--- title: "Adversarial modeler: review, role & definition · Undominated.ai" canonical: https://undominated.ai/agents/trailofbits-adversarial-modeler/ description: "Pins Trail of Bits' differential-review agent that models high-risk changes and says not to run on low or medium risk diffs." --- # Adversarial modeler: review, role & definition · Undominated.ai > Pins Trail of Bits' differential-review agent that models high-risk changes and says not to run on low or medium risk diffs. [← Explore all agents](/agents/) SECURITY REVIEW / Trail of Bits # Adversarial modeler Pins Trail of Bits' differential-review agent that models high-risk changes and says not to run on low or medium risk diffs. [Use this definition ↓](#setup)[Original source ↗](https://github.com/trailofbits/skills/tree/8e54050ae63c2f62224c35b1616efe71865b4777/plugins/differential-review/agents) SOURCE REVIEW Reviewed 2026-10-14 Evidence 5 linked sources Publisher Trail of Bits Licence [CC-BY-SA-4.0 ↗](https://github.com/trailofbits/skills/blob/8e54050ae63c2f62224c35b1616efe71865b4777/LICENSE) Revision 8e54050ae63c [Read what was—and wasn’t—checked ↓](#review) “Activates only for high-risk changes, requires a concrete impact for each finding, and declares Read, Grep, Glob, and Bash.” [Trail of Bits · upstream description ↗](https://github.com/trailofbits/skills/blob/8e54050ae63c2f62224c35b1616efe71865b4777/plugins/differential-review/agents/adversarial-modeler.md) Our analysis follows below. 01 / THE REASONING ## Why this made the selection. - The catalogue has the differential-review skill and not this adversarial-modeler agent. - The file limits itself to high-risk changes and says greenfield code without a baseline is out of scope. ### A good fit for - A high-risk diff that already has a baseline, when the host can load this Trail of Bits plugin - Supply a high-risk diff, baseline and explicit attacker model. - Receive a reachable attack scenario with code locations, preconditions and concrete impact. ### Weigh up before choosing - This review did not execute the resource. - The agent points at adversarial.md, patterns.md, and methodology.md under the skill directory. Those files were not opened. - Frontmatter includes Bash. This review did not run any command the agent might run. - The plugin directory has no LICENSE file at this commit. The grant used here is the repository root CC-BY-SA-4.0 text, which matches the README. - The source requests exact impact quantities but they must be computed from evidence, never invented; Bash enables execution despite the source-review workflow. 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/differential-review/agents/adversarial-modeler.md - README.md - LICENSE - Round 2: pinned definition, output contract and permission/dependency boundaries reread on 2026-10-14 - plugins/differential-review/skills/differential-review/adversarial.md (selected input/output and dependency-contract sections; not a full implementation audit) - plugins/differential-review/skills/differential-review/methodology.md (selected input/output and dependency-contract sections; not a full implementation audit) ### Our findings - The agent says every finding needs a concrete impact and that vague warnings are not findings. - It says to activate for high-risk changes such as auth, cryptography, external calls, value transfer, weaker validation, or access-control changes, and not for low or medium risk, docs, tests, or formatting. - Its tools are Read, Grep, Glob, and Bash. The plugin LICENSE path returns 404 at this commit; the root license is CC-BY-SA-4.0. - The role is scoped to high-risk differential review and rejects greenfield work lacking a baseline. - It cross-checks changed behavior against existing trust boundaries, invariants and git history before calling something a regression. ### Not established by this review - The resource was not executed and its tools were not called. - The linked methodology files were not opened. - No repository diff was reviewed with this agent. The review applies to the material and revision named here. A newer upstream release can change its behavior. 03 / PUT IT TO WORK ## Use the role in your project. [Upstream setup instructions ↗](https://github.com/trailofbits/skills/blob/8e54050ae63c2f62224c35b1616efe71865b4777/README.md) DOCUMENTED COMMAND /plugin marketplace add trailofbits/skills Copy command ↗ Copying does not execute this command. It may retrieve a newer version than the reviewed source. - The README's Claude Code install starts with the command above and then /plugin menu. - The marketplace table names differential-review as the security-focused diff plugin. This review did not install it. - Codex instructions in the same README use codex plugin marketplace add trailofbits/skills and a placeholder plugin name. - Supply the matching differential-review references, bound git/source access and independently reproduce a claimed exploit in an authorized test environment. ### Before you start - Claude Code or Codex marketplace support - A git baseline if the agent is used as written ### Compatibility Claude Code · OpenAI Codex ### Review agent with Bash declared and a high-risk-only scope - Read and search the tree the host provides - Run shell commands if the host honors the Bash tool - Does not authorize deploying an exploit ### Cost model CC-BY-SA-4.0; the agent was not run 04 / FOLLOW THE EVIDENCE ## The source trail. Our notes are separate from the original resource. Check upstream before adopting a new version. - [adversarial-modeler ↗](https://github.com/trailofbits/skills/blob/8e54050ae63c2f62224c35b1616efe71865b4777/plugins/differential-review/agents/adversarial-modeler.md) Checked 2026-10-14 https://github.com/trailofbits/skills/blob/8e54050ae63c2f62224c35b1616efe71865b4777/plugins/differential-review/agents/adversarial-modeler.md Supports: scope, tool list, and when not to use - [Repository README ↗](https://github.com/trailofbits/skills/blob/8e54050ae63c2f62224c35b1616efe71865b4777/README.md) Checked 2026-10-14 https://github.com/trailofbits/skills/blob/8e54050ae63c2f62224c35b1616efe71865b4777/README.md Supports: marketplace install line and CC-BY-SA-4.0 statement - [Repository LICENSE ↗](https://github.com/trailofbits/skills/blob/8e54050ae63c2f62224c35b1616efe71865b4777/LICENSE) Checked 2026-10-14 https://github.com/trailofbits/skills/blob/8e54050ae63c2f62224c35b1616efe71865b4777/LICENSE Supports: Attribution-ShareAlike 4.0 International - [Pinned inspected source: plugins/differential-review/skills/differential-review/adversarial.md ↗](https://github.com/trailofbits/skills/blob/8e54050ae63c2f62224c35b1616efe71865b4777/plugins/differential-review/skills/differential-review/adversarial.md) Checked 2026-10-14 https://github.com/trailofbits/skills/blob/8e54050ae63c2f62224c35b1616efe71865b4777/plugins/differential-review/skills/differential-review/adversarial.md Supports: Dependency setup and source-review boundary - [Pinned inspected source: plugins/differential-review/skills/differential-review/methodology.md ↗](https://github.com/trailofbits/skills/blob/8e54050ae63c2f62224c35b1616efe71865b4777/plugins/differential-review/skills/differential-review/methodology.md) Checked 2026-10-14 https://github.com/trailofbits/skills/blob/8e54050ae63c2f62224c35b1616efe71865b4777/plugins/differential-review/skills/differential-review/methodology.md Supports: Dependency setup and source-review boundary KEEP COMPARING ## Other approaches to consider. Related by category or shared topics. These are alternatives to inspect, not a measured quality order. [### Exploitability verifier ↗ Pins the fp-check phase-2 agent that stays read-only while deciding whether a suspected bug is exploitable, and that skips a proof the bug class does not need.](/agents/trailofbits-exploitability-verifier/)[### Rust review FP judge ↗ Pins the rust-review judge that runs after dedup, aborts when it cannot obtain a finding list, and must not invent the SARIF file.](/agents/trailofbits-rust-review-fp-judge/)[### Rust review worker ↗ Pins the rust-review worker that aborts a malformed or primer spawn, writes findings only inside its scope root, and does not filter them by severity.](/agents/trailofbits-rust-review-worker/) [AI Tools ↗](/tools/)[Skills ↗](/skills/)[Agents ↗](/agents/)[MCP Servers ↗](/mcp-servers/)[Workflows ↗](/workflows/) ## Continue your investigation - [Find reusable instructions](/skills/) - [Inspect connections](/mcp-servers/) - [Build a toolkit](/tools/) - [Choose the model](/compare/)