claude-review
Skill15 starsChanged 8 months agoApache-2.0View on GitHub
Spawn a separate Claude Code agent to review code changes made in the current session or branch. Uses the Agent tool with worktree isolation so the reviewer gets an independent read-only copy. Use when the user says "claude review", "self-review", "/claude-review", or when a task plan calls for post-task review.
21 other projects use a version of this.
Adopted by
Most recently taken up by
- KenmegeTheEducatingGP/codex-claude-companiontheir own version, first seen Nov 16, 2026
- ryanportfolio/Harness-Firmwaretheir own version, first seen Nov 15, 2026
- KroxiDev/engineering-skillstheir own version, first seen Nov 15, 2026
- ryanportfolio/HarnessFirmware.comtheir own version, first seen Nov 12, 2026
- ryanportfolio/Corewise.Academytheir own version, first seen Nov 12, 2026
- AhsanT4riq/claude-plugin-codextheir own version, first seen Nov 8, 2026
Used with
Projects with it also use addskill, apply-firmware and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
ryanportfolioSkilladdskill
Create, import, update, or install repo or personal skills: runtime ownership, resources, discovery checks.
Also in 6 projects that use this
ryanportfolioSkillapply-firmware
Bring current folder (plain folder, existing repo, old or partial firmware copy) up to full Harness Firmware in place. Mirror external repo → adopt-repo.
Also in 6 projects that use this
- +10 more from ryanportfolio
Other versions21 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 21
- xiaolai/cc-suite★ 56same name
- smithjohn78390/smith--skills★ 41same name
- yangliu2060/smith--skills★ 40same name
- mryll/skills★ 8same name
- plannotator/dev-skills★ 7same name
- KroxiDev/engineering-skills★ 4same name
- sandover/agent-skills★ 4same name
- abdurshd/skills★ 2same name
- Kenmege/codex-claude-companion★ 2same name
- KenmegeTheEducatingGP/codex-claude-companion★ 2same name
11 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository63 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- agent-browserSkill
- ast-grepSkill
- branch-review-before-prSkill
- browse-qaSkill
- browser-useSkill
- browseSkill
- bugfixSkill
- build-dmgSkill
- codemapSkill
- code-simplifySkill
- codex-reviewSkill
- commitSkill
- cost-estimateSkill
- create-prSkill
- find-agentsSkill
- find-bugsSkill
- frontend-design-ui-uxSkill
- git-merge-expertSkill
- git-merge-expert-worktreeSkill
- kiro-reviewSkill
- laravel-filamentSkill
- laravelSkill
- lokeiSkill
- map-project-monorepoSkill
- map-projectSkill
- nextjsSkill
- plan-founder-reviewSkill
- plan-to-task-list-with-dagSkill
- pr-retroSkill
- run-parallel-agents-feature-buildSkill
- run-parallel-agents-feature-debugSkill
- rustSkill
- secretsSkill
- startSkill
- update-agent-learningsSkill
- update-claude-learningsSkill
- update-claude-settingsSkill
- update-skill-learningsSkill
- ast-grepSkill
- branch-review-before-prSkill
- codemapSkill
- code-simplifySkill
- commitSkill
- cost-estimateSkill
- create-prSkill
- find-bugsSkill
- frontend-design-ui-uxSkill
- git-merge-expertSkill
- git-merge-expert-worktreeSkill
- map-project-monorepoSkill
- map-projectSkill
- plan-founder-reviewSkill
- plan-to-task-list-with-dagSkill
- pr-retroSkill
- run-parallel-agents-feature-buildSkill
- run-parallel-agents-feature-debugSkill
- startSkill
- update-agent-learningsSkill
- The first 60 shown.
The file
Show its rules and text
We split this skill into the rules it gives an agent
21 rules. 0 are common, 21 are distinctive, and 0 are very specific to this project.
0
Common
Most of the 20 edited versions we read keep these.
21
Distinctive
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 12
Before reporting ANY finding, you ABSOLUTELY MUST:
- Distinctive: line 14
Read the complete diff (every changed line, not just file names)
- Distinctive: line 15
Read surrounding context to verify the issue is real
- Distinctive: line 16
Check if the issue is already handled elsewhere in the changed code
- Distinctive: line 17
Search for existing tests that cover the scenario
- Distinctive: line 18
Never report stylistic issues as bugs — only real defects and vulnerabilities
- Distinctive: line 32
After completing a task from a DAG plan
- Distinctive: line 39
No changes exist on the branch (nothing to review)
- Distinctive: line 56
Full branch: all commits since divergence from default branch
- Distinctive: line 57
Last commit: only the most recent commit (git show --stat HEAD)
- Distinctive: line 60
If $ARGUMENTS contains a scope hint (e.g., "last commit", "uncommitted"), use that.
- Distinctive: line 65
Read the diff to understand what areas need review focus:
- Distinctive: line 83
Use the Agent tool to launch an independent reviewer in a worktree:
- Distinctive: line 129
When the agent returns, extract findings and present to the user:
- Distinctive: line 143
If the user wants to fix:
- Distinctive: line 144
Read each cited file to verify the finding is real
- Distinctive: line 146
Run tests and typecheck
- Distinctive: line 153
Add fixed issues to the "Previously known issues" list
- Distinctive: line 157
Critical: Always maintain the exclusion list between rounds.
- Distinctive: line 182
After a task agent completes, run claude-review on just that task's changes:
- Distinctive: line 230
False positives: Always read the cited code before acting.
“Common” is measured: kept by at least 6 in 10 of the 20 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
<EXTREMELY-IMPORTANT>
Before reporting ANY finding, you **ABSOLUTELY MUST**:- Read the complete diff (every changed line, not just file names)
- Read surrounding context to verify the issue is real
- Check if the issue is already handled elsewhere in the changed code
- Search for existing tests that cover the scenario
- Never report stylistic issues as bugs — only real defects and vulnerabilities
Reporting without verification = false positives, wasted developer time, eroded trust
This is not optional. Every finding requires disciplined verification.
</EXTREMELY-IMPORTANT>
Claude Review
Spawn a separate Claude Code agent to independently review code changes.
When to Use
- User says "claude review", "self-review", "/claude-review"
- After completing a task from a DAG plan
- User wants a second opinion from Claude itself (not codex or kiro)
- As part of the review trifecta:
/claude-review,/codex-review,/kiro-review
When NOT to Use
- User specifically wants codex or kiro review (use those skills instead)
- No changes exist on the branch (nothing to review)
- User wants to make changes (this is review-only)
Workflow
Step 1: Determine the Review Scope
# Detect default branch
DEFAULT_BRANCH=$(git symbolic-ref refs/remotes/origin/HEAD 2>/dev/null | sed 's|refs/remotes/origin/||' || echo main)
# What changed?
git diff --stat $DEFAULT_BRANCH..HEAD
git diff --name-only $DEFAULT_BRANCH..HEADDetermine scope:
- Full branch: all commits since divergence from default branch
- Last commit: only the most recent commit (
git show --stat HEAD) - Uncommitted: staged + unstaged changes
If $ARGUMENTS contains a scope hint (e.g., "last commit", "uncommitted"), use that. Otherwise default to full branch review.
Step 2: Build Review Context
Read the diff to understand what areas need review focus:
# Get changed directories to identify subsystems
git diff --name-only $DEFAULT_BRANCH..HEAD | sed 's|/[^/]*$||' | sort -u
# Count changes per area
git diff --stat $DEFAULT_BRANCH..HEADIdentify:
- Which packages/modules changed
- Whether changes are security-sensitive (auth, crypto, secrets, permissions)
- Whether changes involve concurrency, I/O, or shell execution
- What test files exist vs what source files changed
Step 3: Spawn the Review Agent
Use the Agent tool to launch an independent reviewer in a worktree:
Agent tool call:
subagent_type: general-purpose
isolation: worktree
prompt: |
You are performing a code review of changes on branch [branch-name]
compared to [base].
## Changed files
[list of changed files from Step 2]
## Review focus
Based on the changes, focus on:
1. [Area 1 — derived from actual changes]
2. [Area 2 — derived from actual changes]
3. [Area 3 — derived from actual changes]
## Previously known issues (do NOT re-report)
[List any issues already fixed in prior rounds]
## Instructions
1. Read the full diff: git diff [base]..HEAD
2. For each changed file, read it completely and check for:
- Logic errors, off-by-one, null access, missing error handling
- Security issues: injection, auth bypass, credential leaks
- Race conditions, TOCTOU, concurrent modification
- Edge cases: empty input, unicode, special characters
- Missing or incorrect test coverage
3. Verify each finding against surrounding context before reporting
4. Report findings as a markdown table:
| # | Priority | File:Line | Description | Suggested Fix |
5. Priority levels: P1 (crash/security), P2 (incorrect behavior), P3 (minor)
6. If you find nothing significant, say so explicitly
Do NOT make any code changes. Report only.Key configuration:
isolation: "worktree"— gives the reviewer a clean read-only copysubagent_type: general-purpose— full tool access for reading code- The prompt must include the specific files and focus areas (not generic)
Step 4: Parse and Report Findings
When the agent returns, extract findings and present to the user:
## Claude Review: [scope description]
| # | Priority | File:Line | Description |
|---|----------|-----------|-------------|
| 1 | P1 | ... | ... |
[N] findings: [x] P1, [y] P2, [z] P3Step 5: Fix Findings (if requested)
If the user wants to fix:
- Read each cited file to verify the finding is real
- Apply minimal fixes
- Run tests and typecheck
- Commit the fixes
Step 6: Iterate (if requested)
For thorough review, run multiple rounds:
- Fix findings from round N
- Add fixed issues to the "Previously known issues" list
- Spawn a new review agent with the updated prompt
- Repeat until findings converge to zero
Critical: Always maintain the exclusion list between rounds.
Example: Full Branch Review
Agent tool call:
description: "Review secrets branch changes"
subagent_type: general-purpose
isolation: worktree
prompt: |
Review all changes on this branch compared to main.
Changed files: vault.ts, injection.ts, proxy/server.ts, command-parser.ts
Focus on:
1. SECURITY: AES-256-GCM implementation, credential leaks in error messages
2. CORRECTNESS: Shell parser with quotes/pipes/subshells
3. EDGE CASES: Concurrent vault access, proxy chunked encoding
Read every changed file. Verify findings against context.
Report as markdown table with Priority, File:Line, Description, Fix.Example: Post-Task Review (DAG integration)
After a task agent completes, run claude-review on just that task's changes:
Agent tool call:
description: "Review TASK-003 implementation"
subagent_type: general-purpose
isolation: worktree
prompt: |
Review the changes in the last commit (TASK-003: Add vault encryption).
Run: git show --stat HEAD
Then: git show HEAD (full diff)
Focus on:
1. Is the AES-256-GCM implementation correct?
2. Are there credential leaks in error messages?
3. Is the key derivation secure?
Report findings as markdown table.Comparison with Sibling Skills
| Skill | Tool | Best For |
/claude-review | Claude Code (Agent) | Deep code understanding, context-aware review |
/codex-review | OpenAI Codex CLI | Independent perspective, repro scripts |
/kiro-review | Kiro CLI | Alternative AI perspective |
Recommendation: For critical code, run all three and cross-reference findings. The overlap between reviewers catches more bugs than any single tool.
Safety Rules
| Rule | Reason |
Always use isolation: "worktree" | Prevents reviewer from modifying your working copy |
| Always list prior fixes in iterations | Prevents re-reporting already-fixed issues |
| Never include actual secret values in findings | Report type and location only (e.g., "hardcoded key at config.ts:17") |
| Always verify findings before fixing | Any reviewer can produce false positives |
| Never skip the diff reading step | Generic prompts get generic (useless) results |
| Be specific about focus areas | Targeted reviews find more bugs than broad ones |
Failure Modes
- Agent returns no findings: Prompt was too generic. Add specific file names and focus areas.
- Agent makes code changes: Forgot
isolation: "worktree"or prompt didn't say "report only" - Repeated findings across rounds: Exclusion list not maintained — add prior fixes
- False positives: Always read the cited code before acting. ~10% of findings are wrong.
- Agent times out: Diff too large. Break into per-file or per-module reviews.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

