to third-party services for secret syncs, rotations, PKI and secret scanning.

infisicalcredentialsoauth

Why we selected itTwo of its reference files are generated from the Infisical source by tools/generate-app-connection-refs.py: the API surface, and the exact method values and credential fields for all 83 connections.

Weigh upOnly 16 connection types accept gatewayId; passing one to any other type is a validation error, not a no-op.

Source reviewed

6 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Security · Infisical Inc.

Infisical Dynamic Secrets Guide

Infisical skill for configuring dynamic secrets: short-lived credentials created per lease for databases, cloud IAM, SSH certificates, Kubernetes service accounts and other providers.

infisicalsecretscredentials

Why we selected itIt covers the providers its description lists, from PostgreSQL and Redis to AWS IAM, GCP IAM, Azure Entra ID, SSH certificates and TOTP.

Weigh upA database in a private network needs an Infisical Gateway, which the skill calls an Enterprise feature (Cloud Enterprise tier or self-hosted Enterprise license).

Source reviewed

7 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Infrastructure · Infisical Inc.

Infisical Gateway Guide

Infisical skill for deploying the Infisical Gateway and relays so that Infisical can reach private databases, directories, APIs and HSMs over outbound-only SSH reverse tunnels.

infisicalnetworkingegress

Why we selected itIt names the exact egress the gateway needs: TCP 2222 to the relay, TCP 443 to Infisical, and TCP 8443 platform-to-relay.

Weigh upGateway is a paid Enterprise feature, and Gateway Pools are an Enterprise feature too.

Source reviewed

7 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Security · Infisical Inc.

Infisical KMS Guide

Infisical skill for creating KMS keys and calling encrypt, decrypt, sign, verify and MAC operations with keys that stay inside Infisical.

infisicalcryptographyaws

Why we selected itIt fixes each key to one keyUsage (encrypt-decrypt, sign-verify or generate-verify-mac) and lists the operations each allows.

Weigh upRotating a key adds a new version for new operations and does not re-encrypt existing ciphertext.

Source reviewed

7 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Containers · Infisical Inc.

Infisical Kubernetes Operator Guide

Infisical skill for running the Infisical Kubernetes Operator: syncing secrets into Kubernetes Secrets and ConfigMaps, pushing secrets back to Infisical, and managing dynamic secret leases in-cluster.

infisicalkubernetessecrets

Why we selected itIt makes the agent establish the CRD API version first: v1beta1 splits the work into InfisicalConnection, InfisicalAuth and InfisicalStaticSecret, while legacy v1alpha1 puts it all in one InfisicalSecret.

Weigh upThe skill documents InfisicalDynamicSecret and InfisicalPushSecret only at v1alpha1, which it calls legacy and deprecating soon.

Source reviewed

7 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Security · Infisical Inc.

Infisical PKI

Pins Infisical's certificate-management skill and its rule to send SSH certificates to a different skill.

infisicalpkicertificates

Why we selected itThe catalogue has Infisical secret scanning and not this PKI skill.

Weigh upThis review did not execute the resource.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewMIT

Identity and access · Infisical Inc.

Infisical Privileged Access Manager (PAM) Guide

Infisical skill for brokering human and AI-agent sessions to databases, servers, Kubernetes clusters and cloud accounts through Infisical PAM, with recording and without handing over a credential.

infisicalaccess-controlapproval

Why we selected itIt covers agentic access, infisical pam agentic access -- <agent>, which opens one local port per account and has the Gateway inject the real credential so none reaches the agent.

Weigh upPAM is a separate Infisical product with its own project type and is an Enterprise feature.

Source reviewed

7 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Security · Infisical Inc.

Infisical Secret Rotation Guide

Infisical skill for rotating credentials that already exist in a third-party system on a fixed interval and writing the new values back into Infisical secrets.

infisicalcredentialssecrets

Why we selected itIt covers the rotation providers its description lists and makes the agent establish whether a provider is dual-phase or single-phase, because that changes the operational advice.

Weigh upRotation is an Enterprise / paid feature, per references/rotation-overview.md.

Source reviewed

7 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Security · Infisical Inc.

Infisical secret scanning

Infisical skill for scanning GitHub, GitLab, and Bitbucket history for leaked credentials, with GitHub tied to a github-radar connection and rotation required before history is rewritten.

infisicalsecretsscanning

Why we selected itThe catalogue has a Vault MCP server and no Infisical secret-scanning skill.

Weigh upThis review did not execute the skill or infisical scan.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewMIT

Security · Infisical Inc.

Infisical Secret Syncs Guide

Infisical skill for configuring Secret Syncs that push secrets from an Infisical project to third-party destinations such as AWS Secrets Manager, GCP Secret Manager, Azure Key Vault, GitHub and Vercel.

infisicalsecretsgithub

Why we selected itIt gives the exact initial sync enum values, overwrite-destination, import-prioritize-source and import-prioritize-destination, and says names such as import-prioritize-infisical do not exist.

Weigh upInfisical is the source of truth: destination secrets not in Infisical may be overwritten depending on initial sync behaviour, and auto-sync is on by default.

Source reviewed

7 sources · 2026-10-21

Read the full review Install command in reviewMIT

Infrastructure · Infisical Inc.

Infisical Self-Hosted Deployment

Infisical skill for deploying and running a self-hosted Infisical instance with Docker, Docker Compose or the Kubernetes Helm chart, including its PostgreSQL and Redis dependencies.

infisicalself-hostdocker

Why we selected itIt states that ENCRYPTION_KEY cannot be recovered if lost and gives its format: openssl rand -hex 16 for standard deployments, a 256-bit base64 key from openssl rand -base64 32 with FIPS.

Weigh upPostgreSQL is the only supported database; the skill says to use 14+ and that Infisical is extensively tested on 16.

Source reviewed

6 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Identity and access · Infisical Inc.

Infisical SSO and User Provisioning Guide

Infisical skill for connecting an identity provider for human login and SCIM user and group provisioning, including SSO enforcement and the /login/admin break-glass portal.

infisicalssoidentity

Why we selected itIt gives the prerequisite chain in order: email domain verification, then SAML or OIDC, then SCIM provisioning, then group mappings to roles.

Weigh upOnly Google and GitHub SSO are free; SAML 2.0 and OIDC need the Cloud Pro tier, SCIM the Cloud Enterprise tier, and references/sso-setup.md lists LDAP login as Paid (Enterprise).

Source reviewed

7 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Infrastructure development · Infisical Inc.

Infisical Terraform Provider

Infisical skill for the Infisical Terraform provider: provider authentication, ephemeral secret reads, data sources, project roles, access approval policies and Terraform Cloud OIDC.

infisicalterraformhcl

Why we selected itIt says provider credentials go inside a nested auth attribute, auth = { universal = { client_id, client_secret } } or auth = { oidc = { identity_id, token_environment_variable_name } }, and that client_id, client_secret or identity_id set directly on the provider block fail at plan time.

Weigh upThe infisical_secrets data source stores secret values in Terraform state, per references/resources-and-data-sources.md; only the ephemeral resource keeps them out, and it requires Terraform 1.10 or later.

Source reviewed

7 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Security · Infisical Inc.

Infisical User Setup Guide

Infisical skill for getting secrets from Infisical into applications and platforms: the CLI, Docker, CI/CD, Kubernetes, the language SDKs and machine identity authentication.

infisicalsecretscli

Why we selected itIts references/agent-bootstrap.md tells an agent running the CLI how to log in, create and link a project, and import .env files without the interactive prompts that hang a subprocess.

Weigh upFollowing agent-bootstrap.md sends .env values to the user's Infisical project with infisical secrets set; the reference requires an explicit yes from the user first and tells the agent not to open, read or print the files.

Source reviewed

7 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

API design · JetBrains

Integration Resilience Engineer

Designs resilient HTTP, messaging and scheduled integrations: timeout budgets derived from the caller’s SLO, retryable-versus-terminal failure classification, idempotency and deduplication, circuit breakers, DLQ behaviour, and distributed locking for scheduled jobs.

retriesidempotencycircuit-breakers

Why we selected itDerives timeout budgets backward from the caller’s SLO and classifies every operation (read-only versus state-changing, idempotent versus not) before any retry policy is written.

Weigh upThe worked context is Kotlin plus Spring; the design sequence and rules transfer, the framework-specific examples do not.

Source reviewed

6 sources · 2026-10-13

Read the full review Install command in reviewMIT (upstream Kotlin backend agent skills)

Documentation and knowledge · Anthropic

Internal Communications

Formats source-backed team updates, internal newsletters and employee FAQs for a defined audience.

communicationsstatus-updatesfaq

Why we selected itThe main file is a router to four example formats rather than one generic memo voice.

Weigh upThe skill says to follow the matching example file for formatting and content gathering; the root file does not contain those rules.

Source reviewed

5 sources · 2026-10-14

Read the full review Install command in reviewApache-2.0

Kusto graph visualization · Microsoft

IRQL graph visualization

Pins Microsoft's IRQL graph skill, which asks before deploying a missing stored graph function and will not invent joins or columns.

kustokqlirql

Why we selected itDeploying a missing required graph function is a database alter, and the file says to ask for confirmation before that deploy.

Weigh upThis review did not execute the resource.

Source reviewed

6 sources · 2026-10-15

Read the full review Upstream setup linkedMIT

Development workflow · getsentry

Iterate on Pull Requests

Resolve actionable CI failures and review feedback while reporting human merge gates separately.

development-workflowgetsentry

Why we selected itThe loop fixes failing checks and high or medium review comments, then commits and pushes. It does not wait for human approval or a merge gate.

Weigh upfetch_pr_checks.py only reads gh. reply_to_thread.py posts review replies through GitHub's GraphQL API. Both need a logged-in gh.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Java migration · GitHub community (awesome-copilot)

Javax to Jakarta Namespace Migration

Distinguish migrated Jakarta EE packages from javax packages that remain in the JDK.

java-migrationgithub

Why we selected itThe definition explicitly exempts javax.sql, naming, crypto, net and other JDK APIs while mapping EE packages and checking residual imports after the migration.

Weigh upThe listed dependency and descriptor versions are examples, not a compatibility matrix for every Tomcat/Jakarta target. Blind global replacement would corrupt JDK imports and miss reflection strings or descriptors.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Testing and reliability · Grafana Labs

k6

Writes a k6 script from a matching example, then validates it by running it rather than by reading it.

k6load-testingperformance

Why we selected itThe workflow is read example, adapt, save, validate, review, and it says to skip a todo list that would only mirror those steps.

Weigh upThe validation table uses k6 run for browser scripts first; otherwise ramping-vus and arrival-rate scenarios use k6 inspect to avoid executing the entire load profile. Other rows run the script. This review did not execute any validation.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

k6 documentation · Grafana Labs

k6 documentation

Pins Grafana's k6 documentation skill, which requires a confirmation before any git push and a clean k6 master run before commit.

k6documentationgit

Why we selected itNo selected skills record uses this source path.

Weigh upThe resource was not executed.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Testing and reliability · Grafana Labs

k6 test maintenance

Treats a k6 edit as behaviour-aware change control, with a validation loop on every edit.

k6load-testingmaintenance

Why we selected itThe core-principle heading is behaviour-aware change control.

Weigh upMaintaining a script does not create a load measurement.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

k6 trend reporting · Grafana

k6 trend analysis

Pins Grafana's k6 trend skill, which reports drift and headroom and refuses to apply script or threshold edits itself.

grafanak6trends

Why we selected itNo selected skills record uses this source path.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Website load testing · Grafana

k6 website performance test

Pins Grafana's website k6 skill, which stops without a named workflow and will not design thresholds until functional tests exit cleanly.

grafanak6load-test

Why we selected itNo selected skills record uses this source path.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Kusto graph queries · Microsoft

Kusto Graph Semantics

Pins Microsoft's Kusto graph skill, which requires the exact command and an explicit confirmation before a graph model or snapshot is created, altered, or dropped.

kustokqlgraph

Why we selected itControl commands that create, alter, or drop graph models and snapshots must be shown with cluster, database, and object, then explicitly confirmed.

Weigh upThis review did not execute the resource.

Source reviewed

6 sources · 2026-10-15

Read the full review Upstream setup linkedMIT

Model evaluation · LangChain

LangSmith Evaluator

Builds LangSmith offline and online evaluators, with rules for name-based deletes and judge model configuration.

langsmithllm-as-judgeevaluation

Why we selected itThe file warns that evaluator delete is name-based and removes every workspace rule with that display name, and says to run get first and never pass --yes unless the user asks.

Weigh upSetup installs the LangSmith CLI by piping an unpinned script from cli.langsmith.com to sh. That script was not inspected.

Source reviewed

3 sources · 2026-10-13

Read the full review Install command in reviewMIT

Delivery automation · addyosmani

Launch and Rollback Planning

Prepare a staged release with monitoring, ownership and rollback criteria.

delivery-automationaddyosmani

Why we selected itThe checklist includes post-launch verification, expiring feature flags and an error-budget gate instead of treating successful deployment as the outcome.

Weigh upThe fixed percentage thresholds and timing examples are templates, not statistically justified universal gates. Automated accessibility/security scans do not certify all behavior.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Testing · trailofbits

LibAFL Fuzzer Architecture

Design a custom fuzzer from observers, feedback, objectives, schedulers and executors.

testingtrailofbits

Why we selected itLibAFL is presented as a Rust library for a custom fuzzer when libFuzzer or AFL++ do not fit the target.

Weigh upCloning and the rustup installer in the install section are network steps. This review did not run them, and the skill does not pin the LibAFL revision it clones.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Observability · Elastic

LLM and Agentic Observability

Answers LLM latency, token, cost and call-chain questions from telemetry already in Elastic, and says so when cost is not instrumented.

llm-observabilityesqltoken-cost

Why we selected itIf no cost field exists, the file says to report that cost is not instrumented and never multiply token counts by a guessed per-token price.

Weigh upThe time-series path needs Elasticsearch Stack 9.4 or Serverless; below that the file falls back to FROM with BUCKET, with a base floor of 8.11.

Source reviewed

5 sources · 2026-10-13

Read the full review Install command in reviewApache-2.0

Development workflow · GitHub community (awesome-copilot)

Local Web Application Testing

Exercise a running web app with Playwright and retain browser evidence for failed flows.

development-workflowgithub

Why we selected itIt drives a local web app with Playwright, preferring the Playwright MCP server and falling back to a local Node install.

Weigh upNative mobile apps are out of scope.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Loki label governance · Grafana Labs

Loki label strategy evaluator

Pins Grafana's Loki label skill, which refuses to drop three correlation labels and will not ship an audit without its disclaimer.

grafanalokilabels

Why we selected itNo selected skills record uses this source path.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Device provisioning · Anthropic

M5Stack onboarding

Pins the M5 onboard skill, which will not guess a serial port or continue a native-USB flash until the user confirms download mode.

m5stackesp32cardputer

Why we selected itIf more than one board is connected, the skill says to ask which port to target and not to guess, and a bare Cardputer must be asked apart from Cardputer-Adv because the wrong image boot-loops the device.

Weigh upThis review did not execute the resource.

Source reviewed

7 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Documentation and knowledge · Sentry

Maintaining AGENTS.md

Keeps agent instruction files short and points at existing docs instead of duplicating them.

agents-mdinstructionsrepositories

Why we selected itThe length cap and the symlink rule are explicit, so the file cannot grow into a second README.

Weigh upThe file must stay at or under 100 lines, with a target under 60.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Documentation · GitHub community (awesome-copilot)

Markdown to HTML Workflows

Choose a Markdown renderer and preserve document structure during HTML conversion.

documentationgithub

Why we selected itThe definition separates conversion, syntax-highlighting, math extensions and static-site tooling; it calls out sanitization and template execution settings as separate concerns.

Weigh upExamples span different tools and versions; stated full-compatibility percentages are not measured here. Markdown rendering does not sanitize untrusted HTML or guarantee safe publication.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Research and documentation · Anthropic

math-proof: siege

Runs an hours-long, multi-round attack on one hard mathematics problem with judge and worker sub-agents and a bookkeeping script, ending in a proof.md whose Status section says what is and is not proved.

proofsreasoningsubagents

Why we selected itThe orchestrating session does no mathematics: each step goes to a fresh math-proof-judge subagent and each query to a fresh worker that sees only that query and the problem, and the judge keeps a ledger of claims marked PROVED, REFUTED or OPEN.

Weigh upLast commit touching plugins/math-proof/skills/siege at revision 41f32e40: 2026-09-30 (aa5654b7).

Source reviewed

7 sources · 2026-10-22

Read the full review Setup example in reviewApache-2.0

Research and documentation · Anthropic

math-proof: solo

Works one hard mathematics problem in the current Claude Code session, writing each settled step to a notes file as it goes and ending with a self-contained proof.md.

proofsreasoningclaude-code

Why we selected itIts tool boundary is written down: the frontmatter disallows WebSearch, WebFetch and AskUserQuestion and pre-approves Read, Write, Edit, Glob, Grep and Bash only for mkdir, cp and cmp, and the body says to use the shell for nothing else.

Weigh upLast commit touching plugins/math-proof/skills/solo at revision 41f32e40: 2026-09-30 (aa5654b7).

Source reviewed

3 sources · 2026-10-21

Read the full review Setup example in reviewApache-2.0

Web development · Matt Pocock

Matt Pocock Codebase Design

Evaluates module boundaries through caller complexity, dependency seams and alternative interface designs.

architectureinterfacesrefactoring

Why we selected itDefines a deep module by what callers can accomplish through a small interface, rather than by its line count.

Weigh upThe adapter and interface rules are design heuristics, not universal requirements; local substitutes do not prove production integration behavior.

Source reviewed

16 sources · 2026-10-15

Read the full review Install command in reviewMIT

Research and documentation · Matt Pocock

Matt Pocock Domain Modeling

Builds a project glossary (GLOSSARY.md) and short architecture decision records while testing domain terms against code and concrete edge cases.

domain-modelingglossaryarchitecture-decisions

Why we selected itDistinguishes canonical domain terms from implementation notes and uses conflicting examples to expose overloaded language.

Weigh upThe root and bounded-context instructions do not fully agree on glossary and ADR placement; agree the target paths and numbering convention first.

Source reviewed

16 sources · 2026-10-15

Read the full review Install command in reviewMIT

Research and documentation · Matt Pocock

Matt Pocock Grilling

Interviews the user through dependency-ordered decisions before the agent acts on a plan or idea.

planningrequirementsdecision-making

Why we selected itOnly asks decisions whose prerequisites are settled, so answers do not depend on unanswered questions from the same round.

Weigh upThe interview has no built-in round cap and can be too demanding for a small task; choose its scope deliberately.

Source reviewed

15 sources · 2026-10-15

Read the full review Install command in reviewMIT

Agent development · Anthropic

MCP Integration for Claude Code Plugins

Explains how a Claude Code plugin bundles MCP servers, through .mcp.json or plugin.json, for stdio, SSE, HTTP and WebSocket servers, and how plugin commands pre-allow the resulting MCP tools.

claude-codeplugin-devmcp

Why we selected itIt covers both ways a plugin can bundle MCP servers, a dedicated .mcp.json at the plugin root or an mcpServers field in plugin.json, with configuration examples for stdio, SSE, HTTP and WebSocket servers.

Weigh upLast commit touching plugins/plugin-dev/skills/mcp-integration at revision 41f32e40: 2026-08-06 (088bee40).

Source reviewed

6 sources · 2026-10-21

Read the full review Upstream setup linkedApache-2.0

Data engineering · dbt-labs

Mermaid dbt Lineage

Render model lineage with explicit local, production or artifact provenance.

data-engineeringdbt-labs

Why we selected itIt draws a Mermaid lineage diagram from dbt MCP, manifest.json, or parsed code, in that order.

Weigh upThe skill asks for the model name instead of guessing it.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Security review · Trail of Bits

Mermaid to ProVerif

Reads a Mermaid sequence diagram of a cryptographic protocol and writes a ProVerif .pv model, with queries for secrecy, authentication and forward secrecy, that can be passed to the ProVerif verifier.

cryptographyformal-methodsprotocol

Why we selected itTranslates a Mermaid sequenceDiagram, typically the output of crypto-protocol-diagram, into a ProVerif .pv model with queries for secrecy, authentication and forward secrecy.

Weigh upIts stated tools are Read, Write, Grep and Glob, with no Bash: it writes the .pv model and the user runs ProVerif on it.

Source reviewed

5 sources · 2026-10-22

Read the full review Upstream setup linkedCC-BY-SA-4.0

Billing · Stripe

Metronome

Stripe's Metronome skill for usage ingestion, contracts, and invoice sync, with confirmation required before writes because sandbox and production share one API host.

metronomestripeusage-billing

Why we selected itThe catalogue has stripe-pay and withholds two other Stripe skills; it has no Metronome skill.

Weigh upThis review did not execute the skill or call api.metronome.com.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Agent Frameworks · Microsoft

Microsoft Agent Framework Azure AI

Shows how to build a persistent Azure AI Foundry agent with the Python agent-framework SDK, DefaultAzureCredential, and hosted tools.

microsoftazure-aiagent-framework

Why we selected itAuthentication is pinned to DefaultAzureCredential rather than connection strings.

Weigh upThe skill says to avoid connection strings and account keys.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewMIT

Cloud operations · Microsoft

Microsoft Azure Cost Management

Connects Azure cost reports to actual billing queries, utilization evidence and separately identified savings estimates.

azurecost-managementfinops

Why we selected itRequires the total bill and service breakdown alongside optimization suggestions.

Weigh upRequires Azure billing/monitoring permissions and supporting CLI or MCP tools.

Source reviewed

9 sources · 2026-10-15

Read the full review Install command in reviewMIT

Cloud operations · Microsoft

Microsoft Azure Diagnostics

Routes Azure incidents through resource health, recent changes, logs and service-specific troubleshooting guides.

azureakslogs

Why we selected itSeparates Container Apps, App Service, Functions, AKS, compute and messaging investigations, each routed to its own reference.

Weigh upThe AKS helper selects Azure resources by arguments but uses the existing kubectl context; confirm both point to the same cluster.

Source reviewed

9 sources · 2026-10-15

Read the full review Install command in reviewMIT

Research and documentation · Microsoft

Microsoft Docs Research

Finds current Microsoft documentation with task-specific queries and retrieves full pages when excerpts are insufficient.

microsoftazuredotnet

Why we selected itAsks for version, platform and task context in documentation searches.

Weigh upThis is a retrieval workflow, not an implementation validator; official documentation still needs to match the installed version and target environment.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Identity and access · Microsoft

Microsoft Entra Agent ID

Pins Microsoft's Entra Agent ID skill, which rejects Azure CLI tokens for these APIs and keeps the AgentID sidecar on localhost.

entraagent-identityoauth

Why we selected itThe catalogue has entra-app-registration from this plugin and not this Agent Identity blueprint skill.

Weigh upThis review did not execute the resource.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Identity and access · Microsoft

Microsoft Entra app registration

Pins Microsoft's Entra app-registration skill, which sends Key Vault secret work elsewhere and says not to hardcode client secrets.

microsoftentraoauth

Why we selected itThe catalogue has Microsoft identity and Key Vault Python SDK skills and not this Entra app-registration procedure.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Microsoft's .NET guide to Azure Functions handlers for Entra ID custom authentication extensions: adding claims at token issuance, prefilling or validating sign-up attributes, and sending one-time passcodes through your own channel.

entra-idclaimsauthentication

Why we selected itIt gives a handler for each of the four events it lists (OnTokenIssuanceStart, OnAttributeCollectionStart, OnAttributeCollectionSubmit and OnOtpSend), each returning the matching WebJobs response type.

Weigh upBoth install lines Microsoft prints for this skill, `npx skills add microsoft/skills --skill azure-sdk-dotnet` and `/plugin install azure-sdk-dotnet@skills`, install the whole azure-sdk-dotnet plugin rather than the authentication events skill alone.

Source reviewed

6 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Research and documentation · Microsoft

Microsoft Wiki Researcher

Traces architecture and data flow through source code to produce an explanation with file references, diagrams and stated uncertainties.

codebase-researcharchitecturedocumentation

Why we selected itRequires code-reading before architectural claims and traces actual calls and data flow rather than trusting directory names.

Weigh upIts git command identifies the current branch, although the prose calls it the default branch; choose the actual revision and host-specific link format yourself.

Source reviewed

7 sources · 2026-10-15

Read the full review Install command in reviewMIT

Development workflow · Trail of Bits

Modern Python

Sets up uv, ruff and ty, and says when not to migrate an existing Python toolchain.

pythonuvruff

Why we selected itThe definition has an explicit When NOT to Use section, so a migration is not the default.

Weigh upCC-BY-SA-4.0 requires share-alike and attribution if you redistribute an adaptation.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Web development · JetBrains

Modern Web Guidance

Routes HTML, CSS and client-side JavaScript work through a searchable corpus of Chrome-authored best-practice guides — container queries, dialogs and popovers, view transitions, scroll-driven animation, forms and autofill, image priority — with Baseline browser-support policy rules for fallbacks.

csscontainer-queriesforms

Why we selected itTurns modern-CSS and forms guidance into a queryable corpus instead of a fixed checklist: the agent searches an action-oriented phrase, gets scored matches with token counts, then retrieves only the guide it needs.

Weigh upThe search workflow shells out to the modern-web-guidance npm package at runtime, so it needs network access and Node; the --skill-version flag only warns when the SKILL.md is stale, it does not pin the package version.

Source reviewed

6 sources · 2026-10-07

Read the full review Install command in reviewApache-2.0

Data pipelines · mongodb

MongoDB Atlas Stream Processing

Plan and operate Atlas streaming workspaces, connections and processors through MongoDB MCP.

data-pipelinesmongodb

Why we selected itThe tool matrix separates discovery, creation, lifecycle changes and teardown; processor modification requires a stopped/created state and PrivateLink is project-scoped.

Weigh upThis requires Atlas API credentials and matching streams tools; workspace teardown cascades. Source retention and billing descriptions must be rechecked for the chosen service version.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Database engineering · mongodb

MongoDB Connection Configuration

Review MongoDB client reuse, pool capacity and timeouts against workload and topology.

database-engineeringmongodb

Why we selected itThe definition requires deployment and workload context before setting pool values and separates monitoring connections from per-server application pools.

Weigh upConnection-memory and setup-time examples are estimates; pool formulas do not model every driver/topology. Timeouts and retries need operation-specific semantics.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Agent integration · mongodb

MongoDB MCP Configuration

Configure MongoDB MCP access for a direct cluster, Atlas API or local Docker environment.

agent-integrationmongodb

Why we selected itThe definition separates data-plane connection strings from Atlas administration credentials and explicitly tells the agent not to collect secret values.

Weigh upPresence of MDB_MCP_READ_ONLY alone does not prove true or enforced read-only mode. Client detection via environment variables is a heuristic, and broad Atlas owner roles may exceed the intended task.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Database engineering · mongodb

MongoDB Natural Language Queries

Translate questions into MongoDB read queries using inspected schema, indexes and bounded samples.

database-engineeringmongodb

Why we selected itThe workflow prefers find for simple retrieval, reserves aggregation for transformations/grouping, and explicitly excludes write pipelines and Atlas Search.

Weigh upSamples may contain sensitive data. The suggestion to replace field existence with not-equal-null changes semantics for fields that exist with null values; validate missing/null cases explicitly.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Databases · MongoDB

MongoDB query optimizer

Official MongoDB skill that limits itself to slow-query and index work and names the MCP tools it expects for explain, indexes, and Atlas Performance Advisor.

mongodbindexesexplain

Why we selected itThe catalogue's MongoDB performance advisor is a GitHub Copilot agent, not this official skill.

Weigh upThis review did not execute the skill or call MongoDB MCP tools.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Databases · MongoDB

MongoDB schema design

Pins MongoDB's schema-design skill, which separates read checks from writes and says write tools need an explicit yes.

mongodbschemamodeling

Why we selected itThe catalogue has MongoDB's query-optimizer skill and not this schema-design skill.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Databases · MongoDB

MongoDB Search and AI

Chooses between Atlas Search, Vector Search, Automated Embedding and hybrid search, and creates indexes only after explicit approval.

mongodbvector-searchhybrid-search

Why we selected itIndex creation follows a fixed sequence: explain the configuration, show the JSON, ask for a name and get explicit approval before calling create-index.

Weigh upThe file says never to recommend $regex or $text for search use cases, which steers toward Atlas Search features whose prerequisites sit in the reference files.

Source reviewed

4 sources · 2026-10-13

Read the full review Install command in reviewApache-2.0

Telemetry export · Microsoft

Monitor OpenTelemetry exporter

Pins Microsoft's Application Insights exporter skill, which uses Entra credentials for ingestion and requires an explicit telemetry shutdown.

application-insightsazureentra

Why we selected itThe file says the connection string only identifies the Application Insights resource and that DefaultAzureCredential provides Entra authentication, and it says to use Entra instead of instrumentation keys.

Weigh upThe resource was not executed.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Plans a staged warehouse migration with SQL translation, source-to-target reconciliation, a first cutover slice and a rollback path.

migrationduckdbdata-validation

Why we selected itSeparates planning with supplied schemas from validation against a connected database.

Weigh upThe Python example imports repository-level scripts/_lib helpers; installing only the skill directory does not provide its full executable dependency tree.

Source reviewed

9 sources · 2026-10-15

Read the full review Install command in reviewMIT

Reviews where credentials live, which boundary isolates tenants, and how roles, share grants and published tables combine.

database-accesstenant-isolationservice-accounts

Why we selected itSeparates platform roles, share recipients and table/view exposure instead of treating all permissions as one control.

Weigh upThis is an architectural review guide, not a compliance certification or a live permission audit.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Development workflow · GitHub community (awesome-copilot)

Multi-file Refactor Planning

Plan a sequenced refactor with repository-specific dependencies, checks and recovery steps.

development-workflowgithub

Why we selected itThe workflow orders contracts/types before implementations and callers and requires rollback or recovery for the riskiest phases.

Weigh upThe definition normally stops for plan review; existing explicit authorization to continue can override that checkpoint. A plan does not establish behavior preservation.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Containers · GitHub community (awesome-copilot)

Multi-stage Dockerfile Design

Separate build dependencies from the runtime image and constrain the build context.

containersgithub

Why we selected itThe source separates dependency/build/test/runtime concerns and recommends copying only necessary outputs, ordering cache layers and removing build tools from the final image.

Weigh upVersion tags are mutable and do not guarantee reproducibility; pin digests where required. Multi-stage structure alone does not keep secrets out of build cache or logs, and Alpine compatibility must be checked.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

AI application development · Neon

Neon AI Gateway

Adds multi-provider LLM calling through Neon's branch-scoped gateway with one credential, resolving model IDs against the live catalog instead of hardcoded lists.

llm-gatewaymodel-routingopenai-compatible

Why we selected itIntegrates with OpenAI, Anthropic, google-genai, Vercel AI SDK, and Mastra by changing only the base URL, with separate recommended and lower-level paths.

Weigh upGateway availability and the served catalog are plan-gated; the skill's own Availability section says to check gating first.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Cloud development · Cloudflare

Next.js on Cloudflare

Builds, migrates and deploys Next.js apps on Cloudflare Workers with vinext, holding the vinext-over-OpenNext default for new projects while preserving existing OpenNext setups and delegating setup to vinext's upstream skills.

nextjscloudflare-workersvinext

Why we selected itTakes a position with a preservation rule: vinext is the default for new Next.js projects on Workers, existing OpenNext setups are preserved during unrelated maintenance, and an explicit user choice always wins.

Weigh upA routing skill: the concrete setup and migration steps live in cloudflare/vinext's own skills, which must be installed separately for the full workflow.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Package manager auditing · Grafana

npm supply-chain audit

Pins Grafana's package-manager audit, which is read-only, stops when package.json is missing, and stops again when every check passes.

grafananpmpnpm

Why we selected itNo selected skills record uses this source path.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Observability · Addy Osmani

Observability and instrumentation

Requires a definition of working before adding a log line, and a check that the telemetry itself is real.

loggingmetricstracing

Why we selected itTelemetry starts from the questions an on-call engineer will ask, not from adding logs.

Weigh upMetrics labels must come from small fixed sets; user IDs and other unbounded values are forbidden.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Cloud operations · JetBrains

Observability Integrator

Designs instrumentation that answers operational questions instead of filling dashboards: metrics tied to user journeys with a cardinality budget, structured logs with stable correlation identifiers and PII redaction by policy, trace propagation across async and coroutine boundaries, deliberate sampling, and liveness/readiness/startup health semantics.

observabilitymetricsstructured-logging

Why we selected itStarts from the operational questions the instrumentation must answer — is the service healthy, which journey is slow, which dependency is failing, where is saturation growing — and explicitly rejects vanity dashboards.

Weigh upThe worked context is Kotlin plus Spring (actuator exposure, coroutine propagation); the rules transfer, the framework specifics do not.

Source reviewed

6 sources · 2026-10-13

Read the full review Install command in reviewMIT (upstream Kotlin backend agent skills)

Release preparation · trailofbits

Open-source Release Preparation

Prepare repository content, build instructions and packaging for a deliberate public release.

release-preparationtrailofbits

Why we selected itcheck_readiness.sh flags tracked filenames and presence indicators without scanning their contents; detect_org.sh checks remotes and recent author/committer email domains.

Weigh upThe readiness script checks file/name presence rather than contents or security. Organization detection from remotes/commit emails is heuristic; licence selection and fresh-repository recommendations need the actual ownership and disclosure context.

Source reviewed

6 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

AI application development · OpenAI

OpenAI Agents SDK Workflow

Builds a small Agents SDK application around an explicit workflow contract, then checks the real agent path and optional local deployment.

openaiagents-sdktools

Why we selected itStarts with a single agent and adds tools, delegation or sandboxing only when the application needs them.

Weigh upA companion credential gate can pause work before implementation even when a key already exists.

Source reviewed

6 sources · 2026-10-15

Read the full review Upstream setup linkedProprietary (declared in plugin manifest)

AI application development · OpenAI

OpenAI API Troubleshooting

Classifies a failed OpenAI request using its actual transport or API error and routes credentials, billing, throttling and access issues separately.

openaiapierrors

Why we selected itDistinguishes exhausted quota from ordinary rate limiting even when the HTTP status is the same.

Weigh upIt relies on companion documentation and credential skills; it does not provision or rotate keys itself.

Source reviewed

6 sources · 2026-10-15

Read the full review Upstream setup linkedProprietary (declared in plugin manifest)

Creates or reuses a repository threat model bound to the actual target and revision, preserving supplied context and explicit review gaps.

threat-modelingcodex-securitytrust-boundaries

Why we selected itBinds reusable models to repository/version identity and avoids mixing scan-specific context into a shared cache.

Weigh upThis is one phase in the Codex Security plugin, not a standalone full security scan.

Source reviewed

8 sources · 2026-10-15

Read the full review Upstream setup linkedProprietary (declared in plugin manifest)

Data and research · OpenAI

OpenAI Data Analytics Notebooks

Turns SQL or Python analysis into a traceable notebook whose conclusions are tied to executed outputs and stated assumptions.

jupytersqlpython

Why we selected itRecords query links, input versions, filters and cohort definitions next to the analysis.

Weigh upWarehouse and operational-log placeholders depend on the host’s configured connectors; exports or supplied data are fallback paths.

Source reviewed

5 sources · 2026-10-15

Read the full review Upstream setup linkedProprietary (declared in plugin manifest)

Agent development · OpenAI

OpenAI Evaluate Skill

Connects a local skill review to structural checks, budget explanations and an explicitly separate live benchmark workflow.

skillsevaluationtoken-usage

Why we selected itDistinguishes static instruction-size estimates from observed usage collected by real agent runs.

Weigh upStatic scores and budget bands are tool heuristics, not proof of task quality or safety.

Source reviewed

8 sources · 2026-10-15

Read the full review Upstream setup linkedMIT (declared in plugin manifest)

Observability · grafana

OpenTelemetry with Grafana

Connect application telemetry through direct OTLP export or an Alloy/Collector pipeline.

observabilitygrafana

Why we selected itIt wires OpenTelemetry export to Grafana Cloud or a local Alloy collector and checks auth with an empty OTLP POST.

Weigh upThe empty-body probe is not a valid trace; it only distinguishes auth from a bad path.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Testing · trailofbits

OSS-Fuzz Integration

Prepare an open-source project for continuous fuzzing and reproduce its harnesses locally.

testingtrailofbits

Why we selected itThe skill separates a local helper.py run from public enrollment, which it describes as selective.

Weigh upThe documented local path clones OSS-Fuzz and runs Docker via helper.py. This file does not contain those images and does not submit the enrollment pull request.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Infrastructure · hashicorp

Packer Azure Image Builder

Prepare Azure managed or Compute Gallery images with Packer.

infrastructurehashicorp

Why we selected itThe source distinguishes managed images from gallery destinations and warns that gallery image versions cannot be overwritten.

Weigh upThe Managed Identity example actually enables Azure CLI authentication; it is not evidence of managed-identity configuration. Example subscription-wide Contributor scope is broader than a dedicated build resource group.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMPL-2.0

Image building · HashiCorp

Packer Windows builder

Pins HashiCorp's Packer Windows builder, which provisions over WinRM, deletes guest temp files during cleanup, and says a failed build may leave resources running.

packerwindowswinrm

Why we selected itThe file says a failed build may leave resources running and that cleanup has to be verified. Its cleanup provisioner deletes guest temp files and the Windows Update cache, and it stops and starts wuauserv.

Weigh upThe in-guest cleanup provisioner does not remove cloud resources. The file's own note says a failed build may leave resources running and that this must be checked separately.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMPL-2.0

Web performance · Addy Osmani

Performance optimization

Starts from a measurement workflow and names Core Web Vitals targets instead of guessing a speedup.

performanceweb-vitalsqueries

Why we selected itIts five steps (measure, identify, fix, verify, guard) keep a change only when the re-measurement beats run-to-run variance with tests still green.

Weigh upIts See Also section points to ../../references/performance-checklist.md. The README says a per-skill npx install does not copy that repository-level directory.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Databases · Pinecone

Pinecone quickstart

Pinecone's onboarding skill: either a semantic-search demo on a new integrated index through the Pinecone MCP and a bundled upsert script, or a Pinecone Assistant built from the user's own documents.

vector-dbembeddingsreranking

Why we selected itPath A is a fixed, checkable loop: list-indexes to test the key, create-index-for-model for a serverless index quickstart-skills (aws, us-east-1, llama-text-embed-v2, text mapped to chunk_text), a bundled upsert of nine records with pinecone==9.1.0, search-records queries that share no keywords with the records, and optional bge-reranker-v2-m3 reranking.

Weigh upThe skill directory skills/pinecone-quickstart last changed in a commit dated 2026-08-10 (git log at the pin); the pinned revision 885dc539 is dated 2026-09-18.

Source reviewed

9 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Framework development · JetBrains

Pinia

Reference-index over Pinia store design: defining stores, state, getters and actions, storeToRefs destructuring, plugins, composing stores without circular dependencies, SSR hydration, HMR and testing with @pinia/testing.

piniavuestate-management

Why we selected itSplits the parts of Pinia that agents get wrong into nine reference files — storeToRefs reactivity, composing stores without circular dependencies, using stores outside components, SSR hydration — rather than compressing them into prose.

Weigh upThe mirror carries the January 2026 generation, based on Pinia v3.0.4; the upstream antfu/skills generator has since produced a Pinia v4.0.3 revision (2026-09-25) that the pinned mirror does not include.

Source reviewed

6 sources · 2026-10-07

Read the full review Install command in reviewMIT (upstream antfu/skills)

Database review · PlanetScale

PlanetScale Postgres safety review

Reviews a PlanetScale Postgres database and recommends roles, pg_strict, backups, and Traffic Control without applying them.

postgresplanetscalereview

Why we selected itThe skill reviews PlanetScale Postgres for roles, pg_strict, backups, traffic control, and network posture, and it says not to apply changes.

Weigh upThe file does not apply schema, role, backup, or network changes.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewMIT

Database review · PlanetScale

PlanetScale Vitess safety review

Review a PlanetScale Vitess database and recommend safety posture. The skill's own first instruction is not to apply changes.

planetscalevitessschema

Why we selected itThe file is a review that recommends posture and says not to apply changes.

Weigh upThe purpose section says not to apply any changes.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewMIT

Documents one pattern for per-project Claude Code plugin settings, a .claude/<plugin-name>.local.md file with YAML frontmatter and a Markdown body, and how hooks, commands and agents read it.

claude-codeplugin-devconfiguration

Why we selected itIt documents a single, concrete convention: settings and state in .claude/<plugin-name>.local.md in the project root, YAML frontmatter plus a Markdown body, kept out of git through a .claude/*.local.md line in .gitignore.

Weigh upLast commit touching plugins/plugin-dev/skills/plugin-settings at revision 41f32e40: 2026-02-05 (2438937b).

Source reviewed

7 sources · 2026-10-21

Read the full review Upstream setup linkedApache-2.0

Agent development · Anthropic

Plugin Structure for Claude Code

Describes the Claude Code plugin layout: the .claude-plugin/plugin.json manifest, root-level commands, agents, skills and hooks directories, .mcp.json, auto-discovery and ${CLAUDE_PLUGIN_ROOT} paths.

claude-codeplugin-devmanifest

Why we selected itIt states the layout rules a plugin loader relies on: the manifest lives at .claude-plugin/plugin.json, only its name field is required, and the commands, agents, skills and hooks directories sit at the plugin root, not inside .claude-plugin/.

Weigh upLast commit touching plugins/plugin-dev/skills/plugin-structure at revision 41f32e40: 2026-02-05 (2438937b).

Source reviewed

7 sources · 2026-10-21

Read the full review Upstream setup linkedApache-2.0

Dependency reduction · Grafana

pnpm dependency reduction

Pins Grafana's pnpm reduction skill, which runs a read-only check first, stops off pnpm, and requires approval before adding a dependency.

grafanapnpmdependencies

Why we selected itNo selected skills record uses this source path.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Model development · Hugging Face

Portable Triton Kernels

Writes and benchmarks portable Triton kernels for NVIDIA and AMD GPUs: block-size discipline, autotune configuration, fp16/bf16/fp8 numerics, masked loads, reductions, correctness tests against a PyTorch reference and publishing to the Kernels Hub.

tritongpukernels

Why we selected itIts hard constraints encode the silent-failure traps: never autotune BLOCK_SIZE when it controls the reduction axis because partial rows give wrong results silently, and masked loads need a safe other value.

Weigh upTriton-specific: WebGPU/WGSL and raw CUDA C++ are out of scope, and the portability claim is NVIDIA plus AMD only.

Source reviewed

5 sources · 2026-10-07

Read the full review Install command in reviewApache-2.0

Security review · Trail of Bits

Post-patch validation

Requires a baseline-versus-patched evidence contract before a security fix is called done.

patchesverificationsecurity

Why we selected itThe description says to use it after a patch exists and before accepting it as fixed.

Weigh upCC-BY-SA-4.0 applies to adaptations you redistribute.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Databases · Timescale

Postgres

Timescale's postgres skill from pg-aiguide: a dispatcher that sends the agent to nine reference guides for schema exploration, table design, pgvector and hybrid search, TimescaleDB hypertables and schema migrations.

postgresqlsqlschema

Why we selected itTimescale publishes it in timescale/pg-aiguide (frontmatter license Apache-2.0, author tigerdata; NOTICE reads "Copyright 2025 Timescale, Inc., d/b/a Tiger Data"), and the README prints a single-skill install line for it: `npx skills add timescale/pg-aiguide --skill postgres`.

Weigh upThe skills/postgres directory last changed in a commit dated 2026-09-25; the pinned revision 187be00d is dated 2026-10-07.

Source reviewed

13 sources · 2026-10-21

Read the full review Install command in reviewApache-2.0

Pull request repair · Trail of Bits

PR improver

Pins the branch loop that refuses the base branch, stops on an empty diff, will not replace a missing reviewer, and does not commit.

pull-requestreviewworkflow

Why we selected itThe skill refuses to run when the current branch is the base and stops when the diff is empty.

Weigh upThe resource was not executed.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Databases · Prisma

Prisma Client API

Covers Prisma Client model queries, relation filtering, interactive transactions, and batching semantics.

prismaormtypescript

Why we selected itDetails explicit relation filtering semantics (some, every, none) and type-safe pagination patterns (cursor-based vs offset-based) with TypeScript type examples.

Weigh upAssumes a generated Prisma Client in node_modules/@prisma/client or a custom output directory; does not execute code generation.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Application deployment · Prisma

Prisma Compute

Pins Prisma's Compute skill, which checks auth before app mutations and separates a non-promoting build from a production deploy.

prismacomputedeploy

Why we selected itA non-empty service token is the auth source, an empty token should fail closed, and a failed workspace must not fall through to another cached session.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Database adapters · Prisma

Prisma driver-adapter implementation

A Prisma ORM 7 driver-adapter contract: one connection per transaction, cleanup hooks that do not issue a second commit, and original database errors kept intact.

prismasqltransactions

Why we selected itIt specifies the Prisma 7 SQL adapter contract, including one dedicated connection per transaction and commit hooks that must not send a second COMMIT.

Weigh upThe skill is a protocol guide; it does not ship a working adapter or run Prisma.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewMIT

Database migrations · Prisma

Prisma MongoDB upgrade

Pins Prisma's MongoDB upgrade skill, which refuses a Prisma 7 upgrade and refuses an implicit move from MongoDB to SQL.

prismamongodbupgrade

Why we selected itThe catalogue has Prisma's v7 upgrade, client API, CLI, and driver-adapter skills and not this MongoDB decision file.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Databases · Prisma

Prisma ORM 7 CLI

Pins Prisma's ORM 7 CLI skill and its warning not to use these commands for Prisma 8 or for the public-beta Platform CLI.

prismadatabasemigrations

Why we selected itThe catalogue has Prisma ORM setup and client-API skills and not this ORM 7 CLI reference.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewMIT

Databases · Prisma

Prisma ORM Setup

Guides configuring Prisma ORM across PostgreSQL, MySQL, SQLite, SQL Server, CockroachDB, MongoDB, and Prisma Postgres.

prismaormdatabase-setup

Why we selected itConfigures provider-specific datasource blocks and connection string arguments across 7 distinct database engines without conflating parameters.

Weigh upDirects new projects towards Prisma ORM 8 forward-compatibility while maintaining v6/v7 fallbacks; does not execute database scaffolding.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewMIT

Database migrations · Prisma

Prisma ORM v7 upgrade

Numbered Prisma ORM v6 to v7 migration that requires SQL driver adapters and stops the path for MongoDB.

prismaormtypescript

Why we selected itThe file is a numbered upgrade: prisma-client generator, prisma.config.ts, required SQL driver adapters, and a client import from the generated path.

Weigh upThe file states Latest stable Prisma ORM version: 7.6.0 and requires Node.js 20.19.0+ and TypeScript 5.4.0+. This review did not query the npm registry.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Database setup · Prisma

Prisma Postgres setup

Pins Prisma's Postgres setup skill, which will not delete another resource to get past a quota block and keeps the connection out of source and chat.

prismapostgresprovisioning

Why we selected itIf provisioning is blocked by permissions or quota, the file says to report the blocker and not to delete another resource or switch workspaces.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Cloud operations · JetBrains

Production Incident Responder

Runs production incident response as mitigate-first: state impact, pick the fastest reversible mitigation, preserve evidence while the symptom still exists, then diagnose — with heuristics for retry storms, cache stampedes and config-only regressions.

incident-responsehotfixrollback

Why we selected itOrders the response deliberately — reversible mitigation before root-cause work — and names the anti-patterns: reflexive restarts that destroy evidence and amplify connection storms, scaling the app tier when the database is the bottleneck.

Weigh upFramed for Kotlin plus Spring services; the response sequence is stack-generic but dashboard and tooling references assume JVM-ecosystem observability.

Source reviewed

6 sources · 2026-10-13

Read the full review Install command in reviewMIT (upstream Kotlin backend agent skills)

Observability · grafana

Profilecli Source Insights

Retrieve Pyroscope profiles and map hot functions to the corresponding source checkout.

observabilitygrafana

Why we selected itIt queries an existing Pyroscope server with profilecli and maps hot functions back to this checkout only when the module path matches.

Weigh upThe skill does not start profiling; it reads profiles that already exist.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Security review · Trail of Bits

Project Wycheproof

Loads Project Wycheproof vectors and maps valid, invalid, and acceptable results onto pass, fail, and warning, instead of treating a custom algorithm as in scope.

cryptographytest-vectorswycheproof

Why we selected itThe result field has three values: valid should succeed, invalid should fail, and acceptable may succeed but is called non-ideal.

Weigh upThe skill says to use other tools for timing measurement, fuzzing for new bugs, and not to expect vectors for experimental algorithms.

Source reviewed

2 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Observability · Grafana Labs

Prometheus label strategy

Designs labels before a cardinality incident, and forbids dropping a label that makes a series unique.

prometheuslabelscardinality

Why we selected itThe first heading is the one rule: never drop a label that makes a series unique.

Weigh upA label audit is not a Grafana Cloud bill.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Model evaluation · Promptfoo

Promptfoo evals

Official Promptfoo skill that writes, runs and refines declarative eval suites with deterministic assertions first and a CI gate.

evaluationsevalci

Why we selected itDefines a non-redteam eval workflow with deterministic assertions before model-graded rubrics, inlined source evidence, calibrated known-good and known-bad controls, and an exit-code CI gate.

Weigh upLast commit 2026-10-15, the same day as the review; the bundle ships four skills and this candidate is the eval-authoring one only.

Source reviewed

3 sources · 2026-10-15

Read the full review Upstream setup linkedMIT

Metrics · Grafana Labs

PromQL query patterns

PromQL skill that puts rate before aggregation, keeps the le label inside histogram_quantile, and guards a zero denominator.

promqlprometheusgrafana

Why we selected itThe primary file shows rate before sum, histogram_quantile with by (le, service), a comparison that drops zero denominators, and a recording-rule replacement checked against the Prometheus query API.

Weigh upThe curl examples default to http://localhost:9090 and say to add basic auth for Grafana Cloud. No query was sent. The 0.999 figure is a placeholder in the file, not a measured SLO.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

API design · GitHub community

Protobuf gRPC API Review

Reviews .proto and gRPC changes as long-lived contracts across binary wire, JSON and text formats, generated APIs and runtime behavior, with mixed-version rollout tracing and a four-rung verdict.

protobufgrpcapi-compatibility

Why we selected itCompatibility is evaluated on four independent dimensions (binary wire, named formats, generated source API, behavior and operations) instead of a single wire-compatibility verdict.

Weigh upVerdicts are only as good as the declared compatibility envelope; the skill itself requires lowering confidence when encodings, consumers or deployment order are unknown.

Source reviewed

7 sources · 2026-10-15

Read the full review Install command in reviewMIT

Infrastructure development · HashiCorp

Provider actions

Guides Terraform Plugin Framework action registration, configuration and lifecycle-trigger integration.

terraformactionsprovider

Why we selected itThe first action in a provider requires ActionData to be set in Configure next to the existing resource data.

Weigh upActions do not produce state that resource.TestCheckResourceAttr can read; the skill says to use a PostApplyFunc against the API.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewMPL-2.0

Infrastructure development · HashiCorp

Provider configuration

Builds a provider Configure method as a credential chain and names every source a failed login tried.

terraformauthenticationprovider

Why we selected itThe inspected headings include a credential provider chain and diagnostics that name every source tried.

Weigh upMPL-2.0 is file-level copyleft for a modified redistribution of this file.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMPL-2.0

Infrastructure development · HashiCorp

Provider framework migration

Starts by asking whether to migrate at all, then muxes the provider and verifies before claiming the port is done.

terraformmigrationprovider

Why we selected itThe first heading is Decide Whether to Migrate at All.

Weigh upMPL-2.0 applies to modified redistributions of this file.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMPL-2.0

Testing · HashiCorp

Provider test patterns

Writes Plugin Framework acceptance tests, including state checks and import verification.

terraformacceptance-testsprovider

Why we selected itThe description names TestCase, TestStep, ConfigStateChecks and import testing.

Weigh upMPL-2.0 applies to a modified redistribution of this file.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMPL-2.0

Infrastructure development · HashiCorp

Push to HCP Packer registry

Pins HashiCorp's Packer skill for pushing build metadata to an HCP Packer bucket, including the environment variables the build expects.

packerhcpregistry

Why we selected itThe catalogue has HashiCorp's aws-ami-builder Packer skill and not this HCP Packer registry skill.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMPL-2.0

App Service deployment · Microsoft

Python App Service deploy

Pins Microsoft's Python App Service deploy skill, which forbids health probes after deploy and forbids az webapp up.

azureapp-servicepython

Why we selected itAfter deploy returns, the skill forbids log tail, curl, Invoke-WebRequest, and any other health probe, then says to stop the turn.

Weigh upThis review did not execute the resource.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Databases · Qdrant

Qdrant clients SDK

Qdrant skill that lists the official client SDKs with their repositories and installation lines, points to the REST and gRPC API references, and fetches code examples from a remote snippet search.

qdrantvector-searchgrpc

Why we selected itSKILL.md names six officially supported client SDKs, links each to its qdrant GitHub repository, and prints an installation line for five of them: pip install qdrant-client[fastembed], npm install @qdrant/js-client-rest, cargo add qdrant-client, go get github.com/qdrant/go-client and dotnet add package Qdrant.Client.

Weigh upskills/qdrant-clients-sdk last changed in a commit dated 2026-06-05; the pinned revision is dated 2026-10-21. The README disclaimer says the skills are under active development and that content and structure may change between versions.

Source reviewed

4 sources · 2026-10-21

Read the full review Upstream setup linkedApache-2.0

Data and analytics · dbt Labs

Querying the dbt information schema

Answers from the project metadata, with two routes, instead of guessing the DAG.

dbtmetadatalineage

Why we selected itThe description limits the skill to a dbt v2 project's own metadata.

Weigh upA metadata query is not a production build.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Documents · Trail of Bits

React-PDF

Generates PDF documents with @react-pdf/renderer in TypeScript and JSX — flexbox layout, inline SVG, registered local fonts, Knuth-Plass typesetting — with a when-not-to-use map and crash-avoidance rules for fonts and async rendering.

pdfpdf-generationreact

Why we selected itStates when NOT to use it — reading PDFs, filling forms, HTML conversion — routing each to pdfplumber, pypdf/pdftk or WeasyPrint instead of stretching one tool across every PDF task.

Weigh upNode-only: @react-pdf/renderer renders server-side or in scripts; this is not a browser-PDF skill.

Source reviewed

6 sources · 2026-10-07

Read the full review Install command in reviewMIT

Observability · Anthropic

Receipts

Builds a personal Claude Code impact report from this machine's session transcripts and local git history: what shipped, in which projects, and each project's share of usage, as a Markdown report and an HTML receipt in the home directory.

claude-codereceiptsgit-history

Why we selected itMining is a local Node script, scripts/mine-transcripts.mjs, that reads ~/.claude/projects/**/*.jsonl and runs read-only git commands; the README says the only data that reaches the model is the git user name, aggregate counts and project names.

Weigh upLast commit touching plugins/receipts/skills/receipts at revision 41f32e40: 2026-10-08 (0baf7d19).

Source reviewed

4 sources · 2026-10-21

Read the full review Install command in reviewApache-2.0

Database engineering · redis

Redis Cluster Key and Read Design

Plan hash-tag placement and replica reads for sharded Redis workloads.

database-engineeringredis

Why we selected itHash-tag and replica references make key co-location and eventual consistency explicit; entity-level tags prevent unrelated tenant identifiers from sharing a slot.

Weigh upThe blanket statement that all pipelines require one slot is overbroad: behavior depends on cluster-aware client routing and transaction mode. Shared tags can create hot slots; replica reads can be stale.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Database engineering · redis

Redis Connection Engineering

Review Redis pooling, batching, incremental scans and client timeout choices.

database-engineeringredis

Why we selected itThe definition separates intentional blocking queue operations from server-blocking full scans and distinguishes nontransactional batching from atomic transactions.

Weigh upRetry-on-timeout can repeat non-idempotent writes after an ambiguous response. Pool exhaustion, multiplexing and caching support are client-specific; the numeric timeout examples are starting points.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Database engineering · redis

Redis Data Modeling

Choose Redis structures and key namespaces from access patterns.

database-engineeringredis

Why we selected itThe data-type table maps operations to structures and contrasts per-field Hash updates with whole-object string rewrites; key hierarchy is a separate design step.

Weigh upNamespace prefixes aid organization but do not enforce tenant isolation. JSON, vector and search capabilities require compatible server editions and versions.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Agent memory · redis

Redis Iris Agent Memory

Integrate session events and searchable long-term memory with Redis Cloud Agent Memory.

agent-memoryredis

Why we selected itThe definition distinguishes session and long-term tiers; the bulk reference uses caller IDs and partial-success results, while promotion explicitly permits delayed or missing extraction.

Weigh upPromotion is asynchronous and a successful event write does not establish a searchable fact. The bulk example slices the first batch only; iterate remaining batches and validate ID characters before use.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Observability · redis

Redis Observability Review

Select Redis metrics and diagnostic commands for latency, memory and connection incidents.

observabilityredis

Why we selected itThe definition groups memory, connection, throughput and persistence metrics and offers command-level investigation; the separate metrics example lacks the zero-denominator guard used in the main definition.

Weigh upFixed hit-ratio and memory thresholds are examples, not universal alerts; sustained blocked clients may be expected for queues. The metrics reference divides by zero when there are no hits or misses.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Search and retrieval · redis

Design lexical, filtered and vector retrieval with Redis Search indexes.

search-and-retrievalredis

Why we selected itThe definition distinguishes filter-then-vector retrieval from lexical/vector score fusion and separates retrieval from aggregation and index-management commands.

Weigh upThe quoted exact-match speedup and HNSW recall figures are not guarantees. New commands require version checks; index creation does not mean backfill is ready for an alias switch.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Security · Redis, Inc.

Redis security

Pins Redis's production-hardening skill for authentication, ACLs, and network binding, and records that any one of those layers is not enough on its own.

redisacltls

Why we selected itThe catalogue has the Redis semantic-cache skill and not this security skill from the same repository.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Caching · Redis, Inc.

Redis semantic cache

Redis skill for LangCache cache-aside search and set on Redis Cloud, with separate caches per task and stated similarity thresholds.

redislangcachesemantic-cache

Why we selected itThe catalogue has a Redis MCP server and no LangCache skill.

Weigh upThis review did not execute the skill or call LangCache.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Data pipelines · Redpanda

Redpanda streaming

Redpanda reference skill for using the broker through its Kafka-compatible API: rpk topic and record commands, producer and consumer settings, transactions, client compatibility, tiered storage and licensed enterprise topic features.

redpandastreamingtransactions

Why we selected itSKILL.md works through the Kafka API with rpk commands (topic list, create, produce, consume, alter-config, delete) and a franz-go Go snippet, plus producer and consumer property tables, the transaction call sequence with isolation.level=read_committed, and tiered storage settings.

Weigh upskills/streaming last changed in a commit dated 2026-10-20 (an update to references/SOURCES.md); the pinned revision is dated 2026-10-20.

Source reviewed

5 sources · 2026-10-21

Read the full review Upstream setup linkedApache-2.0

Creative production · Remotion

Remotion Captions

Routes Remotion caption work through one typed Caption contract: transcribing audio with whisper.cpp, displaying TikTok-style paged captions with word highlighting, and importing existing .srt subtitle files.

captionssubtitlessrt

Why we selected itDefines one typed Caption contract (text, startMs, endMs, timestampMs, confidence, pageBreakAfter) and routes transcription, display and SRT import through it, so caption data cannot drift between pipeline stages.

Weigh upRemotion-only: the Caption type and packages are Remotion’s, so this is not a general subtitle toolkit.

Source reviewed

7 sources · 2026-10-15

Read the full review Install command in reviewRemotion License (custom, source-available; free for individuals and small companies)

Creative production · Remotion

Remotion Video Workflows

Routes video work to focused Remotion guides for creation, preview, rendering, media and reusable video applications.

videoremotionreact

Why we selected itKeeps new-project scaffolding separate from adding a composition to an existing project.

Weigh upThis is a router into a larger reference package, so copying the definition alone loses the working guidance.

Source reviewed

8 sources · 2026-10-15

Read the full review Install command in reviewNo redistribution license found

Development workflow · Superpowers

Requesting Code Review

Prepares a requirements-aware review brief around an explicit base and head revision.

code-reviewrequirementssubagents

Why we selected itGives the reviewer concrete requirements, implementation context and a bounded Git comparison.

Weigh upA base-to-head comparison excludes uncommitted changes; include those separately when they are in scope.

Source reviewed

5 sources · 2026-10-14

Read the full review Install command in reviewMIT

Code review · Trail of Bits

Review Walkthrough

Turns the committed diff of the current branch into a standalone HTML page of ordered review steps, each with an explanation and severity-tagged findings anchored to diff lines.

code-reviewdiff-reviewgit-diff

Why we selected itIts bundled renderer compares every step with the complete captured patch, so a step whose per-file diff was split, rewritten, omitted or invented is refused rather than shown.

Weigh upRuns only on explicit invocation (disable-model-invocation: true): /review-walkthrough:review-walkthrough in Claude Code, $review-walkthrough in Codex.

Source reviewed

9 sources · 2026-10-22

Read the full review Upstream setup linkedCC-BY-SA-4.0

Development workflow · getsentry

Reviewer-Facing PR Writing

Write or refresh PR titles and bodies from the full branch change and its reviewer impact.

development-workflowgetsentry

Why we selected itNew pull requests are created as drafts after the whole branch diff is read, not from the latest commit alone.

Weigh upCreate and update both use authenticated gh, so they publish the title and body. The skill does not merge or request reviewers.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Languages and tooling · GitHub community

Ruff Recursive Fix

Enforces Python code quality with scoped Ruff runs: safe fixes first, unsafe fixes second with diff review each pass, then targeted manual remediation under explicit stop criteria.

pythonrufflinting

Why we selected itFix passes are ordered by blast radius: safe autofixes with diff review and formatting first, unsafe fixes only if findings remain, then minimal manual edits.

Weigh upUnsafe autofixes can change behavior; the workflow's value depends on actually performing its mandated diff reviews rather than waving them through.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Security review · Trail of Bits

Rust review

Reviews Rust unsafe, FFI and concurrency, and lists rationalisations the reviewer must reject.

rustunsafesecurity

Why we selected itThe definition has both When to Use and When NOT to Use headings.

Weigh upCC-BY-SA-4.0 requires share-alike and attribution for an adaptation you redistribute.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Testing · trailofbits

Ruzzy Ruby Fuzzing

Create Ruby and native-extension fuzz harnesses with coverage tracing and sanitizer-aware execution.

testingtrailofbits

Why we selected itThe handbook is specifically for Ruzzy, and it says pure Ruby needs a second tracer script that C-extension fuzzing does not.

Weigh upLD_PRELOAD is required for the sanitizer and must not be exported, because the file says that interferes with other programs.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Machine learning · huggingface

SageMaker Serving Image Selection

Choose a region-compatible serving container from the AWS catalogue and inspect architecture, runtime and mirroring requirements.

machine-learninghuggingface

Why we selected itIt picks a SageMaker serving image and prefers a compatible Hugging Face curated family over a newer generic engine tag.

Weigh upThe skill does not create the SageMaker endpoint.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewApache-2.0

Testing and reliability · Trail of Bits

SARIF Parsing

Extracts and summarizes SARIF findings with reusable jq examples and a Python helper.

sarifstatic-analysisci

Why we selected itExplains run structure and severity inheritance instead of treating every absent result.level as the same severity.

Weigh upThe helper is not a complete SARIF validator: invocation-level ruleConfigurationOverrides and extension rule components are not resolved.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Databases · JetBrains

Schema Migration Planner

Plans database schema changes as phased expand, dual-write, backfill, switch-read, contract sequences with explicit lock, replication and rollback-safety analysis, instead of a single DDL statement.

sqlschema-migrationexpand-contract

Why we selected itEncodes expand/contract migration as a planning contract: classify the change as additive, destructive, semantic or data-moving, decide whether old and new application versions must coexist, then phase the rollout and treat cleanup as a separate step behind an explicit proof threshold.

Weigh upFramed for Kotlin plus Spring services; the workflow is dialect-generic but the tooling notes lean on Flyway and Liquibase in that ecosystem.

Source reviewed

6 sources · 2026-10-13

Read the full review Install command in reviewMIT (upstream Kotlin backend agent skills)

Code review · Trail of Bits

Second Opinion

Captures one patch (uncommitted work, a branch against a base, or a single commit) and has the Codex, Antigravity or Gemini CLI review it, then reports each reviewer's findings and where two reviews agree or differ.

code-reviewdiff-reviewopenai

Why we selected itWhen two providers are chosen, both receive the same captured patch, and the skill summarises their agreements and disagreements "without turning agreement into proof".

Weigh upSends the captured diff, the user's focus and applicable project instructions to OpenAI (codex exec) or Google (agy, gemini) through CLIs the user installs and signs in to.

Source reviewed

8 sources · 2026-10-22

Read the full review Install command in reviewCC-BY-SA-4.0

Security · Sentry

Secret Serialization Review

Finds a credential field that survives generated serialization, and the sink that stringifies the whole object.

secretsserializationdataclasses

Why we selected itThe exclusion table says dataclasses.field(repr=False) does not block asdict, Pydantic Field(exclude=True) does not block repr, and neither flag blocks vars, __dict__ or pickle.

Weigh upThe repository README describes the pack as skills for Sentry employees. The documented whole-pack install includes skills this catalogue has withheld.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Security review · Addy Osmani

Security and hardening

Starts with a threat model and a three-tier boundary, not with a generic vulnerability list.

securityowasphardening

Why we selected itIt starts with a threat model: trust boundaries, named assets and a STRIDE pass over each boundary, with abuse cases written next to use cases.

Weigh upEach control links to references/hardening-patterns.md inside the skill, and the review checklist to ../../references/security-checklist.md, which the README says a per-skill npx install does not copy.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Security review · Cloudflare

Security Audit

Source-first security audit with independent verification, a coverage ledger and a sandbox requirement before any target code runs.

security-auditvulnerability-researchsandboxing

Why we selected itTarget-controlled builds, tests and processes run only inside an OS-enforced sandbox with no external network, an allowlisted environment and resource limits; without every control the lead stays needs_validation.

Weigh upFull audit mode needs a host with parallel sub-agents and Node.js for the bundled validators. The companion Markdown files and both validators were not opened for this review.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Security · trailofbits

Security Audit Preparation

Prepare a frozen, buildable codebase and evidence package for an external security review.

securitytrailofbits

Why we selected itThe checklist identifies copied code and modifications, documents actors/privileges and on-chain/off-chain assumptions, and treats a reproducible frozen source as an assessment input.

Weigh upCoverage targets and preparation timelines are planning suggestions; a clean scanner report or high coverage does not establish audit readiness by itself.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Security · trailofbits

Semgrep Repository Scan

Plan and execute language-scoped Semgrep scans with explicit partial, failed and uncovered scan states.

securitytrailofbits

Why we selected itThe scan skill treats the user's request to scan as not enough approval, and run-scans.sh is what puts --metrics=off on every generated command.

Weigh upOSS Semgrep, as this file describes it, cannot track data flow across files. Third-party rules are cloned into the output directory and then deleted.

Source reviewed

4 sources · 2026-10-14

Read the full review Upstream setup linkedCC-BY-SA-4.0

Security and privacy · Trail of Bits

Semgrep Rule Creator

Turns a concrete bug pattern into a Semgrep rule with positive and negative test cases.

semgrepstatic-analysisregression-tests

Why we selected itRequires vulnerable and safe examples before rule implementation, then runs Semgrep validation and rule tests.

Weigh upA passing fixture suite establishes behavior on those examples; it does not prove complete detection or an acceptable false-positive rate on other repositories.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Security · trailofbits

Semgrep Rule Variants

Port one verified rule to target languages with applicability review and language-specific positive and negative fixtures.

securitytrailofbits

Why we selected itThis skill ports one finished Semgrep rule to named languages. A new rule from a bug description is sent to semgrep-rule-creator instead.

Weigh upallowed-tools includes WebFetch and Write. Resume does not survive a new session, because runs live under that session's directory.

Source reviewed

3 sources · 2026-10-14

Read the full review Upstream setup linkedCC-BY-SA-4.0

Research and documentation · Sentry

Sentry API Endpoint Documentation

Aligns a Sentry API endpoint’s runtime response, TypedDict definitions and drf-spectacular OpenAPI schema.

openapidjango-rest-frameworksentry

Why we selected itChecks runtime serialization against types and schema, including the difference between an absent field and a nullable one.

Weigh upThe file paths, ownership checks and validation commands are specific to Sentry; this is not a drop-in guide for every Django API.

Source reviewed

8 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Documentation editing · Functional Software, Inc. dba Sentry

Sentry brand guidelines

Writes Sentry UI copy and defaults to Plain Speech unless the context calls for Sentry Voice.

copytonesentry

Why we selected itPlain Speech is the default, and Sentry Voice is limited to surfaces such as 404s, empty states, and onboarding.

Weigh upThe handbook links are references. The skill does not fetch them and does not check a rendered page.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Developer Tools · Sentry

Sentry Claude Settings Audit

Audits repository tooling to generate least-privilege Claude Code settings.json read-only allowlists.

sentryclaude-codepermissions

Why we selected itInspects package manifests and monorepo configurations to discover project-specific read-only test, lint, and type-check commands.

Weigh upGenerated settings recommendations require human confirmation before copying into .claude/settings.json.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Code review · Functional Software, Inc. dba Sentry

Sentry code review

Reviews a change for runtime, query, security, design, and test gaps, and does not block on style.

reviewsecuritysentry

Why we selected itThe checklist separates defects that should block a change from style comments that should not.

Weigh upThe file is a checklist. It has no gh command, so it does not submit a GitHub review or change the code.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Code editing · Functional Software, Inc. dba Sentry

Sentry code simplifier

Refines recently modified code for clarity and says to leave its behaviour unchanged.

refactorclaritysentry

Why we selected itThe skill's first rule is that behaviour, outputs, and features stay the same while the shape of the code changes.

Weigh upThe definition assumes CLAUDE.md standards but specifies no fallback when that file is absent.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Git commits · Functional Software, Inc. dba Sentry

Sentry commit message

Drafts one conventional commit on a feature branch, and stops if the branch is still main or master.

gitcommitsentry

Why we selected itThe skill stops a commit that is still on main or master unless the user explicitly asked for that, and it asks for one reviewable change per commit.

Weigh upIt tells the agent to pass paragraphs as separate git commit -m arguments and not to open an editor. It does not push the commit or open a pull request.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Git branches · Functional Software, Inc. dba Sentry

Sentry create branch

Creates one local branch with git checkout -b and does not say to push or to commit.

gitbranchsentry

Why we selected itThe workflow creates the branch without asking for a name confirmation, and it only checks out the default branch when the user says so.

Weigh upWith no local diff it invents a generic description such as repo-maintenance. It does not commit, push, or open a pull request.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Testing and reliability · Sentry

Sentry Find Bugs

Reviews a branch diff against surrounding code and existing tests, reporting only substantiated issues and explicit coverage limits.

code-reviewgit-diffsecurity-review

Why we selected itRequires the complete diff to be read before findings and follows changed behavior into surrounding implementations and tests.

Weigh upThe initial command compares committed branch history and omits uncommitted edits; inspect those separately if they are in scope.

Source reviewed

7 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Development workflow · getsentry

Sentry Frontend Queue Triage

Prepare a bounded archive plan for the Sentry internal javascript issue queue and apply only the approved rows.

development-workflowgetsentry

Why we selected itThe skill only archives issues in the sentry/javascript project, and only with untilEscalating, after the user approves the displayed plan.

Weigh upIt needs the Sentry MCP at mcp.sentry.dev and the update_issue tool. If that tool is missing it stops.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Review queue · Functional Software, Inc. dba Sentry

Sentry GitHub review requests

Lists unread review-requested notifications for open pull requests tied to named teams.

githubreviewssentry

Why we selected itThe skill's only declared tool is Bash, and both the skill and the script require an authenticated gh CLI.

Weigh upThe default organisation is getsentry. Another org has to be passed explicitly.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Development workflow · getsentry

Sentry Presentation Builder

Create a branded React slide deck with charts only where supplied measurements support them.

development-workflowgetsentry

Why we selected itCharts are allowed only for numbers that are already in the source. The skill forbids estimated or invented series.

Weigh upSKILL.md forbids category pills above headings, while references/design-system.md says tags are used on every slide. Those two files disagree.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

AI application development · Sentry

Sentry Prompt Optimizer

Starts from a written contract before rewriting a prompt, and names the model family as a choice rather than a default.

promptsprompt-evalstoken-cost

Why we selected itOptimization starts from a written contract and an eval slice, not from a rewrite.

Weigh upThe failure-mode list includes editing before the eval target is defined.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Pull requests · Functional Software, Inc. dba Sentry

Appends an Issues block to the current pull request after reading its body.

githubpull-requestsentry

Why we selected itThe skill edits an existing pull request body and stops when the branch has no pull request.

Weigh upIt uses authenticated gh, so it can change the pull request body on GitHub. It does not open the pull request, create the GitHub issue, or create the Linear ticket.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Documentation editing · Functional Software, Inc. dba Sentry

Sentry SRED work summary

Collects one person's GitHub, Notion and Linear links for a year and groups them in a private Notion document.

notiongithublinear

Why we selected itThe workflow collects a person's GitHub, Notion, and Linear links for a fixed February-to-January window and writes them to a private Notion page.

Weigh upNotion and Linear access are MCP connections, and GitHub may be gh. The skill does not file a tax claim or compute an amount.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Code editing · Functional Software, Inc. dba Sentry

Sentry typing exclusion worker

Removes an assigned batch of modules from a mypy exclusion list, then fixes and checks only that batch.

mypypythontyping

Why we selected itThe worker is limited to one assigned mypy batch and a fixed summary for a manager.

Weigh upOnly assigned exclusion entries may be removed, and cross-team modules need explicit approval.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Observability · Anthropic

Session Report

Runs a bundled Node analyzer over local Claude Code transcripts and fills an HTML template with token use by project, subagent type and skill, cache breaks and the most expensive prompts, plus a few findings and suggestions.

claude-codetoken-usageprompt-caching

Why we selected itIts numbers come from a bundled script, analyze-sessions.mjs, which breaks token use down by project, subagent type and skill and lists cache breaks and the top prompts; the model pastes that JSON into the template and adds 3-5 findings, expressed as shares of total tokens, and 1-4 suggestions.

Weigh upLast commit touching plugins/session-report/skills/session-report at revision 41f32e40: 2026-04-09 (9dc3809e).

Source reviewed

6 sources · 2026-10-21

Read the full review Upstream setup linkedApache-2.0

Model evaluation · Braintrust

Size an Eval Dataset

Computes or audits eval sample sizes, detectable effects and clean-trial counts, with clustering and pairing assumptions stated.

evaluationsample-sizestatistics

Why we selected itIt forbids returning a sample size without stating effect size, baseline rate, risk tolerance, clustering, pairing, confidence and power, and requires a sensitivity range rather than one number.

Weigh upThe body cites anchors in reference.md, which is not in the skill directory at this revision. The header points to references/power-analysis.md, which was not opened.

Source reviewed

5 sources · 2026-10-13

Read the full review Install command in reviewApache-2.0

Agent development · Anthropic

Skill Development for Claude Code Plugins

Guides writing a skill that ships inside a Claude Code plugin: directory layout, a third-person trigger description, an imperative SKILL.md body and progressive disclosure into references/, examples/ and scripts/.

claude-codeplugin-devskills

Why we selected itIt adapts the skill-creator method to plugin skills, which is what separates it from the general skill-creator: plugin skills are created by hand under plugin-name/skills/<skill-name>/ instead of with init_skill.py, ship with the plugin rather than as separate ZIP files, and are reviewed with plugin-dev's skill-reviewer agent.

Weigh upLast commit touching plugins/plugin-dev/skills/skill-development at revision 41f32e40: 2026-02-05 (2438937b).

Source reviewed

5 sources · 2026-10-21

Read the full review Upstream setup linkedApache-2.0

Skill review loop · Trail of Bits

Skill improver

Pins the skill loop that stops when plugin-dev is missing, refuses an inline rewrite, and does not commit.

skillreviewworkflow

Why we selected itSeveral matching skills are a question to the user, and a missing plugin-dev reviewer stops the run instead of a self-review.

Weigh upThe resource was not executed.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Security review · Sentry

Skill scanner

Audits a skill before install, with a bundled script and explicit confidence levels.

skillssupply-chainreview

Why we selected itA Bundled Script heading means the check is not only prose.

Weigh upThe bundled script was not executed in this review.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Security review · Trail of Bits

Slicing Code Context

Uses Trailmark to cut a bounded, graph-informed slice of source around a function, class, caller, callee, call path, entry point or line range and hands it to a smaller worker model that may only propose edits.

code-understandingcall-pathssubagents

Why we selected itBuilds a deterministic Trailmark slice packet and gives a worker only the task and that packet; the worker may propose changes but not apply them.

Weigh upWorkers may only propose changes; the skill is not for tasks that need direct worker edits.

Source reviewed

6 sources · 2026-10-22

Read the full review Upstream setup linkedCC-BY-SA-4.0

Review smart-contract architecture, implementation and verification against a structured guideline set.

securitytrailofbits

Why we selected itThe detailed assessment distinguishes storage layout, inheritance order, initializer front-running, function shadowing and direct implementation use.

Weigh upThis is an advisory checklist, not a security certificate. Compiler behavior and proxy semantics require version-specific inspection; not every project should adopt the same migration pattern.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Security · trailofbits

Smart Contract Maturity Assessment

Assess a blockchain codebase against an explicit nine-category maturity rubric.

securitytrailofbits

Why we selected itThe criteria require more than event declarations for auditing maturity and compare arithmetic precision and low-level code against explicit reference behavior.

Weigh upRubric scores are reviewer judgments, not measured probabilities of safety. Unobserved off-chain processes must remain unknown until corroborated; an average can hide a critical weakness.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Security · trailofbits

Smart Contract Security Workflow

Combine static analysis, feature checks, invariants and manual review in a smart-contract workflow.

securitytrailofbits

Why we selected itThe workflow separates known-pattern scans from state/authorization diagrams, property design and manual review of privacy, transaction ordering and external assumptions.

Weigh upDetector/spec counts in the definition are version-dependent. A completed symbolic or fuzzing run only supports the properties and explored paths actually checked.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Security review · Trail of Bits

Solana vulnerability scanner

Pins Trail of Bits' Solana scanner, which refuses a short coverage table, keeps the version-scoped row, and does not treat passing lints or an Anchor attribute as a full check.

solanaanchorsecurity

Why we selected itA short or empty coverage table is an incomplete scan, and the version-scoped pattern stays in the table with the targeted solana-program version instead of being dropped.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Development workflow · addyosmani

Source-Driven Implementation

Ground version-sensitive code changes in official documentation and explicit source citations.

development-workflowaddyosmani

Why we selected itThe workflow orders sources by authority and expressly excludes model-targeted directives and unrelated outbound endpoints from copied documentation examples.

Weigh upOfficial documentation can be incomplete or refer to another version; a citation is not runtime proof. Retrieved instructions must not expand the authorized task.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Security · trailofbits

Specification to Code Compliance

Compare authoritative requirements with implementation evidence and retain undecidable or unchecked requirements.

securitytrailofbits

Why we selected itThe skill compares code with a document that says what the code is supposed to do. It is not a general bug hunt, and it is not a documentation rewrite.

Weigh upDeclared tools are Workflow, Task, Read, Grep, and Glob. There is no WebFetch in the frontmatter. The check writes local report files and is not meant to be done inline for a long requirement list.

Source reviewed

3 sources · 2026-10-14

Read the full review Upstream setup linkedCC-BY-SA-4.0

Planning and delivery · Matt Pocock

Specification to Dependency-Aware Tickets

Splits an agreed plan into small vertical slices, with explicit blocking relationships and a review before publication.

ticketsdependenciesplanning

Why we selected itAsks for user agreement on the proposed breakdown before creating the tickets.

Weigh upTicket publication is not transactional: partial failures or repeated runs can leave duplicates and incomplete dependency links.

Source reviewed

16 sources · 2026-10-15

Read the full review Install command in reviewMIT

Observability · Axiom, Inc.

SPL to APL

Translate a Splunk SPL fragment into Axiom APL, including the argument reversal and the time filter the SPL picker does not carry over.

splunkaplqueries

Why we selected itThe translation rules include argument order and time filters that a mechanical rename would miss.

Weigh upAPL join is described as preview, limited to 50k rows, and only inner, innerunique, and leftouter.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewMIT

Designs and audits Spring Security configurations: explicit public-endpoint allowlists, JWT and OAuth2 resource-server trust boundaries with issuer, audience and clock-skew validation, ordered filter chains, method security aligned with request security, and CSRF treated as a trust-model decision.

spring-securityauthenticationauthorization

Why we selected itAudits before it configures: enumerate public endpoints explicitly, model who can do what to which resource under which tenant before writing matcher code, and prefer deny-by-default so new endpoints start closed.

Weigh upSpring Security plus Kotlin specifics; servlet versus reactive stack differences are flagged but only the Spring ecosystem is covered.

Source reviewed

6 sources · 2026-10-13

Read the full review Install command in reviewMIT (upstream Kotlin backend agent skills)

Databases · GitHub community

SQL Object Impact Analysis

Before a schema change, traces every reference to a table, column, stored procedure, view or trigger across both SQL objects and application code, and produces a structured blast-radius report with confidence-classified findings.

sqlschema-changesimpact-analysis

Why we selected itClassifies every finding as Direct, Indirect or Uncertain and forbids presenting an Uncertain finding as Direct — the report must state the search method used and its limits instead of implying completeness.

Weigh upStatic text search cannot resolve every dynamic-SQL construction or reflection-based ORM mapping; the skill says so itself and requires the report to say so.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Documentation · Stripe

Stripe docs CLI

Pins Stripe's skill for reading docs through the Stripe CLI, including the login step it requires for gated pages.

stripedocscli

Why we selected itThe catalogue has Stripe billing and payments skills and not this documentation-lookup skill.

Weigh upThis review did not execute the resource.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Payments · Stripe

Stripe Pay

Stops a business-to-business Stripe transfer until the user has seen the exact command and confirmed the handle, amount and currency.

stripetransferscli

Why we selected itA transfer is not run until the user has seen the exact command and confirmed it.

Weigh upThe skill forbids sending money unless the user explicitly asks, and it forbids -y before confirmation.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewMIT

Substrate pallet audit · Trail of Bits

Substrate vulnerability scanner

Pins the Substrate scanner's rule that a short or empty coverage table is an unfinished scan, and that ensure_signed is not a privileged origin.

substrateframepallet

Why we selected itA coverage table with fewer than every pattern row, or an empty verdict, must be reported as incomplete, and not having looked is not an n/a.

Weigh upThe resource was not executed.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Databases · Supabase

Supabase Development

Covers the full Supabase surface beyond Postgres tuning: Auth flows, client and SSR integration, CLI and MCP use, declarative or imperative schema changes, and structured debugging.

supabaseauthmigrations

Why we selected itSeparates declarative schemas from imperative migrations and pairs each path with committing guidance instead of treating all schema edits alike.

Weigh upTriggers on any Supabase task, so it can load for work the narrower Postgres skill would handle better; pick the skill per task.

Source reviewed

6 sources · 2026-10-13

Read the full review Install command in reviewMIT

Connects Postgres schema and query decisions to worked SQL examples, connection guidance and row-level security checks.

postgresqlsqlindexes

Why we selected itPairs incorrect and corrected SQL with query-plan examples instead of offering only generic database advice.

Weigh upIllustrative speedups and blanket indexing rules are not measurements of your workload; inspect actual plans and write costs.

Source reviewed

7 sources · 2026-10-13

Read the full review Install command in reviewMIT

Testing and reliability · Jesse Vincent / Superpowers

Superpowers Receiving Code Review

Checks review feedback against the code and project constraints before making and testing each requested change.

review-feedbackregressionsverification

Why we selected itTreats reviewer suggestions as hypotheses to verify, with explicit checks for existing behavior, platform constraints and unused functionality.

Weigh upIts conversational tone rules are a style preference, and its pause-on-ambiguity rule may be too restrictive for independent review items.

Source reviewed

7 sources · 2026-10-14

Read the full review Install command in reviewMIT

Testing and reliability · Jesse Vincent / Superpowers

Superpowers Systematic Debugging

Moves debugging from reproducible evidence through a single hypothesis to a verified fix, with a stop point for repeated failed attempts.

debuggingroot-causereproduction

Why we selected itAsks for evidence at component boundaries and backward tracing of bad values before proposing a patch.

Weigh upThe bundled polluter helper assumes npm tests, hides their output and swallows their failing exit statuses.

Source reviewed

8 sources · 2026-10-14

Read the full review Install command in reviewMIT

Testing and reliability · Jesse Vincent / Superpowers

Superpowers Test-Driven Development

Makes an observed failing behavior test the entry point for implementation and checks that assertions can detect real production mistakes.

tddregression-teststest-design

Why we selected itRequires verifying why a test fails before changing implementation, then running the project suite after the fix.

Weigh upThe workflow is deliberately prescriptive: it tells an agent to discard implementation written before tests and seek permission for exceptions.

Source reviewed

7 sources · 2026-10-14

Read the full review Install command in reviewMIT

Testing and reliability · Jesse Vincent / Superpowers

Superpowers Verification Before Completion

Requires a fresh command result or direct check for each completion claim and distinguishes passing tests from fulfilling requirements.

verificationevidencebuilds

Why we selected itMaps common claims to appropriate evidence, including compiler/build status, original bug reproduction and requirements checks.

Weigh upThis is a behavioral checklist, not an automatic enforcement mechanism or runtime test harness.

Source reviewed

6 sources · 2026-10-14

Read the full review Install command in reviewMIT

Research and documentation · Jesse Vincent / Superpowers

Superpowers Writing Plans

Turns an agreed specification into file-specific implementation tasks with verification steps and a structured plan review.

implementation-planningspecificationstests

Why we selected itMakes each task name concrete files, expected behavior and meaningful failing/passing checks rather than leaving implementation placeholders.

Weigh upDetailed code in a plan can become stale as the repository changes; planned checks are not evidence that tests have run.

Source reviewed

8 sources · 2026-10-14

Read the full review Install command in reviewMIT

Security and privacy · Trail of Bits

Supply Chain Risk Auditor

Collects dependency identity, advisory and repository-maintenance evidence while retaining unassessable cases.

dependenciessupply-chainprovenance

Why we selected itKeeps ecosystem and resolved package versions attached to evidence and distinguishes unknown results from clean checks.

Weigh upDocumented project inputs include npm package-lock/shrinkwrap, Python uv.lock and supported Go module files; other lock formats need a separate extraction step.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Databases · SurrealDB

SurrealQL

SurrealDB skill for writing, adapting and troubleshooting SurrealQL statements, with statement tables linked to the docs and instructions to validate and format queries with the SurrealDB CLI and surql-fmt.

graphschemaqueries

Why we selected itTables map SurrealQL query, schema and control-flow statements, from SELECT and LIVE SELECT to DEFINE TABLE, DEFINE INDEX and BEGIN / COMMIT, to their surrealdb.com docs pages, with references for querying, schema and values.

Weigh upThe skill directory skills/surrealql last changed in a commit dated 2026-06-16; the pinned revision is dated 2026-09-29. SKILL.md metadata says version 0.2.0.

Source reviewed

3 sources · 2026-10-21

Read the full review Install command in reviewMIT

Framework development · Svelte

Svelte Core Best Practices

Svelte-maintained guidance for reactivity, component composition and integration with browser APIs.

sveltereactivitycomponents

Why we selected itExplains when to derive state instead of synchronizing it with effects, and when reactive wrappers or subscriptions are needed.

Weigh upThe references span several Svelte releases and include experimental async behavior. Check the installed compiler version and required flags before applying an example.

Source reviewed

17 sources · 2026-10-13

Read the full review Install command in reviewMIT

Observability · Grafana Labs

Synthetic monitoring checks

Separates a reliability probe from a load test, and names the assertion mistake that leaves probe_success green.

synthetic-monitoringk6probes

Why we selected itThe first heading says reliability monitoring, not load testing.

Weigh upNo probe was deployed in this review.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Development workflow · GitHub community (awesome-copilot)

Task Context Map

Map implementation files, dependencies, tests and reference patterns before changing code.

development-workflowgithub

Why we selected itThe template keeps files to modify, direct dependencies, tests and reference patterns separate and includes explicit API/schema/configuration risk checks.

Weigh upA search-based map can miss runtime, generated and external dependencies; the map is a starting hypothesis rather than exhaustive coverage.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMIT

Telegram access control · Anthropic

Telegram access

Pins the Telegram access skill, which resolves a state directory, refuses channel-originated edits, and will not approve a pairing without its code.

telegramaccess-controlpairing

Why we selected itThe skill refuses access changes that arrive as a Telegram or Discord message and says those mutations must not be downstream of untrusted input.

Weigh upThis review did not execute the resource.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Telegram channel setup · Anthropic

Telegram configure

Pins the Telegram setup skill, which stores a bot token with mode 600 and says pairing should end once the allowlist is known.

telegrambot-tokenallowlist

Why we selected itThe skill resolves the state directory, writes the bot token there, and says to chmod 600 the env file because the token is a credential. The tool list includes that chmod.

Weigh upThis review did not execute the resource.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Framework development · Temporal Technologies

Temporal Developer

Temporal SDK guidance across seven languages, centred on replay determinism, versioning and wrapping LLM calls in activities.

temporaldurable-executionworkflows

Why we selected itIt explains history replay and why a non-determinism error blocks a workflow, then routes to core and language-specific determinism, versioning and pitfalls references.

Weigh upMost guidance sits in the references directory; only SKILL.md and references/core/ai-patterns.md were read. The README lists Rust as Public Preview and PHP as not yet supported.

Source reviewed

4 sources · 2026-10-13

Read the full review Install command in reviewMIT

Cloud operations · GitHub community

Terraform AzureRM Set Diff Analyzer

Separates false-positive Terraform plan diffs — order-only changes in Set-type attributes — from real resource changes, using a stdlib-only Python analyzer over terraform show -json output with CI-ready exit codes.

terraformazureplan-review

Why we selected itShips a runnable analyzer rather than prose: scripts/analyze_plan.py is standard-library-only Python 3.8+ that classifies order-only versus real changes and exposes --exit-code semantics for CI gating.

Weigh upAzureRM-focused: the position-comparison mechanism is general Terraform, but the attribute reference and examples cover Azure resources only.

Source reviewed

9 sources · 2026-10-15

Read the full review Install command in reviewMIT

Infrastructure development · HashiCorp

Terraform ephemeral resources

Implement short-lived values with Open, optional Renew/Close, private lease handles and explicit non-persistence boundaries.

terraformprovidersecrets

Why we selected itIt implements Terraform 1.10 ephemeral resources whose values are never written to state or plan.

Weigh upEphemeral values cannot be stored in regular resource attributes.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMPL-2.0

Infrastructure development · HashiCorp

Terraform module refactor

Extract Terraform modules while explicitly migrating existing state addresses.

terraformmodulesstate

Why we selected itThe definition prefers state list/show-json over treating raw state format as a stable interface and distinguishes moved blocks from manual state migration.

Weigh upState contains plaintext sensitive values even in JSON output. Terraform tests defaulting to apply can create real infrastructure; address migration requires a reviewed plan and recoverable state.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMPL-2.0

Infrastructure · hashicorp

Terraform Policy Authoring and Tests

Author or migrate policy HCL with version-specific mock schemas and targeted pass/fail policy tests.

infrastructurehashicorp

Why we selected itIt routes Terraform Policy authoring and tests by the installed tfpolicy minor line, 0.2.x or 0.3.x.

Weigh upThe skill does not write .tftest.hcl or general Terraform configuration.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMPL-2.0

Runs one Terraform provider TestAcc against a test account, with an explicit timeout and a flip check for a false pass.

terraformacceptance-testsgo

Why we selected itThe procedure is bounded: confirm the credentials point at a test account, run one TestAcc with TF_ACC=1 and a timeout above Go’s 10-minute default, and discover extra variables from PreCheck and os.Getenv rather than guessing.

Weigh upAcceptance tests create real infrastructure and can cost money. The skill says so, and sweepers are left to a separate provider-test-patterns skill if that skill is installed.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMPL-2.0

Infrastructure · hashicorp

Terraform Provider Documentation

Generate Registry documentation from implemented schema descriptions, focused templates and reusable HCL examples.

infrastructurehashicorp

Why we selected itIt generates Terraform Registry docs from schema descriptions and tfplugindocs templates, and it forbids pages for objects the provider does not implement.

Weigh upThe skill does not publish the provider; it only prepares docs and names the tag and manifest checks.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMPL-2.0

Infrastructure development · HashiCorp

Terraform Provider Resources

Implements Plugin Framework resources and data sources, including drift, import and the acceptance tests a submission needs.

terraformprovidergo

Why we selected itThe scope names CRUD, plan modifiers, validators, waiters, import and acceptance-test coverage as one workflow.

Weigh upMPL-2.0 is file-level copyleft for modified redistributions of this file.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMPL-2.0

Infrastructure · hashicorp

Terraform Provider Scaffold

Scaffold a Plugin Framework provider with configuration schema, unknown-value guards and environment credential fallback.

infrastructurehashicorp

Why we selected itIt scaffolds a Plugin Framework provider only after confirming a new terraform-provider-* workspace.

Weigh upThe skill stops if the user does not want a new workspace.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMPL-2.0

Infrastructure development · HashiCorp

Terraform search and import

Runs Terraform Search only after the provider is shown to support that resource type, and switches to a manual import when it does not.

terraformimportdiscovery

Why we selected itThe skill says to verify list-resource support before writing a query, and to stop the search path when the type is unsupported.

Weigh upSearch needs an initialized working directory and a provider that implements list for that type.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewMPL-2.0

Infrastructure development · HashiCorp

Terraform Stacks

Covers Terraform Stacks component and deployment files, including the required Terraform version file.

terraformstackshcl

Why we selected itThe definition names .tfcomponent.hcl and .tfdeploy.hcl as distinct files.

Weigh upMPL-2.0 applies to modified redistributions of this file.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMPL-2.0

Development workflow · HashiCorp

Terraform style guide

Applies HashiCorp's own HCL conventions for file order, naming, variables and outputs.

terraformstylehcl

Why we selected itThe headings cover file organisation, formatting, naming, variables and outputs as separate rules.

Weigh upStyle agreement is not a plan or an apply.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewMPL-2.0

Development workflow · GitHub community (awesome-copilot)

Test Gap Audit

Map risky behavior to existing tests and report confirmed gaps separately from heuristic leads.

development-workflowgithub

Why we selected itIt audits missing or weak tests and stays read-only unless the user asks for new tests.

Weigh upThe skill does not install dependencies or start services unless the user asks.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Agent development · Trail of Bits

Testing Handbook Skill Generator

Maintainer workflow that reads sections of the Trail of Bits Testing Handbook (appsec.guide) and writes or refreshes the matching Claude Code skills inside the trailofbits/skills testing-handbook-skills plugin.

skillsagent-developmentfuzzing

Why we selected itIt maps handbook sections to four skill types (tool, fuzzer, technique and domain), each with its own template under templates/.

Weigh upIt maintains the trailofbits/skills repository itself: it writes under plugins/testing-handbook-skills/skills/ and edits the repository root README.md, so it needs a checkout of that repository.

Source reviewed

8 sources · 2026-10-22

Read the full review Upstream setup linkedCC-BY-SA-4.0

AI application development · Together AI

Together AI

Routes Together AI work to domain guides, with rules against retrying unrecoverable errors and leaving billable resources running.

together-aiinferencefine-tuning

Why we selected itDedicated endpoints, containers and clusters bill while they run; the file says to create them only when asked, delete what was created and confirm it is gone.

Weigh upModel IDs in the guides are examples checked 2026-10-07, per the file. The frontmatter still lists embeddings and reranking, while domains/embeddings.md says Together does not currently offer them.

Source reviewed

4 sources · 2026-10-13

Read the full review Install command in reviewMIT

Security · trailofbits

Token Integration Review

Review token implementations and protocol integrations for non-standard transfer and approval behavior.

securitytrailofbits

Why we selected itThe checklist separates missing return values, transfer fees, rebasing, hooks and owner-controlled changes, and asks whether the protocol handles those behaviors rather than relying on token reputation.

Weigh upNamed-token examples and sample reports are illustrative and may be stale or internally inconsistent. Safe wrappers do not resolve rebasing, upgradeability or accounting assumptions automatically.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

TON contract audit · Trail of Bits

TON vulnerability scanner

Pins the TON scanner's refusal to omit a coverage row, and its requirement to cite the stored address a Jetton sender is compared against.

tonfuncjetton

Why we selected itAn empty verdict or a missing pattern row is an incomplete scan, and not having looked is not an n/a.

Weigh upThe resource was not executed.

Source reviewed

4 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Security · Trail of Bits

Trail of Bits Agentic Actions Auditor

Traces external issue or pull-request input into AI-enabled GitHub Actions and reports the permissions that could amplify an injection path.

github-actionsprompt-injectionci-security

Why we selected itChecks environment intermediaries and reusable-workflow inputs as well as prompt text.

Weigh upStatic guidance only: it does not execute an attack, fix workflows or certify that CI is secure.

Source reviewed

5 sources · 2026-10-14

Read the full review Upstream setup linkedCC-BY-SA-4.0

Security · Trail of Bits

Trail of Bits Audit Context Building

Builds a cited dossier of function invariants, assumptions and dependencies before a separate vulnerability-hunting pass.

auditcode-understandinginvariants

Why we selected itFollows callees and unsuccessful paths instead of trusting names such as validate or authorize.

Weigh upThe skill is a router into a plugin workflow and named function analyzer; a copied standalone SKILL.md is insufficient. No download is offered for that reason: workflows/audit-context.js and agents/function-analyzer.md sit outside the skill directory.

Source reviewed

7 sources · 2026-10-13

Read the full review Upstream setup linkedCC-BY-SA-4.0

Reviews security-relevant changes against their baseline, removed protections, callers and concrete attacker scenarios.

diff-reviewgit-historysecurity

Why we selected itUses git history to ask why a removed check existed and whether a previous fix is being undone.

Weigh upThe methodology includes checking out the baseline and head; use a suitable worktree so the review does not disrupt uncommitted work.

Source reviewed

6 sources · 2026-10-13

Read the full review Attributed definition availableCC-BY-SA-4.0

Code correctness · Trail of Bits

Trail of Bits Dimensional Analysis

Coordinates unit discovery, source annotations and arithmetic review, with a manifest recording which files were validated or blocked.

unitsfixed-pointarithmetic

Why we selected itTracks semantic units separately from decimal scaling across arithmetic and function boundaries.

Weigh upThe workflow edits source comments and writes root-level manifests; it is not a read-only spot check.

Source reviewed

7 sources · 2026-10-14

Read the full review Upstream setup linkedCC-BY-SA-4.0

Languages and tooling · Trail of Bits

Trail of Bits Modern C++

Pushes new C++ toward C++20, 23, and 26 idioms and a hardened libc++, and stops when the user requires an older standard or the file is C.

cppmodern-cppmemory-safety

Why we selected itWhen NOT to Use respects an explicit older-standard constraint and excludes pure C, build-system questions, and projects that are not primarily C++.

Weigh upAn embedded or legacy ABI that must stay on an older standard is a stated reason not to apply the skill.

Source reviewed

2 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Testing and reliability · Trail of Bits

Trail of Bits Mutation Analysis

Interprets mutation campaign results and turns survivors into qualified test-gap reports or separately reproduced bug findings.

mutation-testingtest-gapsmewt

Why we selected itSeparates test-gap analysis from bug hunting, so an uncaught mutation is not automatically presented as a vulnerability.

Weigh upThis selection focuses on existing-report analysis. Setup and optimization need separate review: examples hardcode src/, retain a pre-optimization acceptance check and mix counting formats. Do not copy their campaign estimates or purge steps unchanged.

Source reviewed

15 sources · 2026-10-13

Read the full review Install command in reviewCC-BY-SA-4.0

Testing and reliability · Trail of Bits

Trail of Bits Property-Based Testing

Chooses meaningful invariants and generators for tests that explore an input domain, while rejecting tautological or vacuous assertions.

property-testinghypothesisinvariants

Why we selected itExplains how tests can pass while restating implementation or filtering away nearly every input.

Weigh upProperty selection still needs a valid specification; an incorrect invariant can generate misleading failures.

Source reviewed

6 sources · 2026-10-13

Read the full review Attributed definition availableCC-BY-SA-4.0

Security · Trail of Bits

Trail of Bits Sharp Edges

Examines APIs and configuration for cases where ordinary developer mistakes can disable a security boundary.

api-designsecure-defaultsconfiguration

Why we selected itLooks at zero, empty, negative and mixed-up values rather than only intended happy-path usage.

Weigh upThis is a design-review method, not a general implementation bug scanner.

Source reviewed

5 sources · 2026-10-13

Read the full review Attributed definition availableCC-BY-SA-4.0

Security review · Trail of Bits

Trailmark

Parses a codebase into a queryable graph of functions, classes and calls and runs pre-analysis passes for blast radius, taint, privilege boundaries and entry points, so later trailmark skills can ask structural questions instead of reading files.

code-understandingcall-pathsblast-radius

Why we selected itBuilds one queryable source/binary graph and runs preanalysis() for blast radius, entry points, privilege boundaries and taint, storing the results as annotations and named subgraphs.

Weigh upIt is static analysis; the skill lists runtime behaviour analysis under When NOT to Use.

Source reviewed

5 sources · 2026-10-22

Read the full review Upstream setup linkedCC-BY-SA-4.0

Security review · Trail of Bits

Trailmark Finding Triage

Triages a single finding, SARIF result, weAudit annotation, suspicious function or report excerpt with Trailmark reachability, entrypoint-path, taint, privilege-boundary, blast-radius and caller/callee evidence, returning one of four verdicts: Promote, Needs manual review, Deprioritize or Blocked.

triagesarifblast-radius

Why we selected itTriages one finding with reachability, entrypoint paths, taint, privilege-boundary, blast-radius and caller/callee evidence from the graph.

Weigh upTrailmark Finding Triage's SKILL.md declares allowed-tools that include Write; it writes an evidence packet, and evidence_packet.py can write JSON with --out.

Source reviewed

6 sources · 2026-10-22

Read the full review Upstream setup linkedCC-BY-SA-4.0

Security review · Trail of Bits

Trailmark Review Gate

Runs a Trailmark structural gate over a branch, pull request, fix commit, release diff or ref range and returns PASS, WARN, FAIL or UNKNOWN for new entrypoints, lost validation, privilege-boundary drift, blast-radius growth and newly reachable sinks.

diff-reviewpull-requestblast-radius

Why we selected itApplies PASS/WARN/FAIL/UNKNOWN gate rules over a branch, PR, fix commit, release diff or ref range, running engine.preanalysis() on both snapshots.

Weigh upTrailmark Review Gate's SKILL.md declares allowed-tools that include Write; it writes a gate packet (references/output-format.md) for the branch reviewer.

Source reviewed

5 sources · 2026-10-22

Read the full review Upstream setup linkedCC-BY-SA-4.0

Security review · Trail of Bits

Trailmark Structural Analysis

Builds a Trailmark graph and runs engine.preanalysis() to report complexity hotspots, taint, blast radius, privilege boundaries and attack surface, with extra 0.4+/0.5+ fields when that Trailmark is installed.

blast-radiusgraphsecurity-audit

Why we selected itBuilds a graph and runs engine.preanalysis() for hotspots, taint, blast radius, privilege boundaries and attack surface, enriched by 0.4+/0.5+ data when present.

Weigh upTrailmark Structural Analysis's SKILL.md declares allowed-tools Bash, Read, Grep and Glob, with no Write.

Source reviewed

5 sources · 2026-10-22

Read the full review Upstream setup linkedCC-BY-SA-4.0

Security review · Trail of Bits

Trailmark Summary

Runs a Trailmark summary over a codebase and returns the auto-detected languages, an entry-point count and a dependency list as a quick structural overview.

code-understandinggraphsecurity-audit

Why we selected itReturns auto-detected languages, an entry-point count and a dependency list, and reports a gap rather than fabricate any of the three.

Weigh upTrailmark Summary's SKILL.md declares allowed-tools Bash, Read, Grep and Glob, with no Write.

Source reviewed

5 sources · 2026-10-22

Read the full review Upstream setup linkedCC-BY-SA-4.0

Security review · Trail of Bits

Trailmark Variant Neighborhood

Expands one confirmed or suspected vulnerability into a Trailmark graph neighborhood of variant candidates — siblings, shared callers and callees, common sinks, common entrypoint paths, interface implementations and overrides — then ranks them for follow-up.

variant-analysistriagegraph

Why we selected itExpands one confirmed issue into a graph neighborhood of variants: shared callers and callees, sinks, entrypoint-path neighbours, interface and override siblings, and taint or boundary peers.

Weigh upTrailmark Variant Neighborhood's SKILL.md declares allowed-tools that include Write; it writes a handoff packet (references/output-format.md).

Source reviewed

5 sources · 2026-10-22

Read the full review Upstream setup linkedCC-BY-SA-4.0

Machine learning · huggingface

TRL Training API

Select TRL trainers and dataset shapes, then configure SFT or online preference/reward training in Python or the CLI.

machine-learninghuggingface

Why we selected itIt maps each stable TRL trainer to a dataset shape and says the chat template is applied by the library.

Weigh upThe skill does not launch a training job.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewApache-2.0

Data engineering · dbt-labs

Troubleshoot dbt Platform Jobs

Diagnose failed platform runs from scoped run logs, node artifacts and source history.

data-engineeringdbt-labs

Why we selected itIt diagnoses dbt platform job failures from run logs and the Admin API, and it forbids changing a test just to make it pass.

Weigh upThe skill is not for local dbt development errors.

Source reviewed

4 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Languages and tooling · JetBrains

Turborepo

Monorepo build-system discipline for Turborepo: package tasks instead of root tasks, turbo.json task registration, cache outputs and hashing inputs, environment variables as cache keys, --affected runs, CI setup, and tag-based package boundaries.

turborepomonorepoturbo-json

Why we selected itLeads with the mistake every Turborepo setup makes — task logic in the root package.json — and shows the correct package-scripts plus turbo.json registration as wrong/right code pairs, including why root tasks defeat parallelization.

Weigh upPinned to the vendored snapshot of Turborepo 2.8.18-canary.7 (synced 2026-03-16); Turborepo releases quickly and flag or boundaries behaviour may drift from newer versions.

Source reviewed

10 sources · 2026-10-13

Read the full review Install command in reviewMIT (skill-local, Vercel copyright)

Databases · Turso

Turso Database

Turso's skill for its in-process SQLite-compatible engine: hard constraints first, then decision trees to feature and SDK reference files, plus CLI, SQL and MCP quick references.

sqlitesqlvector-search

Why we selected itStates ten constraints before any code is written: BETA status, single-process file access, no WITHOUT ROWID tables, no VACUUM, UTF-8 only, WAL as the default journal mode with the legacy delete, truncate and persist modes unsupported, full-text search behind a compile-time feature, encryption behind a flag, experimental MVCC, and lower-is-closer vector distance.

Weigh upThe skill directory skills/turso-db last changed in a commit dated 2026-03-25 (git log at the pin); the pinned revision 34ced52f is dated 2026-07-22.

Source reviewed

4 sources · 2026-10-21

Read the full review Upstream setup linkedMIT

Evidence and evaluation · Undominated.ai

Undominated · 2-bit activation outlier audit

Audits 2-bit weight quantization (QuIP#, AQLM) activation outlier preservation and SQNR thresholds.

serving2-bitquantization

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates measured SQNR against acceptable signal limits; does not compute matrix inversions.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · A dev POST keeps its slash

Refuse a dev POST path with no trailing slash. trailingSlash always 404s that path in dev. Production is a different path shape.

trailing-slashdevpost

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upA production path without a slash can pass. This checker sends no request.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · A load return is not the catalogue

Refuse a load() return labelled catalogue. The catalogue is imported by the page. Putting it in the load return inlines it into every prerendered page.

pagescatalogueprerender

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upThe label is caller-supplied. A pass does not read a page.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · A marginal block is not priced

Refuse a marginal-block bill. A whole request uses the rate of the rung that contains its input length. This checker does not price either mode.

pricingcontextbilling

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upNo amount is read. A whole-request label can pass, and the checker still does not price it.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · A missing announcement is not safe

Refuse a safety claim built from none-announced, unknown, or unrated. Those labels do not establish that a subject is safe.

safetylabelsclaims

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upThe result never sets established to true. A false or null claim stays unestablished. No source is fetched.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · A models list is not a chat call

Refuse a claim that chat works because /v1/models returned the id. A list is not a completion. This checker sends no request.

modelschatnim

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upThe result never says chat was established. Another evidence label can pass and does not prove a completion.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · A prompt is not the catalogue

Refuse a prompt whose scope is the whole catalogue. Retrieval stays top-k. No catalogue is copied.

promptcatalogueretrieval

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upA top-k scope can pass. This checker does not count rows or read the catalogue.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · A query is not a passage

Refuse an embedding call whose input_type does not match the role. A query and a passage are different requests. No vector is computed.

embeddingsquerypassage

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upMatching query or matching passage can pass. No text is embedded.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · A rollback is not the baseline

Refuse to record a rolled-back or unproven deploy as the accepted baseline. The baseline advances only after a deploy has proven good.

deploybaselinerollback

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upThis checker does not contact a host. A pass on a supplied proven outcome does not prove the deploy.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · A token is not PUBLIC_

Refuse a provider token in a PUBLIC_ variable or in browser code. A server environment is outside this refusal and is not approved by a pass.

secretsbrowserpublic-prefix

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upA server environment can pass. A pass does not approve the file the token lives in.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · A zero base is not defined

Refuse a multiple whose base is zero. The multiple is not defined. This checker does not emit a quotient.

pricingmultiplebase

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upA positive base can pass, and the result still does not define the multiple. Amounts are not a field.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Activation Recomputation Tax Audit

Audits selective vs full gradient checkpointing memory savings vs compute FLOP taxes for long-context prefill.

checkpointingrecomputationvram

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upCalculates theoretical FLOP addition based on attention and MLP re-forwarding.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Adversarial suffix detection audit

Checks a supplied perplexity-ratio assertion and anomaly threshold over local numeric input; a flag is a heuristic signal for review, not proof of an attack.

evaluationsecurityjailbreak

Why we selected itPairs a bounded numeric audit with an offline Python checker and an explicitly synthetic threshold-breach example.

Weigh upDoes not measure perplexity, inspect prompts, execute a target model or establish detector accuracy. Input provenance and threshold suitability remain unverified; attack_detected means only that the unrounded ratio exceeded the supplied threshold.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Advertised-route audit

Refuse a crawl-list URL whose English-only path gained a locale prefix, or whose ?format=md sibling was never written.

seocrawllocale

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upChecks a supplied URL list against a supplied relative file list. It does not read a build directory, fetch URLs, or treat a present file as correct content. Locale codes are two letters and English-only paths must be listed exactly.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Agent consensus cost audit

Audits token costs and accuracy ROI of multi-agent consensus, majority voting, and debate protocols.

pricingconsensusmajority-voting

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates accuracy deltas against cost multipliers; does not conduct voting rounds.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Agent trajectory cost audit

Audit multi-turn agent execution loops for quadratic token accumulation, retry loops, tool call overhead, and cumulative trajectory dollar costs.

agentstrajectorycost

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upCalculates costs from supplied turn token counts; does not execute multi-step agent actions.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · AI resource intake audit

Review a skill, agent or MCP server for pinned source identity, licensing, requested permissions and observable verification before adding it to a trusted collection.

skillsagentsmcp

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic local checks over supplied evidence; not a guarantee of source truth or production suitability.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · Air-Gapped Inference Economics Audit

Calculates on-premise hardware CapEx amortization, rack power, and hosting TCO per 1M tokens to verify self-hosting economic viability.

air-gappedsovereigntco

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic financial amortization model. Requires explicit electricity and CapEx inputs.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Alias resolution

Resolve an explicitly supplied alias to one distinct concrete target, refusing unresolved or conflicting candidate evidence. Use before treating a latest-style pointer as a concrete model ID.

identityaliascatalogue

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upChecks one caller-supplied candidate set, refusing unresolved, conflicting and self-targets. It does not verify target existence, candidate completeness, recency or recursively resolve aliases.

Source reviewed

3 sources · 2026-10-07

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · An add_header keeps the snippet

Refuse an nginx location that sets add_header without the shared header snippet. add_header in a location discards inherited headers.

nginxheaderssnippet

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upA location that sets no header can pass. This checker does not read an nginx file.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Apple unified memory bandwidth audit

Audits Apple Silicon unified memory bandwidth utilization and token generation throughput efficiency.

servingapple-siliconmetal

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upCalculates theoretical memory bandwidth utilization; does not profile Metal shaders directly.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · arena.ai is not scraped

Refuse an arena.ai scrape. The Hugging Face dataset pin can pass. Any other origin can pass and is not approved.

arenalicencehuggingface

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upA Hugging Face dataset label can pass. This checker does not fetch either origin.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Artificial Analysis is not republished

Refuse redistribution of Artificial Analysis. A pass for any other source does not approve that source. No series is copied.

licensingredistributionartificial-analysis

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upLeaving it unreprinted can pass. Another source can pass and is not approved.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · ASCII art jailbreak evasion audit

Audits ASCII art spatial obfuscation and safety alignment bypass resistance in prompt inputs.

evaluationsecurityascii-art

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates observed character dispersion and safety classifier scores; does not modify prompt tokens.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Audio streaming chunk latency audit

Audits continuous audio streaming inference chunk sizes and Real-Time Factor (RTF) latency bounds.

servingaudiostreaming

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upCalculates RTF ratios based on chunk durations; does not process PCM audio streams.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Auto Prompt Distillation Audit

Calculates prompt instruction compression efficiency and downstream task execution fidelity.

distillationprompt-compactionefficiency

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates token ratio and automated verification test pass rate.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Batch discount audit

Audit batch API SLA windows, advertised vs realized discounts, queue starvation risk, and cancellation policies.

batchpricingdiscounts

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upChecks pricing rate cards and declared turnaround windows; does not submit live asynchronous batch jobs.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Batch Queue Starvation Risk Audit

Calculates batch queue processing delays and verifies turnaround compliance against provider SLA deadlines.

batch-apiqueue-starvationsla

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates queue entry to completion timestamps.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Benchmark cohort audit

Audit a benchmark comparison for matched model identity, missing coverage, restricted populations and misleading confidence-interval interpretations.

benchmarksstatisticscoverage

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic local checks over supplied evidence; not a guarantee of source truth or production suitability.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · Bidirectional speculative audit

Audits bidirectional speculative drafting hit rates during structured JSON and schema generation.

servingstructured-outputsspeculative

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upCalculates grammar hit ratios; does not compile context-free grammars directly.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · BitNet ternary weight audit

Audits BitNet b1.58 ternary weight representation and addition-only kernel energy reduction factors.

servingbitnetternary

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates observed energy reduction factors; does not synthesize CUDA assembly directly.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Browser code does not name loopback

Refuse a browser script that names 127.0.0.1. That address is the visitor's machine. A server-side sidecar may name it.

browserloopbacksidecar

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upServer-side code may name loopback. This checker does not fetch the page.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Burst Concurrency & Queue Degradation Audit

Evaluates queue buffering latency and HTTP 429/503 drop rates under sudden burst step-load profiles.

concurrencyload-testingsla

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upLoad profiles must simulate realistic client traffic patterns with appropriate timeout limits.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Cache multiplier

Keep a cache figure that is a multiplier distinct from an absolute rate.

pricingcacheunits

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDoes not convert a multiplier into a vendor rate.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Cache-tier billing audit

Bill a request against the cache tier that applies to it, so a cached prefix is not charged at the base input rate.

pricingcachingbilling

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic local arithmetic for one request against one supplied rate set; it does not walk a context-tier ladder or apply batch, off-peak or subscription billing rules.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Canary Deployment Drift Audit

Calculates statistical output distribution divergence (PSI/KS-test) between stable baseline and canary model deployments.

canarydistribution-shiftpsi

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates Population Stability Index (PSI) metrics across token length and embedding features.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Carbon intensity scheduling audit

Audits grid carbon intensity temporal variations to schedule batch inference during green energy peaks.

pricingcarbongreen-energy

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upCalculates carbon savings from historical grid carbon intensity metrics; does not control job schedulers.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Chunked Prefill Scheduler Audit

Calibrates chunked prefill chunk sizes to eliminate inter-token latency (ITL) spikes while preserving prefill throughput.

chunked-prefillitl-jittervllm

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates observed ITL jitter and prefill chunk size constraints.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · CI/CD Regression Gate Audit

Automated deployment gate: enforces strict latency, output schema validity, and accuracy thresholds in CI/CD pipelines.

ci-cdregression-gateautomated-deploy

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic pass/fail gate over test evaluation suite outputs.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Concurrency cliff audit

Audit inference throughput degradation, queue build-up, and throughput cliff saturation as concurrency scales.

concurrencythroughputsaturation

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upAnalyzes benchmarked throughput matrix; does not launch concurrent load testing clusters.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Context compaction audit

Audit conversational context compaction, summarization, and message pruning for named entity retention, critical fact loss, and semantic distortion.

contextcompactionsummarization

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upPerforms exact string and entity verification; does not generate text summaries.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Context-rung selection

Select the context-tier rung that contains a request's input length, so the dearest later rung is not applied to a shorter prompt.

pricingcontexttiers

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upSelects one whole-request rung from a supplied ladder. It does not model marginal block pricing or fetch vendor rates.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Context-tier ladder audit

Check a past-this-length price multiple against every rung, so the dearest rung is not applied at the first boundary.

pricingcontexttiers

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic local check of a supplied ladder; it does not fetch vendor prices or model marginal block billing.

Source reviewed

3 sources · 2026-10-07

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · Continuous Batching Scheduler Audit

Audits iteration-level scheduling vs static request batching to verify throughput gains and elimination of tail-latency bubble waste.

continuous-batchingschedulerorca

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEmpirical model based on measured throughput and sequence length variance.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Cross modal KV allocation audit

Audits visual token versus text token KV cache allocation ratios in long multimodal conversations.

servingmultimodalkv-cache

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upVerifies total VRAM footprint across modality caches; does not compute attention weights.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Currency isolation audit

Refuse to add, average or rank amounts whose currencies differ, so a missing code is not treated as USD and no exchange rate is invented.

pricingcurrencytotals

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic isolation of supplied amounts. It does not convert currencies or fetch an exchange rate.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · DARE random drop rescale audit

Audits DARE (Drop And REscale) random drop rate and benchmark capability retention across merged models.

evaluationdaremodel-merging

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates reported benchmark retention rates under selected drop probabilities; does not perform weight arithmetic.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Datacenter power envelope audit

Audits datacenter rack power density, cooling overhead, and PUE compliance under 100% GPU duty cycles.

pricingdatacenterpower

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upCalculates total power draw from IT load and PUE coefficients; does not measure physical current.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Decimal determinism audit

Prove a published rate, a computed bill and a displayed figure agree to the last digit, so no displayed decimal comes from IEEE-754.

decimalprecisionpricing

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic decimal recomputation over supplied rates, quantity and billed values; it does not fetch vendor prices or judge display formatting.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Delivery class

Refuse a dominance claim that treats batch delivery, or a model's own other delivery, as a like-for-like beater of synchronous delivery.

deliverybatchdominance

Why we selected itIt refuses a dominance claim that treats batch delivery, or the same weights on another delivery, as a like-for-like beater of synchronous delivery.

Weigh upThe checker does not read prices, scores, or context.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · DGX SuperPOD TCO Audit

Calculates enterprise GPU cluster Total Cost of Ownership including InfiniBand fabrics, power usage effectiveness (PUE), and rack hosting.

superpodtcodgx

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic financial amortization and utility model.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Do not scale the invoice unit

Refuse a scale factor or a restated invoice unit. A per-thousand row is not turned into a per-million row by multiplying. Amounts are not read.

pricingunitsinvoice

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upA factor is not applied and is not copied out. Amounts are not a field of this checker.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Dominance wording audit

Classify a two-model comparison before publishing it, so a tie or a dropped requirement is not called both better and cheaper.

dominancewordingcapabilities

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic local checks over one supplied pair and the listed requirements; not a guarantee of source truth.

Source reviewed

3 sources · 2026-10-07

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · Dynamic Spot Rate Arbitrage Audit

Calculates real-time price arbitrage across multi-provider spot inference endpoints under strict latency SLOs.

spot-arbitrageroutingpricing

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates observed rate cards and achieved dollar savings.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Dynamic Tool Filtering Tax Audit

Calculates token savings achieved by dynamically retrieving relevant tools into context vs sending static tool catalogs.

tool-filterprompt-tokensdynamic-tools

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates observed token reduction ratios.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Eagle-2 dynamic draft tree audit

Audits Eagle-2 dynamic draft tree pruning and net token throughput multipliers.

servingeagle-2speculative-tree

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates observed tree acceptance statistics; does not execute tree verification kernels.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Elo drift audit

Audit Bradley-Terry win-rate confidence interval bounds, active pair sampling bias, and Elo deflation/inflation drift across benchmark dataset versions.

elobenchmarksstatistics

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upComputes statistical significance over supplied benchmark rating snapshots; does not compute Elo from raw battle votes.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Embodied silicon carbon TCO audit

Audits embodied carbon in GPU manufacturing (Scope 3) amortized per 1 billion inference tokens.

pricingembodied-carbonscope3

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upCalculates amortized Scope 3 grams per 1M tokens based on device lifetime and volume; does not audit fabs.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Equal is not better

Refuse to call an equal score or an equal price an improvement.

comparisonstieswording

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDoes not rank a catalogue.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · EU AI Act Transparency Audit

Checks four supplier declarations in a local JSON file: training compute, copyright summary, energy log, and model card. A true declaration is not an EU AI Act determination.

eu-ai-acttransparencycompliance

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upA pass means each supplied declaration is true. It is not a legal determination and it does not inspect a model card, a paper, or a regulator's file.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Eval contamination and judge audit

Audit benchmark evaluation sets and prompts for training set contamination, n-gram overlap, verbatim memorization, and judge prompt skew.

contaminationbenchmarksngrams

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic text overlap and statistical correlation over supplied samples; does not crawl external web corpora.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Evidence audit

Audit a numeric claim against locally saved source evidence, including denominator and reproducible calculations, before citing or publishing it.

evidenceprovenanceclaims

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic local checks over supplied evidence; not a guarantee of source truth or production suitability.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · False-Positive Refusal Rate Audit

Quantifies false-positive refusal rates on benign software engineering, cybersecurity, and clinical research queries.

safetyrefusalsover-refusal

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upTest datasets must exclude genuinely dangerous dual-use materials to measure true false positives.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Family snapshot

Require family membership to be a subset of the catalogue snapshot.

familiessnapshotsidentity

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDoes not infer family membership from names.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Few-Shot Example Pruning Audit

Calculates the marginal accuracy contribution per few-shot exemplar to eliminate low-ROI prompt tokens.

few-shotexemplarspruning

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates observed benchmark accuracy across example subset ablations.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Filter invariance

Check that filtering preserves the relative order of retained IDs. Use for search or facet behavior that is intended to hide rows without re-sorting them.

rankingfilterspopulation

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upChecks one supplied ordered population and subsequence, including empty arrays. It does not check scores, numeric ranks, frontier membership or dominance labels.

Source reviewed

3 sources · 2026-10-07

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · Final-host audit

Refuse a vendor quote when the final host you recorded is not the host you requested. A redirect is not that vendor. www is not stripped, and this checker sends no request.

provenancepricingaudit

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upCompares the two host strings you supply. It does not send a request, read a body, or strip www. A pass is not a rate card.

Source reviewed

3 sources · 2026-10-07

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · Final-path audit

Refuse a page claim when the final path you recorded is not the path you requested. A redirect to the homepage is not that page. The host is a different checker, and this one sends no request.

provenancepricingaudit

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upChecks recorded path strings in a narrow ASCII subset. It rejects whitespace, doubled leading slashes, URLs, queries, fragments and dot segments; it does not fetch, compare hosts, decode percent escapes, or verify page content.

Source reviewed

3 sources · 2026-10-07

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · FP4 E2M1 microscaling audit

Audits OCP Microscaling Formats (MXFP4 / E2M1) block scaling factors and perplexity degradation delta.

servingfp4e2m1

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upVerifies reported perplexity differentials on standard evaluation benchmarks; does not perform forward passes.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · FP4 Quantization Noise Audit

Calculates Signal-to-Quantization-Noise Ratio (SQNR) and verifies perplexity degradation across FP4 / NVFP4 tensor formats.

fp4nvfp4sqnr

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates observed SQNR and benchmark scores. Does not calibrate activation clipping.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Generated artefacts stay uncommitted

Refuse a commit path the caller has classified as generated. Source can pass. A pass does not prove the path is source.

gitgeneratedartefacts

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upA source label can pass and does not prove the path. This checker does not list the tree.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · GPU Underutilization Cost Audit

Calculates financial penalties of reserved GPU instances operating under low continuous utilization factors.

underutilizationidle-powerreserved-instances

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates utilization duty cycles and contractual hourly rates.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Graph RAG Extraction Tax Audit

Calculates entity-relationship extraction token overhead and graph construction costs compared to vector retrieval.

graph-ragknowledge-graphextraction-tax

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates prompt token inflation for entity and relationship triples extraction.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Guardrail Latency Overhead Audit

Measures the TTFT penalty and token inflation imposed by input and output safety moderation guardrail layers.

guardrailssafetyttft

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upMeasures synchronous guardrail configurations; asynchronous streaming guardrails require separate event analysis.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · H2O heavy hitter eviction audit

Audits Heavy Hitter Oracle (H2O) KV cache eviction and critical prompt token retention under memory limits.

servingh2okv-cache

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates observed accuracy retention under selected token budgets; does not generate prompt responses.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Header-not-row audit

Refuse a unit claim that borrows the table's display header when that row does not state its own unit. A per-million header is not every row's invoice unit. Amounts are not read.

pricingmeteraudit

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upThe header is ignored. rowUnit has to come from that row's own sentence, or be null. It does not read amounts or convert units. A pass is not a vendor quote.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · Hierarchical Memory Consolidation Audit

Audits episodic-to-semantic memory compaction token costs in long-horizon autonomous agents.

agent-memoryconsolidationepisodic-memory

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates compression ratio and preservation flags.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Hint versus handler

Refuse a read-only hint when the handler notes you supply include a write, or when those notes are empty. A hint is not a review of the handler.

mcppermissionsread-only

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upChecks supplied operation names against a true read-only hint. It does not read tool names, open a handler, or start a server. A pass is not proof the server is read-only.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · HTTP/2 Multiplexing & Head-of-Line Audit

Measures TTFT variance and head-of-line blocking under HTTP/2 multiplexing vs HTTP/1.1 connection pools under high stream concurrency.

http2multiplexingttft

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upHTTP/2 stream prioritization may vary depending on intermediary reverse proxy configurations.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Hybrid transformer SSM memory audit

Audits memory footprint and KV cache allocation in hybrid models combining attention and Mamba layers.

servinghybridjamba

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upCalculates layer-wise memory allocations; assumes linear recurrent state sizes.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Identity dedup audit

Collapse duplicate identities inside one population before ranking, so one model or one seller cannot be counted twice.

deduplicationidentitydenominators

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic local check over supplied rows and caller-normalised identities; it does not resolve upstream aliases or verify provenance.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Indicative-rate audit

Refuse a global-rate claim when the excerpt says the figure is indicative, varies by country, starts at a floor, or is a list price beside a credit. One sentence is not every region's rate.

pricingregionaudit

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upLooks for qualifying wording in the excerpt you supply. Negation is not parsed. It does not read amounts or pick a country. A pass is not a vendor quote.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · Indirect Prompt Injection Detector

Audits model vulnerability to untrusted prompt injections embedded in retrieved RAG contexts and web documents.

indirect-injectionrag-securitydata-poisoning

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates detection and mitigation rates across indirect injection test corpora.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Install receipt

Compare an expected file manifest with a sha256sum receipt and, when asked, the bytes in a directory next to the input.

installintegritysha256

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upHashes only the files named inside the input directory. A pass is not evidence that npm or the Skills CLI performed the install.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · IP Infringement Verbatim Memorization Audit

Measures verbatim memorization rates of copyrighted code and text corpora in model completion outputs.

copyrightmemorizationip-risk

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates 50-token exact match n-gram overlap against protected corpora.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · JSON Schema Constrained Decoding Tax Audit

Measures token generation throughput penalty and grammar compilation overhead incurred by FSM-constrained decoding engines.

structured-outputjson-schemaconstrained-decoding

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates measured TPS and initial compilation time. Schema-dependent regex parsing overhead.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Judge Prompt Variance Audit

Measures LLM-as-a-judge score variance, position bias, and length preference bias across automated evaluation runs.

llm-judgeposition-biaseval-variance

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates observed pairwise swaps and position agreement scores.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Keep every ladder rung

Refuse a published context-tier ladder whose rung count differs from the source. Dropping a rung flattens the ladder. Adding a rung invents one. Prices and boundaries are not read.

pricingcontextladder

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upPrices and boundaries are not read. A matching count does not say the boundaries are right.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · KV Cache INT4 Outlier Drift Audit

Audits key/value cache INT4 quantization channel scaling to prevent catastrophic attention drift on long sequences.

int4kv-cacheoutliers

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upAudits maximum absolute error across attention logit distributions.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · KV Cache Memory Sizing Audit

Audits KV cache memory footprint per token and max concurrent sequence limits under FP16, FP8, and INT4 KV caches to prevent serving OOMs.

kv-cachevramhardware

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upHardware memory models assume standard Multi-Head Attention (MHA) or Grouped-Query Attention (GQA). Verify key-value head counts explicitly.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · KV Transfer RDMA Bandwidth Audit

Calculates RDMA network saturation and link concurrency during high-throughput KV cache migration.

rdmakv-cachenetwork

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates theoretical link saturation based on concurrent migration sessions.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Length bias audit

Audit length bias (verbosity bias) in LLM-as-a-judge evaluations and preference datasets, detecting artificial win-rate inflation.

evaluationsbiasllm-judge

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic analysis of battle length deltas and outcomes; does not re-score answers independently.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Length bias removal audit

Audits reward model length bias and synthetic response verbosity inflation in preference datasets.

evaluationdporlhf

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upCalculates Pearson/Spearman correlation between length and reward; does not train reward models.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Licence-boundary audit

Check a redistribution claim against the licence evidence you recorded. A public page is not a licence.

licensingredistributionattribution

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic consistency check of a filled licence note; it does not interpret licence text or fetch the licence.

Source reviewed

3 sources · 2026-10-07

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · Linear Attention Recurrent State Audit

Verifies $O(1)$ constant memory scaling of recurrent and state-space architectures during generation versus Transformer attention.

mambarwkvlinear-attention

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upPure mathematical verification of recurrent hidden state tensor volume.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Live Benchmark Drift Detector

Detects silent provider model weight updates, quant modifications, and temperature drift across production endpoints.

drift-detectormodel-driftregression

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates statistical score deltas against historical baseline confidence intervals.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Locale is not content

Refuse a translated label as proof that the page body is translated.

i18nclaimslocale

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDoes not translate anything.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Long-Context Distraction Audit

Audits retrieval accuracy degradation as irrelevant distractor documents are added to prompts (evaluating signal-to-noise ratio in long contexts).

long-contextdistractionretrieval

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDistractor documents must share lexical domain similarity with target needles to detect realistic confusion.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Lookahead decoding n-gram audit

Audits Lookahead Decoding parallel Jacobi n-gram candidate generation and net wall-clock speedup.

servinglookahead-decodingspeculative

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upProfiles acceptance rates and wall-clock multipliers; does not generate token trajectories.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Loopback needs curl --resolve

Refuse a Host header pointed at loopback. That request sends no SNI and hits the default server. curl --resolve can pass.

nginxsniloopback

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upcurl --resolve can pass. A pass for a named host does not send the request.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · LoRA Cache Swapping Latency Audit

Audits dynamic LoRA adapter loading latency from host system memory to GPU VRAM.

loraswappingpcie

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates measured adapter transfer latency across PCIe.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Mamba-2 SSD chunk audit

Audits Mamba-2 Structured State Space Duality (SSD) matrix chunk sizes and continuous decoding throughput.

servingmamba-2ssm

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upProfiles token throughput at given SSD chunk configurations; does not retrain state matrices.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Many-Shot Jailbreak Resilience Audit

Evaluates model safety adherence under long-context many-shot in-prompt jailbreak demonstrations.

many-shotjailbreakadversarial

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates refusal accuracy on adversarial benchmark suites.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · master is not force-pushed

Refuse a force push whose branch is master. Another branch is outside this refusal, and a pass does not approve the force push.

gitforce-pushmaster

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upA non-force push can pass. A pass for another branch does not approve a force push.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · MCP permission map

Bind a saved MCP tools/list capture to explicit permission labels and reject a read-only claim that a label contradicts.

mcppermissionstools

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upCompares human labels with the capture. It does not start the server or infer permissions from descriptions.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Meter-scope audit

Refuse a unit claim that borrows an invoice unit from a note that does not name that meter, so a knowledge-base billing sentence cannot reprice an inference row.

pricingunitsmeters

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upChecks supplied meter names and allowlisted unit strings. It does not fetch a page, convert units, or read amounts. A pass is not a vendor quote.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Mixed Precision Layer Allocation Audit

Audits selective high-precision preservation on sensitive initial and final layers versus bulk quantized middle layers.

mixed-precisionlayer-allocationfp16

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic parameter weighted average bit calculation.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Model card transparency audit

Audit model card releases against the Foundation Model Transparency Index, checking documentation completeness across pre-training data, compute hours, and limitations.

transparencymodel-cardsgovernance

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upChecks presence of documented transparency indicators; does not verify independent accuracy of training data logs.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Model Cascade Routing Audit

Audits 2-tier cascading routing policies (small draft model with confidence gate to frontier verifier), calculating net cost savings and escalation ratios.

cascaderoutingarbitrage

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upResolution rates must be calibrated on representative domain evaluation splits.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Model Distillation Fidelity Audit

Measures task accuracy retention and output distribution alignment of distilled student models relative to teacher models.

distillationfidelityteacher-student

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluation benchmarks must be out-of-distribution relative to student fine-tuning training datasets.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Model Merge Interference Audit

Measures weight interference and cosine similarity degradation in spherical linear interpolation (SLERP) model merges.

model-mergeslerpinterference

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates layer cosine similarity bounds.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Model migration preflight

Check a proposed model replacement for required modalities, context, tool use, structured output and evaluation coverage before calling it a safe saving.

modelsmigrationcompatibility

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic local checks over supplied evidence; not a guarantee of source truth or production suitability.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · MoE Activation Memory Footprint Audit

Calculates active versus total parameter memory footprint and expert weight swapping overhead on constrained GPU nodes.

moememoryvram

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic memory sizing. Assumes steady-state layer distribution.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · MoE Expert Parallelism All-to-All Audit

Calculates All-to-All collective dispatch and combine communication overhead across nodes in Expert Parallel (EP) clusters.

moeexpert-parallelismall-to-all

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates ring/pairwise All-to-All collective equations.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · MoE Expert Routing Tax Audit

Audits top-k gating entropy, expert load distribution, and token drop rates under finite expert capacity factors.

moemixture-of-expertsgating

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates observed routing frequencies and capacity factors. Does not modify router logits.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Multi-agent error cascade audit

Audits error compounding probability across sequential multi-agent tool execution pipelines.

evaluationmulti-agenterror-cascade

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upComputes joint failure probability based on observed step error rates; does not execute agent tool steps.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Multi-Head Latent Attention Audit

Calculates DeepSeek Multi-Head Latent Attention (MLA) low-rank compression factors and verifies KV cache reduction relative to MHA and GQA.

mladeepseekattention

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upPure mathematical verification of KV projection tensor sizes and per-token memory footprint.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Multi-LoRA Kernel Overhead Audit

Measures batched multi-adapter GPU kernel overhead (S-LoRA) versus single-model inference throughput.

loras-loramulti-tenant

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates measured throughput compared to pure base model baseline.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Multi-Token Prediction Speedup Audit

Calculates acceptance rate thresholds and net speedup of Multi-Token Prediction (MTP) heads vs standard autoregression.

mtpmulti-token-predictionspeculative

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates observed acceptance rate and head compute cost.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Multi-Turn State Drift Audit

Calculates cumulative token inflation and prefill cost scaling across multi-turn agent conversations to verify context compaction policies.

multi-turncontext-compactionstate-drift

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates cumulative turn sums and compacted history savings.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Multilingual Retrieval Parity Audit

Audits cross-lingual embedding alignment and retrieval recall parity across high- and low-resource languages.

multilingualembeddingsretrieval

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upBenchmark queries must be human-curated and idiomatic in the target locale.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Multiplier-aside audit

Refuse a base-rate claim when the excerpt names a multiplier. Priority, fast, and scale factors are not the on-demand rate. The number in 2x is not read as a price.

pricingmultiplieraudit

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upLooks for multiplier wording in the excerpt you supply. Negation is not parsed. It does not read amounts or apply a factor. A pass is not a vendor quote.

Source reviewed

3 sources · 2026-10-07

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · Native Sparse Attention Audit

Audits Native Sparse Attention (NSA) compression ratios, block-sparse top-k selection, and memory bandwidth reduction.

nsasparse-attentionnative-sparse

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates block-sparse index selection math. Does not train sparse kernels.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Needle-in-a-haystack density audit

Audit long-context retrieval recall across varying context lengths, needle depth placements, and distractor similarity levels.

retrievallong-contextneedle-in-haystack

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upAnalyzes supplied retrieval test results; does not execute long-context inference runs.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · nginx is reloaded, not restarted

Refuse an nginx restart. A reload is review until nginx -t has succeeded. This checker does not run nginx.

nginxreloadrestart

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upA tested reload can pass. This checker does not run nginx or reload a host.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · NPU INT8 TPU offload audit

Audits mobile NPU and edge accelerator power efficiency measured in Tokens per Joule.

servingnpuedge

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upCalculates Tokens per Joule metrics; does not interface with hardware power meters.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Offer join

Refuse a price that does not name both the model and the seller.

pricingidentityoffers

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDoes not fetch a price.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Open-weights licence audit

Audit open-weights model licenses against commercial deployment restrictions, Monthly Active User thresholds, synthetic data output constraints, and attribution requirements.

licensingopen-weightscompliance

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upTechnical license boundary check; does not constitute formal legal counsel.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · P99.9 tail latency budget audit

Audits tail latency distribution ratios (p99.9 / p50) to detect micro-stalls and jitter in production serving.

servingtail-latencyp999

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upCalculates percentile ratio bounds from reported latency summaries; does not record live request times.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Page is not a quote

Keep a pricing-page research note from becoming a joined rate. A spelled currency and unit are not a price.

pricingprovenanceword-bar

Why we selected itIt keeps a fetched pricing page from becoming a joined rate: a spelled currency and unit are not a price.

Weigh upThis check does not decide whether a provider publishes prices. A failed fetch is not that finding either.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Parallel Tool Call Fan-Out Audit

Calculates concurrency speedup and total payload token loads for simultaneous multi-tool execution in agent loops.

tool-callparallelismfan-out

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates observed tool execution latencies and serial sum comparisons.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Payback basis

Check switching-cost payback against savings per period using exact arithmetic. Use when a migration claims to pay back in a stated number of periods.

costmigrationarithmetic

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upExact savings-based arithmetic only; repeating results remain rational pairs. Currency and period alignment, taxes, timing, discounting and source truth are not verified.

Source reviewed

3 sources · 2026-10-07

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · PII Scrubbing Latency Tax Audit

Measures token count shifts and millisecond pre-inference latency added by NER-based redaction and token replacement pipelines.

piiprivacyredaction

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates deterministic byte and token drift metrics. Does not judge NER classification precision.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Pipeline parallelism bubble audit

Audits pipeline schedule bubble fraction and microbatch allocation in 1F1B schedules.

servingparallelismpipeline-parallel

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upCalculates 1F1B bubble fraction based on microbatch counts; assumes uniform stage execution latency.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Plan-quote hygiene

Keep unverified subscription quotes out of a USD monthly ceiling. A source sentence is not a price.

plansquotescurrency

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic local checks over a supplied plan table; not a guarantee that a vendor page still matches.

Source reviewed

3 sources · 2026-10-07

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · Population denominator

Require a stated population, a computed population and a claimed count to be the same set, so a statistic on a subset is not published as the whole set.

denominatorspopulationclaims

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upSet equality over supplied membership flags. It does not compute a correlation, a price or a rank.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Population scope

Check that a claimed count uses the population the sentence names.

populationclaimsdenominators

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDoes not recompute a market statistic.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Precision prompt cost estimator

Calculate inference costs across prompt tokens, completion tokens, cached prompt tokens (read/write), reasoning tokens, and multi-turn trajectories using explicit provider pricing ladders.

pricingtokensprompt-caching

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic decimal calculations over supplied token counts and pricing ladders; does not model network egress or batch discounts.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Prefill-Decode Disaggregation Audit

Audits separate prefill and decode worker node topologies and calculates RDMA KV transfer latency vs GPU compute overlap.

disaggregationprefill-decoderdma

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates RDMA network transfer bandwidth vs compute overlap bounds.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Prefix Caching Radix Tree Audit

Audits Radix tree prefix caching efficiency, memory allocation, cache hit rate, and effective TTFT reduction to prevent latency degradation under memory pressure.

prefix-cachingradix-treevllm

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upAssumes deterministic prompt prefix sharing. Does not simulate dynamic sequence eviction under memory fragmentation.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Process reward model tax audit

Audits Process Reward Model (PRM) step-level verification token overhead versus outcome accuracy gains.

evaluationprmreward-model

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates observed benchmark differentials and token counts; does not execute reward models.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Prompt Extraction & Leakage Resilience Audit

Evaluates system prompt confidentiality retention under adversarial prompt extraction and reconstruction attacks.

prompt-leakagejailbreaksecurity

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upSystem prompts containing sensitive credentials violate security policies; never store secrets in prompts.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Prompt Injection Classifier Tax Audit

Quantifies TTFT latency inflation and false-positive rejection rates of pre-flight security moderation guardrails.

prompt-injectionjailbreaksecurity-classifier

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates observed classifier latency and measured false positive rates across benchmark datasets.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Provider Failover Latency Tax Audit

Measures circuit breaker timeouts, retry amplification, and failover latency tax during upstream provider outages.

failovercircuit-breakerresilience

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates measured timeout windows and retry latency.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Provider quote comparison

Compare like-for-like provider quotes with complete context tiers and distinct sellers; recompute workload cost without inventing cache or reasoning discounts.

pricingproviderstier-ladders

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic local checks over supplied evidence; not a guarantee of source truth or production suitability.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · PyramidKV layer budget audit

Audits PyramidKV non-uniform layer cache allocations to verify total VRAM budget compliance.

servingpyramidkvlayer-allocation

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upVerifies layer budget sums against hardware capacity limits; does not compute per-layer attention.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Quantisation label

Keep an unknown quantisation labelled unknown, so a missing weight format is not published as full precision or fp16.

quantisationmodelslabels

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upLabel equality only. It does not inspect weights or translate aliases such as float16 and fp16.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Quantization Degradation Audit

Audits perplexity and benchmark score degradation across quantization levels (BF16, FP8, INT4 AWQ, Q4_K_M) to prevent claiming unquantized performance for quantized serving.

quantizationbenchmarksfp8

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upScores must derive from identical evaluation harness configurations and matching temperature.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Queue queuing theory audit

Audits M/M/c queuing models under variable arrival rates to predict queue buildup before cluster saturation.

servingqueuing-theorymmc

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upCalculates predicted queue wait times from arrival and service rates; does not execute HTTP traffic.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Quote-join audit

Join a vendor quote to a catalogue model only when the provider and the model id match exactly, so a case fold is not an identity.

identitypricingjoin

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upExact string equality only. A join is not a quality, context, modality or price claim, and the checker does not apply an alias table.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · RAG Chunk Overlap Tax Audit

Calculates redundant token costs incurred by sliding-window chunk overlap in RAG ingestion and vector indexing pipelines.

ragchunkingvector-db

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upOverlap reductions must be validated against retrieval boundary preservation.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Rate-unit audit

Refuse a token-price conversion until the currency and the unit are both explicit, so per M or a bare cent figure cannot become a per-million USD rate.

pricingunitscurrency

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic local conversion of supplied decimal strings; a scaled figure is not a new vendor quote and the checker does not guess that M means million.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Read-only claim audit

Reject a read-only label when a listed tool name contains a mutation token, without treating a passing name list as proof the server is read-only.

mcppermissionsread-only

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upChecks name tokens only. A pass is not proof the handlers are read-only, and a name outside the token list still needs the handler opened.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · README is not a licence

Refuse a README sentence as a software licence.

licensingredistributionreadme

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDoes not identify an SPDX licence from prose.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Reasoning Budget Allocator Audit

Audits dynamic reasoning token ceilings to prevent cyclical reasoning explosions while guaranteeing sufficient tokens for complex proofs.

reasoningbudgetcot

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upVerifies token envelope compliance and sentinel termination. Does not grade mathematical correctness of reasoning.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Reasoning token audit

Audit reasoning token expansion ratios, budget caps, separate billing vs output billing, and hidden multiplier surcharges across reasoning models.

reasoningtokenspricing

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic local checks over supplied token counts; does not evaluate model reasoning depth or solution correctness.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Reasoning trace redundancy audit

Audits reasoning trace self-correction loops and verbosity redundancy to prune repetitive reflection tokens.

evaluationreasoningverbosity

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates trace redundancy percentages and fidelity preservation; does not generate answers.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Receipt drift

Fail a verification receipt whose hashes no longer match the supplied bytes.

receiptshashesverification

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upHashes only the bytes it is given.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Rendered-surface and payload agreement

Prove a rendered surface and the JSON it serves state the same value for every supplied field.

agreementprovenancepayload

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upCompares two supplied snapshots field by field; it does not fetch either surface, prove a release shipped both halves, or establish which side is correct.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Resource release proof

Verify local release artifacts and meaningful public-response receipts without confusing a build, HTTP 200 or registry metadata with a working installation.

releasesverificationdistribution

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic local checks over supplied evidence; not a guarantee of source truth or production suitability.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · Restriction of range

Refuse a correlation or agreement claim whose evidence population is narrower than the population the sentence names.

correlationpopulationbenchmarks

Why we selected itIt refuses a correlation or agreement sentence whose evidence identifiers are a narrower population than the sentence names.

Weigh upIdentity match is not evidence that two benchmarks agree.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Review-gate ratchet

Check that a review gate moved in one direction only, stricter and never looser, so an automated reviewer cannot weaken a check to pass a change.

ratchetreviewgates

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic local check of one rule set against another; it does not edit a gate, judge whether the resulting severities are correct, or run the gate itself.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Ring Attention Interconnect Audit

Verifies communication-to-computation overlap in block-wise Ring Attention across distributed sequence-parallel nodes.

ring-attentionsequence-parallellong-context

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates steady-state block pipelining. Does not model warmup or cooldown steps.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Ring attention latency audit

Audits Ring Attention circular KV cache passing latency against block compute time in 1M+ context inference.

servingring-attentioncontext-parallelism

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upProfiles communication-to-compute ratio; does not modify CUDA streams.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Ring MLA hybrid audit

Audits Ring Attention combined with Multi-Head Latent Attention (MLA) compressed latent states.

servingmlaring-attention

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upCalculates compression ratios of MLA latent projections; does not verify latent reconstruction loss.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · RWKV-6 linear state audit

Audits RWKV-6 linear recurrent state decay parameters and numerical stability across long token horizons.

servingrwkvlinear-attention

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates accumulated drift epsilon against tolerance limits; does not compute forward passes.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Schema Validation Retry Cost Audit

Quantifies wasted token expenditures incurred by recursive error correction loops when models emit invalid tool parameters.

validation-retryerror-recoverytoken-waste

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates retry counts and token billing sums.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Schema-not-quote audit

Refuse a model-rate claim when the excerpt describes a pricing field or a schema. A sentence that says values are in a unit is a field description, not a model's rate. Amounts are not read.

pricingprovenanceaudit

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upLooks for schema phrases in the excerpt you supply. Negation is not parsed. It does not read amounts. A plain pass is not a vendor quote.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · Self-Healing Prompt Optimizer Audit

Audits automated prompt refinement iterations to resolve edge-case test failures without human intervention.

self-healingprompt-optimizationautomated-repair

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates iteration convergence count and test pass rates.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Self-play distillation tax audit

Audits teacher-student self-play distillation token expenditure against student benchmark accuracy gains.

pricingdistillationself-play

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upCalculates financial cost per benchmark point; does not execute training runs.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Self-Speculative Early Exit Audit

Audits layer skipping and early exit draft verification within a single model architecture.

early-exitself-speculativelayer-skip

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates exit layer depth and accuracy retention.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Seller-spread audit

Check a price-spread multiple against distinct seller owners, so one vendor's own service tiers are not counted as competition.

pricingsellersspread

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic local checks over supplied rows and caller-assigned seller owners; not a guarantee of source truth or market coverage.

Source reviewed

3 sources · 2026-10-07

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · Semantic Prompt Pruning Audit

Measures token reduction percentages and verifies mutual information retention in token pruning pipelines.

prompt-pruningcompressiontoken-cost

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates observed token counts and task verification flags.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Semantic Router Latency & Token Tax Audit

Quantifies embedding model latency and static classifier token taxes in semantic routing topologies prior to upstream generation.

semantic-routingembeddinglatency

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upClassifier embedding cache hit rates should be considered under repetitive query distributions.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Sequence parallelism ring audit

Audits sequence parallelism communication overhead across layernorms and attention projection.

servingparallelismsequence-parallel

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates observed sequence communication latency; does not alter GPU kernel launch dispatch.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Serving-mode attribution

Check that an explicit claimed score belongs to its named model or serving mode. Use when a model headline may have inherited a mode-specific score.

benchmarkidentityattribution

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upChecks explicit claimedScore and named-subject equality only. It does not verify benchmark provenance, source truth or shared measurement conditions.

Source reviewed

3 sources · 2026-10-07

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · Shared-lastmod audit

Refuse a crawl list where every page is dated on the build day, so a shared build clock is not published as a per-page fact date.

seositemaplastmod

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upChecks supplied dates only. One page dated on the build day can be a real change. A shared older day can be the day a record began. It does not fetch pages or decide which fact moved.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Significance tie

Treat two scores as tied when their difference falls inside the stated interval.

scoresintervalsties

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDoes not invent an interval.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Single-base audit

Refuse a single-base price claim when the excerpt names both sides of a cache split or a peak split. Spelling USD per million does not make cache-hit and cache-miss one rate.

pricingcacheschedule

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upChecks two splits in the excerpt you supply. It does not fetch a page, read amounts, or decide which sentence is the invoice. Batch, region, and input-versus-output are outside this check. A pass is not a vendor quote.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · Sliding Window Attention Audit

Audits sliding window context buffers and attention sink token retention to bound KV memory in continuous streaming.

sliding-windowattention-sinkstreaming

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates window buffer size and attention sink retention parameters.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · SLO error budget depletion audit

Audits 30-day Service Level Objective (SLO) error budget burn rates to trigger automated traffic shedding.

servingsloerror-budget

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upCalculates multi-window burn rates from telemetry inputs; does not configure Prometheus alert rules.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Slug-is-not-body audit

Refuse a currency or quantity claim whose words appear only in a URL, a slug, or a title. The excerpt is the body. A path named pricing-details-usd does not state USD.

pricingprovenancecurrency

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upReads the excerpt only. A locator is ignored. It does not fetch a page or read amounts. A pass is not a vendor quote.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · SnapKV cluster compression audit

Audits SnapKV prompt attention feature clustering and compressed KV bank generation ratios.

servingsnapkvkv-cache

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upMeasures effective token compression ratios; does not evaluate clustering loss directly.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Source-pin audit

Refuse a source identity that is only a branch, a tag, or a short SHA. A full 40-hex revision is the pin. The name v7 or main does not become that revision.

provenancerevisionaudit

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upChecks the revision string you supply. It does not fetch a repository, resolve a tag, or prove the tree was read. A pass is not a source review.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · Speculative Decoding Arbitrage Audit

Computes draft model acceptance rate thresholds (alpha) and verifies whether speculative decoding delivers net throughput/cost benefit over standard autoregressive decoding.

speculative-decodingthroughputlatency

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upAcceptance rates vary significantly across coding, math, and conversational domains.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Spelled-meter audit

Refuse a currency or quantity claim whose excerpt does not spell the word. A dollar sign is not USD, "per 1M" is not million, and the character 元 is not CNY.

pricingcurrencyunits

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upChecks words in the excerpt you supply. It does not fetch a page, read amounts, or decide which sentence is the rate. A pass is not a vendor quote.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Status-not-page audit

Check a supplied status against a conservative integer-200-only gate and refuse every quote claim based on status alone. No status proves a body was received or read. This checker sends no request.

provenancepricingaudit

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upChecks a conservative status policy only. A 200 result is status-eligible; every other valid status needs review and every quote claim is refused. No status establishes that a response body was received or read; bodyVerified is always false.

Source reviewed

3 sources · 2026-10-07

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · Stream Truncation & Sentinel Detector

Detects silent SSE connection drops in long-running streaming completions where no [DONE] sentinel or incomplete JSON chunk is emitted.

streamingssereliability

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upClient implementations must buffer partial JSON frames across chunk boundaries.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Streaming framing audit

Audit Server-Sent Events (SSE) streaming protocol compliance, chunk serialization overhead, heartbeat keep-alive frequency, and terminal sentinel completeness.

streamingsseprotocol

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upAudits recorded stream traces; does not open live TCP sockets.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · sudo does not read the script from stdin

Refuse sudo bash -s fed by a heredoc. That script is read as the password prompt. sudo -n bash -lc can pass.

sudosshstdin

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upsudo -n bash -lc can pass. This checker does not open an SSH session.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Suggest, do not redirect

Refuse a language redirect, including when Accept-Language is absent. A missing header is not a request for another locale. Suggest, or do nothing.

i18nredirectlanguage

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upAn absent header and a present header are both reviews when the action is redirect. The checker sends no request.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Synthetic data deduplication audit

Audits MinHash LSH and cosine clustering deduplication to prune repetitive synthetic training generations.

evaluationsynthetic-datadeduplication

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upMeasures cluster diversity metrics and pruning ratios; does not train model checkpoints.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Synthetic Eval Diversity Audit

Audits lexical and semantic diversity in synthetic evaluation sets to prevent test-set memorization and benchmark gaming.

synthetic-evaldiversitybenchmark-gaming

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates pairwise n-gram overlap and embedding cosine dispersion.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Synthetic Text Watermark Retention Audit

Measures statistical green/red list token watermarking retention under editing, summarization, and paraphrasing.

watermarkingdetectionai-provenance

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates green-token z-score significance thresholds.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · System prompt extraction guard audit

Audits resistance against system prompt leakage attacks and multi-turn extraction attempts.

evaluationsecurityleakage

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates embedding similarity between candidate outputs and protected system prompts.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · System Prompt Override Resilience Audit

Measures system directive adherence under conflicting user instructions and adversarial persona hijacking.

system-promptoverride-attackpersona-hijack

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates adherence rates on adversarial prompt override datasets.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Task vector orthogonality audit

Audits task vector pairwise cosine similarity to prevent subspace interference in multi-task checkpoints.

evaluationtask-vectorsorthogonality

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upMeasures pairwise cosine similarity; does not compute forward passes.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Tensor parallelism communication audit

Audits Tensor Parallelism All-Reduce communication overhead against NVLink bus saturation.

servingparallelismtensor-parallel

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upCalculates theoretical vs measured All-Reduce communication bus times; does not modify NCCL rings directly.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Tensor Parallelism Communication Tax Audit

Calculates All-Reduce collective communication volume and latency overhead across NVLink and PCIe interconnects to prevent negative TP scaling.

tensor-parallelismall-reducenvlink

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upModels standard Ring All-Reduce collective communication. Does not simulate overlap with custom CUDA kernels.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Test-Time Compute ROI Audit

Calculates cost-per-accuracy return curves for Best-of-N sampling, search depth scaling, and verifier reranking vs base model upgrading.

test-time-computebest-of-nverifier

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates candidate generation and verifier pricing. Assumes calibrated reward model ranking.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Test-time search budget audit

Audits test-time compute search budgets to ensure marginal accuracy gains justify token expenditure.

evaluationreasoningtest-time-compute

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates observed pass@k curves against token budgets; does not generate reasoning trees.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Throughput benchmark audit

Audit throughput and latency claims (tokens/second, TTFT, concurrency, batch sizes) to ensure apples-to-apples comparisons and detect deceptive single-user vs batched measurements.

throughputlatencyttft

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic validation of supplied benchmark measurements; does not execute GPU benchmarks directly.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · TIES merging trim elect audit

Audits TIES-Merging parameter trimming threshold and sign consensus across fine-tuned task vectors.

evaluationmodel-mergingties-merging

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates parameter sign agreement ratios; does not perform checkpoint serialization.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Time-window rate

Refuse an off-peak rate published as the all-day rate.

pricingwindowsclaims

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDoes not fetch vendor windows.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Token blowout circuit breaker audit

Audits runaway loop detection and trips circuit breaker when token expenditures exceed budget ceilings.

pricingcircuit-breakertoken-budget

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upVerifies token expenditure thresholds; does not kill running remote API calls.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Tokenizer density audit

Audit tokenizer vocabulary density, tokens-per-word / tokens-per-byte expansion ratios, and real effective text processing costs.

tokenizersdensityarbitrage

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upCalculations based on provided word and token counts for a benchmark corpus; not a tokenizer training tool.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Tool allowlist

Decide whether a saved MCP permission map fits a task allowlist. A consistent refusal is a pass, not an error to paper over.

mcpallowlisttools

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upUses the labels already stored in the receipt. It does not call tools or widen the allowlist.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Tool schema tax audit

Calculate the hidden static token overhead imposed by injecting JSON schema and OpenAPI tool definitions into prompt payloads on every turn.

toolsschemasoverhead

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic calculation over schema token metrics; does not generate tool definitions.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Tool Selection Recall Audit

Audits tool routing accuracy across large tool inventories and detects hallucinated function invocations.

tool-selectionfunction-callingrecall

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates precision and recall on ground-truth benchmark intent datasets.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Tree Speculative Decoding Audit

Evaluates tree-based draft candidate verification (Medusa/Eagle-2) to verify branch acceptance probabilities.

tree-speculativemedusaeagle

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates tree acceptance statistics across test runs.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · TTFT jitter audit

Audit Time-To-First-Token variance, queuing delays, cold-start penalization, and tail latency distributions (p50, p95, p99).

latencyttftjitter

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upStatistical analysis of supplied latency samples; does not execute live network ping probes.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Ulysses all-to-all bandwidth audit

Audits DeepSpeed Ulysses All-to-All transpose communication overhead versus interconnect saturation.

servingulyssescontext-parallelism

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates observed All-to-All transpose times against interconnect capacity.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Unrated sentinel

Keep a missing score out of a ranked list, so unrated is not scored zero and a scored row is not dropped in silence.

rankingunratedscores

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic check of one supplied table. It does not fetch scores or decide that a measured zero is wrong.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Upstream freshness budget

Measure each upstream source against its own declared staleness budget, so a cache nobody refetched cannot pass for current.

freshnesscacheprovenance

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upDeterministic local check over declared budgets and supplied fetch dates and content hashes; it does not fetch, refresh or validate the sources themselves.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Vector Reranking Latency Tax Audit

Calculates cross-encoder reranking latency overhead and cost versus top-k bi-encoder retrieval recall.

rerankingcross-encoderretrieval

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates observed reranking latency and recall margin.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Verifier Consensus Cost Audit

Calculates token expenditures and marginal certainty of majority voting consensus across parallel reasoning traces.

majority-votingconsensusverifier

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates exact vote distributions and token pricing.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Video frame subsampling audit

Audits spatio-temporal video token subsampling to prune static background tokens in long video streams.

servingvideomultimodal

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upVerifies token reduction ratios and task accuracy metrics; does not decode video frames.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · WebGPU browser inference audit

Audits in-browser WebGPU execution shader compilation latency and first-token response times.

servingwebgpubrowser

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates recorded browser navigation timing milestones; does not initialize WebGPU contexts.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Withdrawal residue

Check an explicitly observed withdrawn numeric field and its named dependents for leftover values. Use after a source value has been removed from a snapshot.

withdrawalevidenceconsistency

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upRequires the explicit observed withdrawnValue and named numeric dependents. It cannot verify withdrawal or discover omitted dependencies; zero remains residue and empty coverage requires review.

Source reviewed

3 sources · 2026-10-07

Read the full review Install command in reviewMIT

Evidence and evaluation · Undominated.ai

Undominated · Withheld provider names

Check a supplied publication denylist by exact provider-name matching. A pass does not approve a provider or establish that the supplied list is complete.

providerspublicationnames

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upOnly names in the caller-supplied withheldNames list are withheld. The checker does not load or verify a production policy; a pass does not approve publication.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Workstation GPU Serving Audit

Calculates local inference throughput and hardware payback periods for desktop RTX workstations versus cloud API spend.

workstationrtxdesktop-inference

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upEvaluates observed local throughput and monthly cloud subscription spend.

Source reviewed

3 sources · 2026-10-08

Read the full review Attributed definition availableMIT

Evidence and evaluation · Undominated.ai

Undominated · Zero Data Retention policy audit

Audit provider Zero Data Retention (ZDR) claims against published Terms of Service, abuse monitoring windows, logging exceptions, and training opt-out guarantees.

zdrprivacygovernance

Why we selected itPairs a bounded workflow with an offline Python checker and explicitly synthetic example inputs.

Weigh upChecks documented policy fields against enterprise requirements; does not perform physical datacenter security audits.

Source reviewed

3 sources · 2026-10-14

Read the full review Attributed definition availableMIT

Data and analytics · dbt Labs

Upgrading dbt

Plans a dbt v1 migration to 1.12 using adapter-specific issue bundles and a recorded verification gate.

dbtmigrationwarehouses

Why we selected itSeparates sequential pre-1.8 changes from detected behavior flags and preserves unresolved outside-repository job changes.

Weigh upIts stated target is 1.12, not a promise to reach whichever release is newest. Starting versions 1.3–1.7 have bundles; later versions have narrower flag handling. Parsing alone does not prove migration correctness.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Model evaluation · Braintrust

Validate an Eval Scorer

Checks an LLM judge or scorer against expert-reviewed labels, using agreement with uncertainty and severity-weighted errors.

evaluationllm-as-judgeagreement

Why we selected itIt requires naming the reference tier first, says a strong-model reference supports iteration only, and calibrates fitness bands to adjudicated human labels.

Weigh upThe body cites fitness bands in reference.md, a file not present in the skill directory at this revision. The header points to references/validation-recipe.md, which was not opened.

Source reviewed

5 sources · 2026-10-13

Read the full review Install command in reviewApache-2.0

Security and privacy · Trail of Bits

Variant Analysis

Expands a confirmed vulnerability into related-code searches while preserving root-cause evidence and triage decisions.

variant-analysiscodeqlsemgrep

Why we selected itStarts from a verified instance and broadens one abstraction at a time, with explicit false-positive review.

Weigh upPattern matches are candidates, not confirmed vulnerabilities; reachability, sanitization and impact still need review.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Testing and reliability · Trail of Bits

Vector Forge

Finds implementations of a cryptographic algorithm, runs mutation testing to locate escaped mutants, generates test vectors that exercise the uncovered paths, and compares mutation kill rates before and after.

test-vectorswycheproofmutation-testing

Why we selected itFinds a crypto implementation, runs mutation testing to identify escaped mutants, then generates test vectors that exercise the uncovered code paths.

Weigh upIt installs the tools it needs: `uv tool install trailmark` when missing, and mutation frameworks per references/mutation-frameworks.md, whose Installation Policy says they MUST be installed (cargo install, brew, go install …@latest, sudo package managers).

Source reviewed

6 sources · 2026-10-22

Read the full review Upstream setup linkedCC-BY-SA-4.0

Cloud operations · Vercel

Vercel Optimize

Runs a metrics-first Vercel audit: collects usage signals before reading source, gates candidates deterministically, then produces ranked cost and performance recommendations.

vercelcost-optimizationperformance

Why we selected itEnforces an observability-first pipeline: no source file is inspected until signals.json exists and a deterministic gate names a route, file, or setting, which blocks speculative rewrites.

Weigh upNeeds Vercel project access and usage metrics; without signals the workflow stops by design and produces nothing.

Source reviewed

8 sources · 2026-10-13

Read the full review Install command in reviewMIT declared in README; no LICENSE file at 063bee94

Web development · Vercel

Vercel React Best Practices

Organizes React and Next.js performance reviews around request waterfalls, bundles, server work and rendering costs.

reactnextjsperformance

Why we selected itPairs rules with incorrect and corrected code, making recommendations easier to inspect.

Weigh upPublished impact labels and example speedups are not measurements of your application.

Source reviewed

7 sources · 2026-10-14

Read the full review Install command in reviewMIT (declared in upstream README)

Web development · Vercel

Vercel React Composition Patterns

Replaces growing boolean-mode APIs with explicit component composition and provider-owned state contracts.

reactcomponent-designcontext

Why we selected itShows complete before-and-after component variants rather than only naming the compound-component pattern.

Weigh upContext and compound components can add indirection to a simple component; apply the pattern where variation warrants it.

Source reviewed

7 sources · 2026-10-14

Read the full review Install command in reviewMIT (declared in upstream README)

Mobile development · Vercel

Vercel React Native Skills

Provides React Native and Expo patterns for scrolling, animations, native navigation and dependency layout.

react-nativeexpomobile

Why we selected itRules are split by priority, with list performance marked critical and each rule living in its own file.

Weigh upSKILL.md is an index; applying a rule without opening its rules/*.md file skips the incorrect and correct examples.

Source reviewed

7 sources · 2026-10-14

Read the full review Install command in reviewMIT (declared in upstream README)

Design and interfaces · Vercel

Vercel Web Interface Guidelines

Reviews interface code against a freshly fetched guideline set and returns file-specific findings.

accessibilityformsui-review

Why we selected itThe review procedure is short and explicit: fetch the guideline file, read the named UI files, and emit line findings.

Weigh upThe rules are not in this repository; every review must fetch an unpinned main URL, so the rule set can change without this commit changing.

Source reviewed

6 sources · 2026-10-14

Read the full review Install command in reviewMIT (declared in upstream README)

Creative production · affaan-m

Video Editing

Runs a layered AI video-editing pipeline — transcript-driven structuring, deterministic FFmpeg cuts from an edit decision list, Remotion composition, selective asset generation, human polish — for turning long recordings into published video.

video-editingffmpegremotion

Why we selected itSeparates the pipeline into layers with strict jobs — the model structures, FFmpeg cuts deterministically, Remotion composes, humans supply taste — instead of pretending one tool does everything.

Weigh upThe later layers call third-party services (ElevenLabs with an API key from the environment, fal.ai, VideoDB, Screen Studio, Descript); only the FFmpeg and Remotion layers run without one.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Languages and tooling · JetBrains

Vite

Vite build-tool guidance pinned to the Vite 8 transition: vite.config.ts discipline, import.meta.glob and asset queries, the plugin API and virtual modules, library mode and SSR builds, the multi-environment API, and Rolldown/Oxc migration.

vitebundlervite-config

Why we selected itPinned to the Vite 8 transition — Rolldown bundler and Oxc transformer replacing esbuild plus Rollup — with a dedicated rolldown-migration reference, which is exactly the migration agents are being asked to perform.

Weigh upThe mirror generation is dated 2026-01-31 against Vite 8 beta; the upstream antfu/skills generator has since regenerated against Vite 8.3.1 stable (2026-09-25).

Source reviewed

6 sources · 2026-10-07

Read the full review Install command in reviewMIT (upstream antfu/skills)

Web development · JetBrains

VitePress

Documentation-site guidance for VitePress: .vitepress configuration, file-based routing and rewrites, Markdown extensions, Vue components inside Markdown, build-time data loaders, dynamic routes, theming, internationalization, SSR compatibility, and deployment recipes.

vitepressdocumentation-sitesstatic-site-generator

Why we selected itCovers the documentation-site generator end to end — config, file-based routing, Markdown containers and includes, Vue-in-Markdown, data loaders, dynamic routes — each routed to its own reference file.

Weigh upThe mirror generation is dated 2026-01-28 and based on VitePress 1.x; the upstream generator has since produced a VitePress 2.0.0-alpha.20 revision (2026-09-25).

Source reviewed

6 sources · 2026-10-07

Read the full review Install command in reviewMIT (upstream antfu/skills)

Testing and reliability · JetBrains

Vitest

Routes an agent through Vitest one topic at a time: configuration shared with the Vite pipeline, the test/describe/expect APIs and hooks, mocking with the vi utilities, snapshot testing, V8 versus Istanbul coverage, fixtures via test.extend, environments, and type-level testing.

vitestunit-testingjest-compatible

Why we selected itSplits the parts of Vitest that agents get wrong into dedicated reference files — vi mocking of modules, timers and dates, snapshot testing, and coverage provider choice — instead of compressing them into prose.

Weigh upThe mirror carries the January 2026 generation, based on Vitest 3.x; the upstream antfu/skills generator has since produced a Vitest 5.0.1 revision (2026-09-25) that the pinned mirror revision does not include.

Source reviewed

6 sources · 2026-10-07

Read the full review Install command in reviewMIT (upstream antfu/skills)

Testing and reliability · vLLM project

vLLM Bench Serve

vllm bench serve reference for measuring throughput, TTFT, TPOT and goodput against vLLM or any OpenAI-compatible endpoint.

vllminference-benchmarklatency

Why we selected itIt documents the load-shape controls that change results: the default --request-rate inf sends every request at once, while a fixed rate, burstiness and ramp-up model steadier traffic.

Weigh upThe table lists defaults of 1000 prompts and no warm-up requests. Pointed at a paid remote API, a run spends quota at the provider's rates, and the file gives no cost warning.

Source reviewed

3 sources · 2026-10-13

Read the full review Install command in reviewApache-2.0

Design and interfaces · Microsoft

VS Code Accessibility

VS Code’s required accessibility discipline for feature work: help dialogs, accessible views and verbosity settings for new interactive surfaces, plus ARIA live-region, keyboard-navigation and labelling rules for any UI change.

accessibilityariascreen-readers

Why we selected itTurns screen-reader support into a buildable contract: every new interactive surface must ship an accessibility help dialog and a verbosity setting, plus an accessible view when it renders rich visual content, with a per-feature checklist that names the registration points.

Weigh upThe implementation sections are written against VS Code’s internal accessibility framework (IAccessibleViewImplementation, AccessibleViewProviderId, accessibilityConfiguration), not portable web APIs.

Source reviewed

5 sources · 2026-10-15

Read the full review Install command in reviewMIT

Localization · GitHub community

VS Code Extension Localization

Localizes VS Code extensions by routing each string to its correct target: package manifest files, walkthrough markdown, or code l10n bundles.

localizationvscode-extensionsi18n

Why we selected itThree localization approaches keyed to resource type (manifest, walkthrough, code strings) instead of one blanket rule applied to all strings.

Weigh upVS Code extensions only; not a general i18n-framework guide and silent on pluralization, ICU formats and RTL layout.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Testing and reliability · GitHub community

Vue 3 Pinia Unit Testing

Writes and reviews unit tests for Vue 3 + TypeScript + Vitest + Pinia codebases: behavior-first assertions on components, composables and stores, with createTestingPinia setups from a checked-in pattern cookbook.

vue-3composition-apipinia

Why we selected itStructures the work around the behavior boundary first — component UI, composable or store — and restricts wrapper.vm access to exceptional cases, enforcing black-box assertions over implementation details.

Weigh upAssumes the Vitest + @pinia/testing + Vue Test Utils stack; projects on Jest or without Pinia are out of scope.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Security review · Trail of Bits

Vulnerability Triage Brocards

Runs an incoming vulnerability report or automated finding through seven pass, dismiss or needs-more-info tests and returns a one-table triage summary with an accept, dismiss or needs-more-info verdict.

triagevulnerability-researchthreat-modeling

Why we selected itEach of the seven brocards is written as a falsifiable quick test, such as whether the report can say "an attacker with [capability] can [action] to achieve [impact]".

Weigh upBy default it stops at the first DISMISS; all seven brocards are evaluated only when the user asks for a full evaluation, although the summary template has a row for each.

Source reviewed

6 sources · 2026-10-21

Read the full review Upstream setup linkedCC-BY-SA-4.0

Model evaluation · Weights & Biases

W&B EvalTable Previews

Converts existing W&B Table artifacts into new EvalTable previews without overwriting source runs, with stated size caps.

wandbevaluationeval-tables

Why we selected itThe workflow runs a read-only scan first, previews one table key per invocation, and reports success only when verify-preview returns verified: true.

Weigh upConversion relies on scripts/table_artifact_to_eval_table.py and references/EVAL_TABLES.md, which were not opened. preview creates W&B runs, EvalTables and Weave evaluations.

Source reviewed

3 sources · 2026-10-13

Read the full review Install command in reviewApache-2.0

Languages and tooling · JetBrains

Wasm Inspector

Inspects WebAssembly binaries through JetBrains’ Hexana MCP server: imports, exports, memories, globals and tables read from the binary itself, ABI diffs between two builds, crash-frame triage, and host or supply-chain import audits.

webassemblywasmbinary-analysis

Why we selected itGrounds every structural claim in tool output read from the binary itself, explicitly preferring the MCP server over objdump/strings/hexdump shell probing — the discipline that keeps wasm analysis from hallucinating.

Weigh upDepends entirely on the bundled Hexana MCP server; the SKILL.md is an operating manual for that tool, not standalone guidance.

Source reviewed

5 sources · 2026-10-07

Read the full review Upstream setup linkedJetBrains Free Plugin License Agreement

Observability · Weights & Biases

Weave Instrumentation

Adds Weave tracing to Python or TypeScript agents one path at a time, with read-back checks of the stored spans.

weavetracingopentelemetry

Why we selected itThe file stops before any edit when no valid W&B login exists, forbids running wandb login in the agent session, and says not to read, print or ask for an API key.

Weigh upMechanism choice and verification rely on six reference files that were not opened for this review.

Source reviewed

3 sources · 2026-10-13

Read the full review Install command in reviewApache-2.0

Databases · Weaviate

Weaviate

Weaviate skill that runs bundled Python scripts against a Weaviate Cloud cluster to list, inspect and create collections, search them, ask the Query Agent, and import files or example data.

weaviatevector-dbvector-search

Why we selected itSKILL.md indexes a reference file per operation (Query Agent ask and search modes, hybrid search as the default, semantic and keyword search, listing, inspecting, exploring and creating collections, fetch and filter, file import, example data). Its recommendations name the bundled script to run with uv run for listing, inspecting, exploring and creating collections, importing, example data and each search mode; fetch and filter appears only as a link to its reference.

Weigh upskills/weaviate last changed in a commit dated 2026-10-16; the pinned revision is dated 2026-10-20.

Source reviewed

8 sources · 2026-10-21

Read the full review Upstream setup linkedBSD-3-Clause

Web development · Addy Osmani

Web Best Practices

Audits a rendered page against Lighthouse best-practices failures and fixes the implicated code — CSP with nonces and a report-only rollout, Trusted Types, SRI-pinned third-party scripts, hidden source maps with sourcesContent stripped, deprecated APIs.

security-headerscsptrusted-types

Why we selected itThe security reference goes beyond checklist headers: CSP as defense in depth with nonces or hashes tested report-only first, Trusted Types as the modern DOM-XSS layer, and an explicit instruction not to send X-XSS-Protection because the legacy auditor itself introduced vulnerabilities.

Weigh upLighthouse best-practices scope: the skill itself states a high score is not proof an application is secure and that this is not a complete security assessment.

Source reviewed

6 sources · 2026-10-15

Read the full review Install command in reviewMIT

Web development · Addy Osmani

Web Quality Audit

Runs an evidence-led site audit across performance, accessibility, SEO, best practices and agentic browsing, routing each check to live traces first with CLI fallbacks, and closes with re-run verification.

web-qualitylighthouseaccessibility

Why we selected itMeasured findings and code-only hypotheses are kept in separate columns, and the Performance section tracks the Lighthouse move to shared Performance Insights instead of requiring retired audit IDs.

Weigh upIts own reference list points at the sibling accessibility skill withheld in wave 1 over a px-versus-pt contrast-table error; follow this audit's ratio-only checks and current WCAG text instead of that table.

Source reviewed

8 sources · 2026-10-15

Read the full review Install command in reviewMIT

Web development · JetBrains

WebGPU Three.js TSL

Develops WebGPU-enabled Three.js applications with TSL node materials and compute shaders, covering renderer setup, WGSL integration, device-loss handling and WebGPU limits through checked-in docs, examples and templates.

webgputhreejstsl

Why we selected itThe compute-shader reference captures the TSL trap that matters most — property assignment on nodes is intercepted, JavaScript variable reassignment is not — with a works/doesn't-works table instead of prose.

Weigh upNo licence file exists in the JetBrains catalog repository or in the original dgreenheck/webgpu-claude-skill repository it snapshots; treat the definition as all-rights-reserved until clarified — this record attaches no redistributable artifact.

Source reviewed

5 sources · 2026-10-07

Read the full review Install command in reviewNot established

Data and analytics · dbt Labs

Working with dbt Mesh

Judges a dbt change by who consumes it, and states that cross-project refs require dbt Cloud Enterprise.

dbtmeshcontracts

Why we selected itA heading says cross-project refs require dbt Cloud Enterprise.

Weigh upEnterprise availability was not verified against an account.

Source reviewed

3 sources · 2026-10-15

Read the full review Install command in reviewApache-2.0

Development workflow · Trail of Bits

Writing Lean proofs

Writes Lean 4 proofs in Mathlib style, gates sorry with the kernel rather than grep, and defers to a project's own CONTRIBUTING file when it conflicts.

leanproofsformal-methods

Why we selected itUnproved obligations are gated with collectAxioms or #print axioms, not grep, because grep matches comments and misses unproved helpers.

Weigh upA verified-software project with its own house style takes precedence over Mathlib conventions; the skill says to read CONTRIBUTING first.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Security review · Trail of Bits

YARA rule authoring

Authors YARA-X rules that must be tested on the intended samples and on goodware, and forbids speculative nocase, wide, and unbounded regex.

yaradetectionmalware

Why we selected itRequires evidence for case or encoding variants before adding nocase or wide modifiers; measure their cost on the target corpus.

Weigh upWhen NOT to Use sends disassembly to Ghidra or IDA, dynamic analysis to a sandbox, network detection to Suricata or Snort, and memory forensics to Volatility.

Source reviewed

2 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

Security review · Trail of Bits

Zeroize audit

Looks for secrets that source claims to wipe but the compiler deleted.

cryptographyzeroizecompilers

Why we selected itThe description names compiler-removed zeroization, not only a missing function call.

Weigh upCC-BY-SA-4.0 applies to adaptations you redistribute.

Source reviewed

3 sources · 2026-10-14

Read the full review Install command in reviewCC-BY-SA-4.0

03 / BEHIND THE SELECTION

The review is part of the product.

A useful recommendation tells you why it fits—and what still needs checking.

What we inspect

We inspect the skill instructions and relevant supporting material, look for a concrete workflow and useful checks, and record dependencies, permissions and limitations. Each review names the files actually examined. Directory popularity is a discovery signal, not a quality verdict.

What a selection means

Each resource has a concrete use case, original selection reasons, named limitations and a dated review record. We favour specific instructions, useful scope and inspectable setup. The list is curated; it is not a census of everything available or an independently benchmarked league table.

Before using it

Read the complete review, check the upstream revision and match the resource to your own host and permissions. Review notes apply to the material named at the recorded date. Installation commands may retrieve a newer upstream version.

Where we looked

Download this index as JSON ↗
Selections we withheld (106)

These are documented decisions from this review, not a blacklist of the wider ecosystem.

  • playwright: The publisher explicitly deprecated this repository and directs current skill/plugin users to openai/plugins. Its source was inspected but it is not selected as a current recommendation. Source ↗
  • security-best-practices: The publisher explicitly deprecated this repository and directs current skill/plugin users to openai/plugins. Its source was inspected but it is not selected as a current recommendation. Source ↗
  • security-threat-model: The publisher explicitly deprecated this repository and directs current skill/plugin users to openai/plugins. Its source was inspected but it is not selected as a current recommendation. Source ↗
  • gh-fix-ci: The publisher explicitly deprecated this repository and directs current skill/plugin users to openai/plugins. Its source was inspected but it is not selected as a current recommendation. Source ↗
  • jupyter-notebook: The publisher explicitly deprecated this repository and directs current skill/plugin users to openai/plugins. Its source was inspected but it is not selected as a current recommendation. Source ↗
  • speech: The publisher explicitly deprecated this repository and directs current skill/plugin users to openai/plugins. Its source was inspected but it is not selected as a current recommendation. Source ↗
  • transcribe: The publisher explicitly deprecated this repository and directs current skill/plugin users to openai/plugins. Its source was inspected but it is not selected as a current recommendation. Source ↗
  • figma-implement-design: The publisher explicitly deprecated this repository and directs current skill/plugin users to openai/plugins. Its source was inspected but it is not selected as a current recommendation. Source ↗
  • Expo EAS Workflows: Withheld pending clearer setup and trigger guidance. The helper depends on import.meta.main but declares no Node.js version requirement. The embedded key list describes a trigger as required, while the current service schema requires only jobs and official documentation allows manual runs without on. Resolve these differences before relying on the installed instructions. Source ↗
  • Neon Postgres Branching: Withheld at the inspected revision: its lifecycle policy returns no changes for every existing branch, while the subsequent guidance says to apply that policy to an existing branch. The branching and reset material remains useful background, but this reconciliation path needs correction before recommending the installable workflow. Source ↗
  • Sentry Skill Inspection: Withheld at the inspected revision: the helper copies a matching source line into its secret-detection evidence without redacting the matched value, despite a masking comment. It also reads files before reporting symlinks. Resolve these output and input-handling concerns before running it on sensitive or untrusted material. Source ↗
  • Neon Postgres Egress Review: Withheld pending clearer operational examples. The bundled compute policy returns no changes for every existing branch while the next command is described as applying it to the current branch. Its query rankings use row and call counts rather than network bytes, and resetting statistics discards the existing measurement window. These need reconciliation before a guided cost-reduction recommendation. Source ↗
  • Sentry Django Access Review: Withheld pending clarification of the examples and reference loading. The DRF reference places an unscoped return before the scoped return in the same function, making the latter unreachable if copied. The main definition also does not link its four useful references. The investigation method has value, but the installable material needs more precise guidance before selection. Source ↗
  • Expo SDK Upgrade: Withheld after inspecting the complete references: the main workflow says to migrate uses before removing expo-av, while the video reference tells readers to uninstall it before installing expo-video. Reconcile this migration order against the target app and current Expo documentation before adoption. Source ↗
  • Stripe Best Practices: Withheld at the inspected revision: required payment and security reference links point to documentation URLs that returned HTTP 404 during this review. Related files exist in the repository, but the installed workflow needs its reference routing checked before recommendation. Source ↗
  • Stripe Upgrade Workflow: Withheld pending reconciliation: the useful live-version and no-downgrade guidance is paired with legacy global/module-level SDK examples, while the companion current best-practices skill explicitly rejects those patterns. Follow current SDK documentation instead of copying the bundled examples unchanged. Source ↗
  • Microsoft Playwright CLI Skill: Withheld at the inspected revision: its test-generation guide requires scenario files inside tests/<group>/ but imports ./fixtures even though the shown fixture lives at tests/fixtures.ts. The same guide gives conflicting sequential/parallel instructions. Correct those examples before relying on its generation workflow. Source ↗
  • Addy Osmani Accessibility: Withheld at the inspected revision: its contrast table uses CSS pixels where WCAG’s large-text exception uses points. Applying that lower size threshold could accept insufficient contrast; verify the current W3C criterion rather than copying the table. Source ↗
  • Microsoft Azure Monitor OpenTelemetry for Python: Withheld pending correction: the main definition presents instrumentations as an allow-list, while its own reference and Microsoft’s current API documentation use instrumentation_options with enable/disable entries. The configuration and default behavior need reconciliation before recommendation. Source ↗
  • Superpowers Git Worktree Setup: Withheld at the inspected revision: the ignore check succeeds when either .worktrees or worktrees is ignored, rather than checking the directory actually selected. This can let an unignored local worktree path pass the stated safety check. Source ↗
  • GitHub Implementation Plan: Withheld pending reconciliation: the required GOAL example is not a declaration under the skill’s own rules, and its separate regex checks miss cross-form duplicates. The advertised template-validation contract therefore does not match the checks provided. Source ↗
  • Datadog Monitor Management: Withheld at the inspected revision: the noisy-monitor audit sorts state-change timestamps rather than alert counts, and its ownership test looks for a literal team: tag rather than a team-prefixed tag. Those examples can misclassify the monitors they claim to audit. Source ↗
  • Datadog Flaky-Test Triage: Withheld pending reconciliation: one step sends an unknown cause to quarantine while a later decision table sends a low-rate unknown cause to escalation. The same inputs can produce different actions, including a shared test-management state change. Source ↗
  • Trail of Bits Modern Python: Withheld at the inspected revision: the requirements migration loop continues after an individual dependency import fails, then instructs removal of the old requirements files and environment. Confirm a complete dependency migration before following that destructive cleanup. Source ↗
  • Hugging Face AWS Context Discovery: Withheld at this revision: the identity branch treats an IAM Identity Center/SSO ARN as evidence that IAM roles cannot be created. Effective permissions depend on the assigned permission set and policies, so the identity alone cannot settle the preflight. Source ↗
  • Hugging Face SageMaker IAM Preflight: Withheld at this revision: the workflow inherits the assumption that an SSO principal cannot create a role and tells the user to stop on that basis. Check actual assigned IAM permissions instead of deriving this restriction from the sign-in type. Source ↗
  • Hugging Face Local Models: Withheld at this revision: the llama.cpp setup still instructs users to run make, while the current upstream Makefile explicitly rejects that removed build path and directs users to CMake. Source ↗
  • Matt Pocock Resolving Merge Conflicts: Withheld for broad use: the compact workflow directs the agent to always resolve, never abort, and stage everything before committing. It does not bound those writes to the requested conflict or protect unrelated work in a shared checkout. Source ↗
  • Svelte Code Writer: Withheld at this revision: the Bash documentation example expands rune names inside double quotes, while the inline-code example preserves an unwanted backslash inside single quotes. Both examples need corrected shell quoting before they are reliable installation/use guidance. Source ↗
  • Terraform Test: The pinned main guide documents terraform test -no-cleanup and a positional test-file invocation, and treats -filter as a run-block name. The current official Terraform test command contract documents option-only invocation and -filter=testfile. Correct upstream commands before recommending this execution guide. Source ↗
  • Security Code Review: The pinned Do Not Flag list excludes code paths that require prior authentication. That blanket exclusion can hide authenticated authorization and privilege-escalation vulnerabilities, contradicting the stated broad security-review scope. The separate Django access review explicitly handles these cases. Source ↗
  • libFuzzer: The recommended byte-to-uint32 harness uses potentially unaligned pointer casts, so a harness can generate its own undefined behavior. The troubleshooting guidance also conflates ASan virtual address reservation with RSS and recommends disabling the RSS guard for out-of-memory. Require corrected harness and resource-limit guidance. Source ↗
  • Prometheus cardinality troubleshooter: Its series-per-instance example uses sum of sample values rather than count of series, and it asserts a fixed 14x benefit per removed label plus conclusive causation from temporal correlation. Those statements can misdirect operational diagnosis; cardinality depends on distinct label combinations. Source ↗
  • Hugging Face Local Models at the reviewed revision: Withheld: the llama.cpp source-build example invokes make, while current upstream build guidance uses CMake. The shown installation path was not accepted as reliable. Source ↗
  • Hugging Face Tool Builder at the reviewed revision: Withheld after reviewing bundled scripts: model-card frontmatter parsing uses fragile string splitting, the private-card helper omits authentication on the card fetch, and the paper-search helper does not perform its claimed paper filter. Source ↗
  • Anthropic MCP Builder bundled evaluation harness: Withheld as a complete installable package: the evaluation loop handles only the first tool call in a multi-call model response and attempts direct JSON serialization of MCP content objects. It also exposes every discovered tool without a read-only enforcement layer or call budget; the planning guide alone does not resolve these runtime concerns. Source ↗
  • clickhouse-best-practices: At ClickHouse/agent-skills commit 356a8c1b9a7392adb389a132f5ad3fec38532a03 the skill says it contains 31 rules and later says a schema review covers all 28 rules. Two rule files were opened and were not the reason for exclusion. The Apache copyright appendix at that commit is the unfilled template. Source ↗
  • langfuse: At langfuse/skills commit b86532b77f16f1e27c7cffa32bb632aa3cd9a352 the description says to query or modify Langfuse data and to invoke the skill for any AI engineering task even when Langfuse is not mentioned. The allowed-tools list is narrower: WebFetch on langfuse.com, curl to langfuse.com, and langfuse-cli api limited to schema, help, list, and get. The reference workflows were not opened. Source ↗
  • building-pydantic-ai-agents: The Pydantic AI building skill at pydantic/skills commit 238d97102650c1caa51f35027aee13c469c59542 prints specific run results, including a RunUsage cost, that were not executed. It also shows logfire.instrument_httpx(capture_all=True), which the same file says can capture prompts and secrets. The reference set the procedure routes to was not line-reviewed. Source ↗
  • langsmith-trace: At langchain-ai/langsmith-skills commit 1fb52e8 the setup calls passing --api-key on the command line the preferred method and repeats --api-key $LANGSMITH_API_KEY in its examples, while the sibling langsmith-evaluator at the same commit says not to place keys in commands or logs. It installs the CLI by piping an unpinned main-branch script to sh. Source ↗
  • langsmith-dataset: At commit 1fb52e8 the file prefers --api-key on the command line, as langsmith-trace does, and tells non-interactive runs to pass --yes, which skips the CLI confirmation before dataset deletes and overwrites. Source ↗
  • liteparse: At run-llama/llamaparse-agent-skills commit 2dcef7c the ranked-search fallback runs ./.claude/skills/effective-liteparse/scripts/search.py, but the skill directory is skills/liteparse, so the documented helper path does not match an installed copy. Its parse example writes to the fixed path /tmp/doc.txt. Source ↗
  • llamaparse: At commit 2dcef7c the image-download step sends LLAMA_CLOUD_API_KEY as a Bearer header to each presigned image URL, passing the account key to whatever host that URL names. The step heading says it requires httpx while the code uses fetch. Source ↗
  • defending-code-reference-harness skills: The README at commit d3bea6b states the repository is not maintained and is not accepting contributions. Its skills (threat-model, vuln-scan, triage, patch and others) are written for the repository's own harness and are run by opening that checkout in Claude Code. Source ↗
  • transformers-to-mlx: No LICENSE file exists in the repository tree at commit 1800e73, and neither the README nor SKILL.md names a licence, so reuse and redistribution terms are not established. Source ↗
  • openrouter-models: No LICENSE file exists in OpenRouterTeam/skills at commit 33597a4 and the README names no licence, so reuse terms for the definition are not established. Source ↗
  • openrouter-benchmarks: No LICENSE file exists at commit 33597a4. The skill also reports Artificial Analysis and Design Arena rankings through OpenRouter's endpoint; this project treats Artificial Analysis redistribution as unresolved. Source ↗
  • azure-aigateway: At microsoft/azure-skills commit 5aa6e04 the quick-reference table states 60-80% cost savings for semantic caching with no source, and the troubleshooting table answers no cache hits by lowering score-threshold to 0.7 without noting that a looser match can return a cached answer to a different prompt. Source ↗
  • huggingface-best: At huggingface/skills commit c3ff942 the error handling falls back to "the curated fallback table above", which the file does not contain, and to models ranked by trendingScore when leaderboards fail. Device budgets use a parameters-per-GB rule of thumb that leaves out context and KV-cache memory. Source ↗
  • databricks-mlflow-evaluation: At databricks/databricks-agent-skills commit f774134 the LICENSE is the Databricks License, which limits use to connecting with Databricks Services and lets Databricks terminate it on notice. The skill names the Apache-2.0 mlflow/skills repository as the canonical source for the generic workflow; build-a-scorer from that repository is listed instead. Source ↗
  • vllm-prefix-cache-bench: Its online path repeats vllm bench serve, which is listed. Its offline path at commit c996234 runs benchmarks/benchmark_prefix_caching.py from a vLLM source checkout and downloads an unfiltered ShareGPT dump from an anonymous Hugging Face account. Source ↗
  • run-models: At replicate/skills commit 2f36e41 the file tells the agent to fire off multiple predictions concurrently without waiting, and has no step that confirms cost before billed predictions run. compare-models from the same repository is listed. Source ↗
  • Hugging Face Local Models: Deferred: the reviewed quick start uses a plain make build, while current llama.cpp build documentation specifies CMake. The setup path needs reconciliation before inclusion. Source ↗
  • Trail of Bits Devcontainer Setup: Not selected as a general isolated-workspace recommendation: the inspected post-install script sets bypassPermissions, the template forwards host credentials, and the container receives NET_ADMIN and NET_RAW. These defaults require a narrower threat model and explicit adaptation. Source ↗
  • MotherDuck DuckDB SQL: Deferred: the syntax reference shows date_sub(date, interval), but the official DuckDB signature is date_sub(part, startdate, enddate). A SQL repair reference needs this example corrected before inclusion. Source ↗
  • Databricks agent skills: LICENSE at databricks/databricks-agent-skills commit e7a6f218af2fc25b4b896d1e8a0cb4858b20cb75 is a Databricks License whose agreement definition points at the Master Cloud Services Agreement. It is not MIT, Apache-2.0, BSD, ISC, CC-BY-4.0, CC-BY-SA-4.0, or MPL-2.0. Source ↗
  • Trigger.dev skills: README at triggerdotdev/skills commit e0f9f87153d1502fd9f6298e2eb1d58cfd7f9871 says the repository is an automatic mirror of skills maintained in the Trigger.dev monorepo. LICENSE and LICENSE.md at that commit return HTTP 404, so this repository has no licence file to review. Source ↗
  • Angular skills: README at angular/skills commit d5cdb148f23a74cd6788d983c931f3b3a9860e6f exists and names angular-developer and angular-new-app. LICENSE, LICENSE.md, and COPYING at that commit return HTTP 404. Skill bodies were not fetched because no licence file was present. Source ↗
  • prisma-postgres: prisma-postgres/SKILL.md at prisma/skills commit be16a8740d01363d13552b317042431ee8dfc581 says it is a deprecated compatibility name for prisma-postgres-setup and tells the reader to load that other skill. It does not contain the setup recipes. The repository LICENSE is MIT; the exclusion is the deprecated alias, not the licence. Source ↗
  • better-auth skills: At better-auth/skills commit 20c9e88a5c007461a703f1c213572b073196113e, root README.md and LICENSE both return HTTP 404. No skill body was reviewed. Source ↗
  • dbt command runner: The preferences say always use --select and never run the whole project without approval, and always use --quiet. The static-analysis examples are dbt run --static-analysis=off and dbt run --static-analysis=unsafe, with neither --select nor --quiet. Source ↗
  • Grafana database observability: The description says the skill creates a least-privilege monitoring user. The PostgreSQL example sets the password to the literal secret, and the MySQL example creates grafana_monitoring@% with SELECT, PROCESS, and REPLICATION CLIENT on *.*, which the same file still calls least privilege. Source ↗
  • AppInsights instrumentation guide: The file says it is reference material and that azure-prepare orchestrates actual changes, and it sends any request to change or add components to azure-prepare. The guidelines then say to manually instrument the app by creating the Application Insights resource and updating the application's code. Source ↗
  • Azure app onboard: The quick reference says the pipeline is self-contained and makes no external skill calls. Step 3 of the same workflow requires invoking the separate azure-app-onboard-prereq skill, while the deploy note forbids invoking azure-deploy. The file therefore both prohibits external skill calls and requires one. Source ↗
  • Azure app onboard prepare: Step 9 writes prepare-plan.json and step 10 returns the orchestrator summary, while step 11 says the four plan checks must all pass before writing. The same step 11 row also says to fix the plan inline on failure, which writes after a failed check. Source ↗
  • Azure app onboard scaffold: The file boundary says the scaffold only writes files and runs no install or build commands. The same file then requires az bicep build and tells the main thread to run the conformance script when a shell exists. Source ↗
  • Microsoft Foundry: The file contradicts itself: Tool Usage Conventions say to reference official documentation URLs instead of embedding CLI command syntax, but the network-isolation section embeds an az cognitiveservices account show command. Source ↗
  • Capacity Discovery: The description says a quota-increase request is out of scope and should be sent to the Azure Portal, but the no-quota path says to hand that same increase request to the quota skill. Source ↗
  • Deploy Model to Optimal Region: The description says custom capacity configuration belongs in the customize skill, but Advanced Usage still shows a deployment create that sets --sku-capacity. Source ↗
  • Azure Prepare: The description says to use the skill only for azd and not for non-azd deployments, but Phase 1 says to choose AZD, AZCLI, Bicep, or Terraform. The description also lists creating and deploying to Azure, while the body says this skill must not run azd up, azd deploy, or terraform apply. Rule 1 says the plan skeleton is the very first action, and Step 0 says invoking a specialized skill is the mandatory first action. Source ↗
  • Azure Quotas: The scripts section says to use check-quota.ps1 or check-quota.sh instead of constructing commands manually for usage queries and capacity calculation. Workflow 2 then constructs az quota show and az quota usage show by hand to calculate available capacity across regions. Source ↗
  • Azure Reliability: The sample re-assessment leaves health probes partial and then says multi-region is offered next. Step 3 says not to ask about multi-region unless zone-redundant compute, ZRS or GZRS storage, and health probes are all on, and not to ask when any core feature is not on. Source ↗
  • IRQL incident response queries: The description says the skill translates natural language hunting questions into IRQL pipelines and lists failed logins as a WHEN trigger, but the body says it is not a natural-language-to-IRQL converter and not to activate for a generic failed-logins query unless the user explicitly asks for IRQL. Source ↗
  • Azure Cosmos DB NoSQL Python: The Security Requirements say to use DefaultAzureCredential in Azure and never store keys in code, and the later Security list says zero secrets in code. The next requirement says to hardcode the well-known emulator key for local development. The emulator samples do not hardcode a key: they pass COSMOS_KEY or settings.cosmos_key, after the authentication block said to avoid account and API keys in every code sample. Source ↗
  • Azure Bot Service SDK: The authentication rules say DefaultAzureCredential works in Azure with no code change and that production should constrain that same chain with AZURE_TOKEN_CREDENTIALS, but best practice 3 says to use DefaultAzureCredential only for local code and a specific token credential in Azure. Source ↗
  • Azure Fabric management SDK: The authentication rules say DefaultAzureCredential works in Azure with no code change and that production should constrain that same chain with AZURE_TOKEN_CREDENTIALS, but best practice 3 says to use DefaultAzureCredential only for local code and a specific token credential in Azure. Source ↗
  • Azure Monitor ingestion SDK: The authentication rules say DefaultAzureCredential works in Azure with no code change and that production should constrain that same chain with AZURE_TOKEN_CREDENTIALS, but best practice 3 says to use DefaultAzureCredential only for local code and a specific token credential in Azure. Source ↗
  • Grafana skill authoring: The score-variance section says to ship only on three consecutive local scores of 100. The later fix section says not to stop until all five local runs hit 100, and it points back to that score-variance section as the reason. Those are different stop conditions for the same local review. Source ↗
  • Grafana Cloud k6 manage: The opening rule says the only header set by hand on Loki queries, browser screenshot and file fetches, and Tempo queries is X-K6TestRun-Id, and that any other header is overwritten or causes conflicts. The screenshot and file section then requires a hand-set Content-Type on the pre-signed URL request and says that header matters. Source ↗
  • Migrate a dbt project across platforms: Success criteria and Step 5 require dbtf compile to finish with 0 warnings and say warnings are blockers. The same step says adapter warnings from profiles.yml are not blockers. Step 4 says to remove platform-specific configs such as snowflake_warehouse when switching targets, while Known Limitations says not to pre-remove those keys until Fusion flags them. Source ↗
  • SRED project organizer: Step 4 creates a private Notion document called "SRED Project Descriptions", and Step 6 links to that plural title. Step 5 says each project document is a child of the "SRED Project Description" document created in Step 4. Source ↗
  • Writing Hookify Rules: plugins/hookify/README.md ends with the heading License and the line MIT License. plugins/hookify/LICENSE is the Apache License, Version 2.0, and it matches the repository LICENSE. The plugin README and the licence file do not name the same licence. Source ↗
  • Math Olympiad Solver: The file says to launch 8-12 attempt agents, calls the tight-budget workflow 8 solvers, and sets the model-tier table to 8, 4, and 3 solvers for Haiku, Sonnet, and Opus. Olympiad verification is called 5-pass, while that table gives Haiku 3 verify passes. Abstention is both up to 3 revise cycles and, in the same table, 2, 3, or 4 revise failures by model. The description says pure reasoning with no tools, while deep mode allows local Bash and Python. Source ↗
  • Cairo vulnerability scanner: Section 6 defines pattern 4 as overconstrained L1-to-L2 interaction. A later rationalization says the boundary between felt252 and the sized integer types is where pattern 4 lives. The file assigns pattern 4 two different subjects. Source ↗
  • Cosmos vulnerability scanner: The embedded agent prompt says to report every pattern in the reference file and not to skip any. The completeness check expects 8 core patterns from sections 1–9 because it excludes section 8 as legacy-only, while the agent table assigns that scanner sections 1–9. Reporting section 8 misses the expected total, and omitting it violates the prompt. Source ↗
  • Hugging Face datasets: Canonical definition already selected in both inventories as huggingface-huggingface-datasets. A newer pin or shorter ID would not be an acquisition. Source ↗
  • Hugging Face community evals: Canonical definition already selected in both inventories as huggingface-huggingface-community-evals. Source ↗
  • Docker Sandboxes network and credentials: Missing from the older integration snapshot but already present in the shared-root catalogue as docker-sandboxes-network-credentials. Do not duplicate under a new ID. Source ↗
  • Cloudflare sandbox stable: Already present in the shared-root catalogue as cloudflare-sandbox-stable; not a new canonical resource. Source ↗
  • Google GKE Batch and HPC: Not accepted in this packet. The inspected definition applies an unpinned Kueue latest-release manifest and an MPI Operator master manifest, creates node pools, and names MCP tools without binding their implementation. Those external executable configurations and their permissions were not fully pinned and inspected. This is a review-coverage deferral, not a finding that the publisher is unsafe. Source ↗
  • Dagster expert skill: Publisher tree discovery only. The large reference surface includes deployment, secret, run-launch and deletion operations, and plugin MCP configuration. Definition, licence and the complete referenced execution boundary were not inspected; not promoted on the strength of the repository name. Source ↗
  • example-command: Reference implementation and example placeholder; not a production security skill. Source ↗
  • example-skill: Reference example template; not a standalone operational skill. Source ↗
  • panel-review: Test fixture in review-panel test suite, not a standalone skill. Source ↗
  • review-pr: Test fixture in code-improver tests directory, not an operational skill. Source ↗
  • Ollama serve: No publisher skill to review at the inspected revision: the repository contains no SKILL.md and no skills, agents or claude directories, so there is no Ollama-published skill covering local model serving. Source ↗
  • Phoenix phoenix-evals: Withheld at the inspected revision: the skill frontmatter claims Apache-2.0 while the repository root LICENSE is Elastic License 2.0, which is non-sublicensable and limits hosted use, so the skill cannot be recorded as Apache-2.0 redistributable pending upstream clarification. Source ↗
  • Ragas evals: No publisher skill to review at the inspected revision: the repository contains no SKILL.md and no skills directory, only contributor slash commands and editor rules, so any Ragas skill found elsewhere is third-party rather than publisher-served. Source ↗
  • Snowflake cortex-code-skills: Withheld as one candidate: the collection is community-contributed for educational and reference use with no correctness guarantee, and it carries no single licence, with per-skill files varying between Apache-2.0 and a service-gated Snowflake Skills License that bars extraction, reproduction, derivatives and distribution. Source ↗
  • FastAPI Router: Overlaps fastapi-fastapi, FastAPI's own skill, which already covers router-level prefix, tags and shared dependencies, Annotated dependencies, and return types versus response_model. This 72-line file does the same job (routers with auth dependencies, response models and status codes) through a template and integration steps that assume a src/backend/app/ layout. Its samples declare response_model together with the same return type, which the FastAPI skill reserves for when the two differ. Its template's update and delete endpoints are documented as requiring authentication and ownership, but the update's ownership check is commented out and the delete has none. Source ↗
  • Pydantic Models: Overlaps fastapi-fastapi, whose skill is for working with FastAPI APIs and Pydantic models and sets its own rules for model fields. This 71-line file's pattern is the same API request and response schema work, split into Base, Create, Update, Response and InDB variants, with integration steps that assume a src/backend/app/models/ layout and matching TypeScript types. Its required fields are written Field(..., alias=...), which the FastAPI skill says not to do. Source ↗
  • Azure Identity library for Java: Withheld pending correction: the Token Caching section (SKILL.md lines 244-258) says "Enable persistent token caching for better performance", but its code calls DefaultAzureCredentialBuilder().enableAccountIdentifierLogging() under the comment "Enable token caching (in-memory by default)" (lines 249-251). The Logging section shows the same call labelled "Log account info" (line 305). A reader who follows the caching advice with that builder turns on what the same file's Logging section calls account-info logging. Source ↗
  • Azure Key Vault Keys library for Rust: Teaches an unsafe data encryption key. The envelope-encryption sample (SKILL.md lines 133-184) generates its "symmetric data encryption key (DEK)" as random::<u32>().to_le_bytes() (lines 158-159): 4 bytes (32 bits), far below a safe size for a key that protects data, so code copied from the sample generates data keys that weak. The value comes from the rand crate (line 144), which the install line (line 28) does not add. The same section says "the private key never leaves the HSM" (line 135) but creates its key encryption key as KeyType::Rsa (line 148), not KeyType::RsaHsm. Source ↗
  • Azure Key Vault Secrets (Java): Teaches unsafe handling of the secrets it manages. Four samples print secret values to standard output: "Secret value: " + value (SKILL.md line 88), the first five characters of every enabled secret while listing (line 131), the async get (line 221) and the error-handling read (line 290). Samples also hardcode literal secret values such as "P@ssw0rd123!" (line 62), "sk_live_abc123xyz" (line 68) and "CertP@ss!" (line 361). The Purge block (lines 177-187) purges old-secret, then calls getDeletedSecret under the comment "Get deleted secret info first" and purges it a second time. Source ↗
  • let-fate-decide: Not selected: the skill picks a direction by drawing a random 12-house Tarot spread, and its description says to prefer the draw over asking clarifying questions when the user sounds casual. Its own SKILL.md says that in security and correctness work the reading is never sufficient by itself, and that cards can suggest what to inspect while only evidence can dismiss a concern. The technical-context lenses translate a reading into which evidence to gather next; the skill produces none of that evidence itself. Source ↗
  • Azure AI Content Understanding SDK for Python: Withheld pending correction: the main definition and its own reference disagree on the custom-analyzer API. SKILL.md (lines 169-211) creates the analyzer with client.create_analyzer(analyzer_id=..., analyzer={...}) and reads result.fields["vendor_name"]. references/non-hero-scenarios.md (lines 62-111) calls client.begin_create_analyzer(analyzer_id=..., resource=ContentAnalyzer(...)), which it says returns an LRO poller to wait on, and reads content.fields["vendor_name"].value_string from result.contents[0]. They also name different base classes for DocumentContent and AudioVisualContent: MediaContent at SKILL.md line 263, AnalysisContent at line 163 of the reference. Neither file says which is current. Source ↗

04 / PRACTICAL ANSWERS

Before you add it.

What is an AI skill?

A skill packages instructions for a repeatable task. It may also include scripts, reference documents or other files. The host decides when and how to use it; a skill does not replace the underlying model or guarantee its output.

What does “source reviewed” mean here?

We examined the material named in the review record and wrote our own reasons, findings and limitations. This is not an installation test, a security certification or a measured improvement in model performance. Untested aspects are listed on each detail page.

How do I install a listed skill?

Open its review and use the documented command or upstream setup guide. Skills installed through the Skills CLI can include supporting files, so copying only SKILL.md may be incomplete. Choose the intended host and project scope, and inspect the permissions before running the skill.

Are the skills ranked by downloads?

No. The default order is alphabetical. These are editorial selections with named reasons and trade-offs, drawn from multiple publishers and sources. Installation counts and star counts do not establish that a skill is right for your task.

Evidence & Ask