---
title: "Second Opinion: review, setup & limitations · Undominated.ai"
canonical: https://undominated.ai/skills/trailofbits-second-opinion/
description: "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."
---

# Second Opinion: review, setup & limitations · Undominated.ai

> 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.

[← Explore all skills](/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.

 [See setup guidance ↓](#setup)[Original source ↗](https://github.com/trailofbits/skills/tree/8e54050ae63c2f62224c35b1616efe71865b4777/plugins/second-opinion/skills/second-opinion)

SOURCE REVIEW

 Reviewed 2026-10-22
 Evidence 8 linked sources
 Publisher Trail of Bits
 Licence [CC-BY-SA-4.0 ↗](https://raw.githubusercontent.com/trailofbits/skills/8e54050ae63c2f62224c35b1616efe71865b4777/LICENSE)
 Revision 8e54050ae63c
 [Read what was—and wasn’t—checked ↓](#review)

“Gets independent code reviews from Codex or Antigravity for uncommitted changes, branch diffs, and commits. Use when the user requests an external review, a second opinion on code, a codex review, a gemini review, an antigravity review, or /second-opinion.”

 [Trail of Bits · upstream description ↗](https://raw.githubusercontent.com/trailofbits/skills/8e54050ae63c2f62224c35b1616efe71865b4777/plugins/second-opinion/skills/second-opinion/SKILL.md) Our analysis follows below.

01 / THE REASONING

## Why this made the selection.

 - When two providers are chosen, both receive the same captured patch, and the skill summarises their agreements and disagreements "without turning agreement into proof".
- Its provider references hold the reviewers back from writing: Codex runs with --sandbox read-only and --ephemeral, Gemini CLI keeps --approval-mode default, and the skill forbids enabling automatic write approval.
- A nonzero exit, missing output, invalid JSON, denied inspection or a plan request is reported as incomplete work, not as a clean review.

### A good fit for

 - Get an outside model's review of staged, unstaged and untracked changes before committing them
- Compare Codex and Antigravity findings on the same branch patch against a named base
- Ask an external reviewer to look at one commit with a security or performance focus

### Weigh up before choosing

 - Sends 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.
- The reviewer CLIs can read beyond the patch: codex exec runs from the repository root and "can inspect local context within its read-only sandbox", and Antigravity keeps the user's permission settings, so other repository files may also reach OpenAI or Google.
- Only the Codex path runs in a read-only sandbox. The Antigravity reference says headless mode alone does not make the workspace read-only.
- Antigravity and Gemini CLI return prose; only Codex returns findings in the bundled JSON schema.
- The Gemini CLI path is kept for Code Assist Standard/Enterprise, paid Gemini API key and Vertex AI accounts; an UNSUPPORTED_CLIENT response means the account must move to Antigravity.
- A patch larger than a CLI or model limit is not truncated; the skill asks for a narrower scope.
- Default models are gpt-5.6-sol at xhigh effort, gemini-3.1-pro-high and gemini-3.1-pro-preview. The publisher records checking the flags against Codex CLI 0.154.0 and Gemini CLI 0.59.0 on 2026-09-14.
- It reports findings only: no fixes are applied and no review is posted to a remote service.
- Last commit touching plugins/second-opinion/skills/second-opinion at the pinned revision: 2026-09-16.
- Not executed: it was not installed or run.

02 / THE REVIEW RECORD

## What we actually inspected.

Source review has boundaries. A clear record is more useful than a “safe” badge.

### Material inspected

 - plugins/second-opinion/skills/second-opinion/SKILL.md
- plugins/second-opinion/skills/second-opinion/references/codex-invocation.md
- plugins/second-opinion/skills/second-opinion/references/antigravity-invocation.md
- plugins/second-opinion/skills/second-opinion/references/gemini-invocation.md
- plugins/second-opinion/skills/second-opinion/references/review-input.md
- plugins/second-opinion/README.md
- plugins/second-opinion/.claude-plugin/plugin.json
- README.md (installation and licence sections)
- LICENSE

### Our findings

 - The description and plugin.json name Codex and Antigravity; the skill body and plugin README add Gemini CLI for accounts that still support it.
- allowed-tools is Bash Read Glob Grep AskUserQuestion; prompt, output and diagnostic files are created with mktemp outside the checkout.
- Untracked files are added to an uncommitted review with git diff --no-index, and a read error stops preparation instead of dropping the file.
- The shared prompt labels user requirements, project conventions and the diff, and tells the reviewer to treat the diff as data, not instructions.
- Plugin version 1.8.1 removed the automatic Codex MCP registration; the skill says not to launch codex mcp-server or substitute codex app-server.
- Codex severity runs from 0 informational to 3 high and is sorted descending.

### Not established by this review

 - Not executed: it was not installed or run.
- No reviewer CLI was installed, authenticated or called.

The review applies to the material and revision named here. A newer upstream release can change its behavior.

03 / PUT IT TO WORK

## Add a skill to your workflow.

[Upstream setup instructions ↗](https://raw.githubusercontent.com/trailofbits/skills/8e54050ae63c2f62224c35b1616efe71865b4777/plugins/second-opinion/README.md)

DOCUMENTED COMMAND

 /plugin install second-opinion@trailofbits Copy command ↗

Copying does not execute this command. It may retrieve a newer version than the reviewed source.

 - Add the Trail of Bits marketplace in Claude Code with `/plugin marketplace add trailofbits/skills`; this adds the whole marketplace, not this skill.
- Install the plugin with `/plugin install second-opinion@trailofbits`; the second-opinion plugin holds this one skill.
- Install and authenticate at least one reviewer CLI from the plugin README's Setup table, then invoke `/second-opinion:second-opinion` with a provider and scope.

### Before you start

 - OpenAI Codex: `npm i -g @openai/codex` and a supported ChatGPT account or API authentication
- Google Antigravity: the `agy` executable from the Antigravity CLI installation guide
- Gemini CLI, optional: `npm i -g @google/gemini-cli` and a supported Code Assist, API-key or Vertex AI account
- Git

### Compatibility

Claude Code

### Reads local diffs; sends them to the chosen OpenAI or Google reviewer CLI

 - Bash, Read, Glob, Grep and AskUserQuestion (allowed-tools)
- Run git to capture uncommitted, branch or single-commit patches, untracked files included
- Create prompt, output and diagnostic temp files outside the checkout
- Run codex exec, agy or gemini, which send the prompt and patch to OpenAI or Google
- Let the reviewer CLI read further workspace files: Codex inside its read-only sandbox from the repository root, Antigravity under the user's existing permission settings
- Use the credentials those CLIs are already configured with; the Gemini reference says not to print them

### Cost model

The skill text is CC-BY-SA-4.0. Each review runs on the user's own OpenAI account (ChatGPT or API) or Google account (Antigravity, Code Assist, Gemini API key or Vertex AI) and counts against that account's quota or billing.

04 / FOLLOW THE EVIDENCE

## The source trail.

Our notes are separate from the original resource. Check upstream before adopting a new version.

 - [Second Opinion SKILL.md ↗](https://raw.githubusercontent.com/trailofbits/skills/8e54050ae63c2f62224c35b1616efe71865b4777/plugins/second-opinion/skills/second-opinion/SKILL.md) Checked 2026-10-22 https://raw.githubusercontent.com/trailofbits/skills/8e54050ae63c2f62224c35b1616efe71865b4777/plugins/second-opinion/skills/second-opinion/SKILL.md Supports: upstreamDescription, summary, whySelected, bestFor, limitations, access, review.findings
- [Codex invocation reference ↗](https://raw.githubusercontent.com/trailofbits/skills/8e54050ae63c2f62224c35b1616efe71865b4777/plugins/second-opinion/skills/second-opinion/references/codex-invocation.md) Checked 2026-10-22 https://raw.githubusercontent.com/trailofbits/skills/8e54050ae63c2f62224c35b1616efe71865b4777/plugins/second-opinion/skills/second-opinion/references/codex-invocation.md Supports: whySelected, limitations, install.prerequisites, access
- [Antigravity invocation reference ↗](https://raw.githubusercontent.com/trailofbits/skills/8e54050ae63c2f62224c35b1616efe71865b4777/plugins/second-opinion/skills/second-opinion/references/antigravity-invocation.md) Checked 2026-10-22 https://raw.githubusercontent.com/trailofbits/skills/8e54050ae63c2f62224c35b1616efe71865b4777/plugins/second-opinion/skills/second-opinion/references/antigravity-invocation.md Supports: limitations, access
- [Gemini CLI invocation reference ↗](https://raw.githubusercontent.com/trailofbits/skills/8e54050ae63c2f62224c35b1616efe71865b4777/plugins/second-opinion/skills/second-opinion/references/gemini-invocation.md) Checked 2026-10-22 https://raw.githubusercontent.com/trailofbits/skills/8e54050ae63c2f62224c35b1616efe71865b4777/plugins/second-opinion/skills/second-opinion/references/gemini-invocation.md Supports: whySelected, limitations, access, access.cost
- [Shared review input reference ↗](https://raw.githubusercontent.com/trailofbits/skills/8e54050ae63c2f62224c35b1616efe71865b4777/plugins/second-opinion/skills/second-opinion/references/review-input.md) Checked 2026-10-22 https://raw.githubusercontent.com/trailofbits/skills/8e54050ae63c2f62224c35b1616efe71865b4777/plugins/second-opinion/skills/second-opinion/references/review-input.md Supports: review.findings, access, bestFor
- [second-opinion plugin README ↗](https://raw.githubusercontent.com/trailofbits/skills/8e54050ae63c2f62224c35b1616efe71865b4777/plugins/second-opinion/README.md) Checked 2026-10-22 https://raw.githubusercontent.com/trailofbits/skills/8e54050ae63c2f62224c35b1616efe71865b4777/plugins/second-opinion/README.md Supports: install, compatibility, install.prerequisites, access.cost, review.findings
- [trailofbits/skills README ↗](https://raw.githubusercontent.com/trailofbits/skills/8e54050ae63c2f62224c35b1616efe71865b4777/README.md) Checked 2026-10-22 https://raw.githubusercontent.com/trailofbits/skills/8e54050ae63c2f62224c35b1616efe71865b4777/README.md Supports: install.steps, license
- [trailofbits/skills LICENSE ↗](https://raw.githubusercontent.com/trailofbits/skills/8e54050ae63c2f62224c35b1616efe71865b4777/LICENSE) Checked 2026-10-22 https://raw.githubusercontent.com/trailofbits/skills/8e54050ae63c2f62224c35b1616efe71865b4777/LICENSE Supports: license

KEEP COMPARING

## Other approaches to consider.

Related by category or shared topics. These are alternatives to inspect, not a measured quality order.

 [### 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.](/skills/trailofbits-review-walkthrough/)[### Django performance review ↗ Review Django code and report only performance issues the reviewer traced. Zero findings is an allowed result.](/skills/getsentry-django-perf-review/)[### Sentry code review ↗ Reviews a change for runtime, query, security, design, and test gaps, and does not block on style.](/skills/getsentry-code-review/)

 [AI Tools ↗](/tools/)[Skills ↗](/skills/)[Agents ↗](/agents/)[MCP Servers ↗](/mcp-servers/)[Workflows ↗](/workflows/)

## Continue your investigation

 - [Choose a toolkit](/tools/)
- [Inspect execution hosts](/agents/)
- [Check tool access](/mcp-servers/)
- [Check project dependencies](/check-repo/)
