agent-review
Skill3 starsChanged 50 days agoMITView on GitHub
Use in /implement after committing, inside the draft PR, to review the committed diff. Dispatches a dependency-free review subagent fed the diff + review criteria and returns structured findings (severity / location / problem / fix). The recipe drives the address→re-review loop with a human exit gate; the pr-review-toolkit ceiling is an opt-in override of this skill.
11 other projects use a version of this.
Adopted by
Most recently taken up by
- facebookresearch/autoform-bottheir own version, first seen Nov 12, 2026
- blamechris/chroxytheir own version, first seen Nov 8, 2026
- khoralabs/skillstheir own version, first seen Nov 1, 2026
- IlyaSemenov/agent-review-skilltheir own version, first seen Oct 31, 2026
- mementomori-co/maker-masterytheir own version, first seen Oct 25, 2026
- alexaiarchitect/google-adk-skillstheir own version, first seen Oct 20, 2026
Used with
Projects with it also use commit.
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.
SentrySkillcommit
Create commit messages following Sentry conventions. Use when committing code changes, writing commit messages, or formatting git history. Follows conventional commits with Sentry-specific issue references.
Also in 3 projects that use this
Other versions11 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 11
- DmNote-App/DmNote★ 4.2ksame name
- foldkit/foldkit★ 1.1ksame name
- Tsuzat/Edra★ 675same name
- facebookresearch/autoform-bot★ 143same name
- blamechris/chroxy★ 8same name
- IlyaSemenov/agent-review-skill★ 4same name
- mementomori-co/maker-mastery★ 2same name
- khoralabs/skills★ 1same name
- ngc-shj/claude-code-config★ 1same name
- alexaiarchitect/google-adk-skills★ 0same name
1 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 repository8 other files it gives its agents.
- capture-to-vaultSkill
- cleanup-artifactsSkill
- dispatch-explorationSkill
- load-contextSkill
- open-draft-prSkill
- plan-fileSkill
- publish-prSkill
- repo-setupSkill
The file
We split this skill into the rules it gives an agent
3 rules. 0 are common, 3 are distinctive, and 0 are very specific to this project.
0
Common
Most of the 9 edited versions we read keep these.
3
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 32
read the diff and the changed files in full — from the worktree: pass its absolute path in the prompt and tell the subagent to read files …
- Distinctive: line 33
check against the criteria and the AC / plan context.
- Distinctive: line 36
Return the findings to the recipe: Do not loop or apply fixes here — the recipe owns the address→re-review loop and the human …
“Common” is measured: kept by at least 6 in 10 of the 9 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
Agent Review
The review step of /implement's repo path. Runs on committed work (never unstaged changes), inside the draft PR. Dispatches a review subagent — fed the diff plus review criteria — and returns structured findings for the recipe to act on.
Dependency-free by default: a fresh install gets working review with nothing beyond dossier-mcp. The richer pr-review-toolkit specialist panel is an opt-in override of this skill (see Project overrides), not a built-in — so /implement never requires the toolkit.
Input
- diff — the committed change to review, as a ref range (default
origin/main...HEAD) or a PR number. - criteria (optional) — focus areas; defaults below.
- context (optional) — the
plan-filepath and the ticket / acceptance criteria, passed to the reviewer so it can check the change against intent.
Default review criteria (the floor)
- Correctness — does the code do what it claims; do its own commands/claims hold up when checked; logic errors, edge cases, off-by-ones.
- AC coverage — does the change actually satisfy the ticket's acceptance criteria.
- Silent failures / error handling — swallowed errors, missing guards, fallback that hides real failures.
- Internal consistency — self-contradiction; drift from the change's own stated contract.
- Convention consistency — does it match sibling code/patterns in the repo.
A project override can replace or extend these. If the AC / plan context isn't provided, the reviewer flags its absence as a gap rather than silently skipping the AC-coverage and intent checks.
Process
- Resolve the diff —
git diff <range>for a ref range, orgh pr diff <number>for a PR number — plus the list of changed files and the context (plan-filepath + AC) if available. - Dispatch a review subagent restricted to read-only tools (no Edit/Write/NotebookEdit — e.g. a read-only agent type), so the no-side-effects contract is structurally enforced, not just instructed. Instruct it to:
- read the diff and the changed files in full — from the worktree: pass its absolute path in the prompt and tell the subagent to read files there (its own cwd may be the main checkout, i.e.
main's versions); - check against the criteria and the AC / plan context;
- be skeptical and concrete — verify claims, don't praise; where a claim is checkable (a command, a path, an invariant), check it rather than trust it;
- return findings in the structured format below, plus a one-line verdict.
- read the diff and the changed files in full — from the worktree: pass its absolute path in the prompt and tell the subagent to read files there (its own cwd may be the main checkout, i.e.
- Return the findings to the recipe. Do not loop or apply fixes here — the recipe owns the address→re-review loop and the human ready-to-publish gate (a primitive can't host an interactive gate).
Output / contract
- In: a committed diff (+ optional criteria / context).
- Out: a list of findings, each
{ severity: blocker | major | minor | nit, location, problem (1 sentence), suggested fix (1 sentence) }, plus a one-line verdict (mergeable as-is / mergeable with minor fixes / needs changes before merge). If a severity level has no findings, say so. - Side effects: none — read-only review that dispatches a subagent. No commits, no file edits, no PR state changes.
Project overrides (the ceiling)
The toolkit ceiling is an override, not a default (per the dependency posture):
- User-level override (
~/.claude/skills/agent-review/) — dispatchpr-review-toolkit's panel (code-reviewer,silent-failure-hunter,type-design-analyzer,comment-analyzer,pr-test-analyzer) and aggregate their output into the same structured-findings format. The user with the toolkit installed gets the rich review everywhere; a fresh user isn't forced into the dependency. - Project-level override — swap a stack-specific reviewer (e.g. a TypeScript- or Ansible-aware agent) or tune the criteria for the repo.
Overrides must honor the contract — input a diff, output structured findings, and stay read-only / side-effect-free — so the recipe's review loop keeps working unchanged.
Future scope
workflow-reviewer— a thin bundled reviewer agent tuned to a terse, critical, why-not-what style, replacing the general-purpose subagent as the floor. This is the self-sufficiency direction: the floor becomes good enough that the toolkit ceiling is rarely needed.
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.

