Differential review skill for Claude: security-review a PR or diff

Makes Claude security-review a PR, commit or diff using git history, caller counts and test coverage, and write a report file.

Installs: 9.4K; on skills.sh
Stars: 7.9K; on GitHub
Files: 6; besides SKILL.md
Updated: Oct 2026; last repo commit

Claude works through a set of phases: it builds baseline context, triages the diff by risk, reads the changed code with git blame, checks test coverage, counts callers to size the blast radius, digs into high-risk changes and models an attacker. The depth scales with codebase size, and the last step is a markdown report.

The plugin also adds a /differential-review:diff-review command and an adversarial-modeler agent for high-risk changes. It runs git and shell commands in your repo, and the methodology includes checking out the baseline commit.

Good for

Security reviewers checking pull requests for new vulnerabilities
Maintainers who suspect a change re-introduces an old bug
Teams that want to see which changed code has no test

Install differential-review

Claude Code

Copy this prompt.

Install the differential-review skill in Claude Code. Run this command:

/plugin marketplace add trailofbits/skills
/plugin install differential-review@trailofbits

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Pick one route. The plugin route installs the differential-review plugin from the Trail of Bits marketplace: 1 skill, 1 command (/differential-review:diff-review) and 1 agent (adversarial-modeler), with no hooks or MCP config files in the plugin folder. The skills CLI and folder routes install just the skill, without the command or agent. The skill runs git and shell commands, so there is no Claude app upload route.

Built from the package data we hold. Not yet tested by us. Source.

What's in the skill

When it loads

Performs security-focused differential review of code changes. Adapts analysis depth to codebase size, uses git blame for context, calculates blast radius by counting callers, checks test coverage of modified code, and generates a markdown report. Use when reviewing a PR, commit, or diff for security vulnerabilities, checking whether a change re-introduces a previously fixed bug, asking what else a change could break, or finding which modified code has no test covering it.

Sections in SKILL.md

Core Principles
Rationalizations (Do Not Skip)
Quick Reference
Codebase Size Strategy
Risk Level Triggers
Workflow Overview
Decision Tree
Agents
Quality Checklist
Integration
Example Usage
Quick Triage (Small PR)
Standard Review (Medium Codebase)
Deep Audit (Large, Critical Change)
When NOT to Use This Skill
Red Flags (Stop and Investigate)

Key facts

Author: Omar Inuwa, for Trail of Bits (named in the plugin README)
License: CC BY-SA 4.0 (repo README)
Installs on skills.sh: 9,392 (Nov 2, 2026)
GitHub stars: 7,937 for the whole trailofbits/skills repo (Nov 2, 2026)
Last repo commit: Oct 28, 2026
Audits on skills.sh: Gen Agent Trust Hub: pass, Socket: pass, Snyk: pass
What it includes: 11 files in the plugin folder including SKILL.md: 4 reference guides, an agent, a command, a manifest, a README, an icon and an openai.yaml (no scripts)
Part of: The differential-review plugin: 1 skill, 1 command and 1 agent
Plugin version: 1.1.4 (plugin.json)

What it includes

Builds baseline context, including a checkout of the baseline commit
If the audit-context-building skill is available, methodology.md runs git checkout of the baseline commit and then that skill on the baseline code.
Try: “Review the security implications of this PR: git diff main..feature/auth-changes”
Runs code · Pre-analysis
Sizes the change and scores files by risk
Uses git diff --stat, git log and, for PRs, gh pr view. SMALL, MEDIUM and LARGE codebases get DEEP, FOCUSED or SURGICAL treatment.
Runs code · Phase 0: Triage
Reads diffs and git blame, looks for regressions
Uses git log -S and git blame on removed code, especially code from security or fix commits.
Runs code · Phase 1: Changed code
Finds changed code that no test covers
Compares changed files with test files and greps for tests that name the changed functions.
Runs code · Phase 2: Test coverage
Counts callers of changed code
Counts callers with grep; SKILL.md flags 50 or more callers with a high-risk change as an immediate escalation.
Runs code · Phase 3: Blast radius
Five Whys root cause on high-risk changes
Used for HIGH RISK changes, optionally with audit-context-building.
Read · Phase 4: Deep context
Attacker model and exploit scenarios
Run in the main flow, or delegate to the differential-review:adversarial-modeler agent, whose tools are Read, Grep, Glob and Bash.
Runs code · Phase 5: Adversarial analysis
Writes <PROJECT>_DIFFERENTIAL_REVIEW_<DATE>.md
reporting.md lists the output locations in order: the current working directory, your Desktop, then ~/.claude/skills/differential-review/output/.
Write · Phase 6: Report
Command: <pr-url|commit-sha|diff-path> [--baseline <ref>]
The plugin command passes the target and optional baseline to the skill.
Runs code · /differential-review:diff-review

From the official docs

Permissions

Medium risk

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

It can check out another commit
methodology.md has Claude run git checkout of the baseline commit before building context, which changes your working tree. Commit or stash your work first, or run it in a separate clone or worktree. github.com
It writes a report file outside the repo if needed
reporting.md tries the current directory first, then your Desktop, then ~/.claude/skills/differential-review/output/. Check where the file landed. github.com
The plugin adds a command and an agent
The plugin includes the /differential-review:diff-review command and an adversarial-modeler agent with Read, Grep, Glob and Bash tools. It also lists related skills (audit-context-building, issue-writer) that it hands off to if installed. github.com
The skill lists patterns for attack scenarios
adversarial.md builds exploit scenarios for the code under review. Use it on code you are authorized to review. github.com

Security advisories

No published security advisories as of Nov 2, 2026.

Technical details

Turning on read-only	No read-only mode. SKILL.md allows Read Write Grep Glob Bash.
Sign-in and keys	No key or account. gh pr view needs the GitHub CLI signed in to read a PR.
Scopes	Runs git, gh and grep commands in your repository through Bash Can run git checkout of a baseline commit Writes a markdown report file to your project folder, Desktop or ~/.claude

Questions

What does the differential-review skill do?

It makes Claude do a security-focused review of a PR, commit or diff. It uses git history to spot re-introduced bugs, counts callers to size the impact, checks which changed code has no test, and writes a markdown report.

How do I install it?

Use the install box at the top of this page. The plugin route also adds a command and an agent; the skills CLI route installs only the skill.

Does it change my repository?

It writes a report file. methodology.md also has Claude run git checkout of the baseline commit, which changes your working tree, so use a clean checkout or a worktree.

How do I run it?

Ask in plain language, for example to review the security implications of a PR or a git diff range, or use the plugin command /differential-review:diff-review with a PR URL, commit SHA or diff path and an optional --baseline ref.

Is it only for smart contracts?

SKILL.md describes it as security review of PRs, commits and diffs. Its pattern commands and examples lean on Solidity (.sol) files, and the size count also covers .rs, .go and .ts files, so check how well it fits your language.

When should I not use it?

SKILL.md lists greenfield code with no baseline, documentation-only changes, formatting or linting changes, and cases where you only want a quick summary.

Is it safe?

It runs git, gh and grep through Bash and writes a report file, so run it in a clone you can discard. The plugin also adds an agent that has Bash access. Use it only on code you are allowed to review.

Who made it?

The plugin README names Omar Inuwa as author, and it ships in the Trail of Bits skills repository, which Trail of Bits publishes under CC BY-SA 4.0.

Information

updated Nov 2, 2026

Repository: trailofbits/skills
Folder: plugins/differential-review/skills/differential-review

trailofbits/skills skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 8

differential-review SKILL.md · github.com
differential-review plugin README · github.com
methodology.md · github.com
reporting.md · github.com
Trail of Bits marketplace.json · github.com
adversarial.md · github.com
adversarial-modeler agent · github.com
patterns.md · github.com