agentleFS
doivamongdoivamong/agent-workbench/.claude/skills/awb-review/SKILL.md

awb-review

Skill2 starsChanged 6 months agoMITView on GitHub

WHAT: review a code change in three passes — does it meet the spec, is it well built, and can an adversary break it. USE WHEN: the user says "review my changes / this PR / what I just wrote", or before committing a non-trivial change. DO NOT TRIGGER: a broad "audit the whole codebase" request; research/compare questions; when no code has been written yet.

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    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
  1. Distinctive:

    Stop reviewing as the author.

    line 45
  2. Distinctive:

    If this is security- or money-adjacent, how would I exploit it?

    line 49

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Three-stage review

Announce on activation: "Using awb-review — three passes: spec, quality, adversarial."

A checklist review catches typos. A staged review catches the things checklists miss: wrong assumptions, security holes, and race conditions. Run the passes in order; stop early only if a blocking issue makes later passes moot.

Stage 1 — Spec compliance

Does the change actually do what was asked? Compare the diff against the stated goal. Flag anything implemented that wasn't asked for (scope creep) and anything asked for that's missing.

Philosophy check (this repo): does the change oversell a guard, drop a "what it does NOT do" line, blur SHIPS vs BLUEPRINT, let a star/adoption framing or a domain identifier creep in, or otherwise contradict PHILOSOPHY.md? Its "what would betray this" list is the checklist — a regex can't catch this, so it lives here.

Stage 2 — Quality

Walk the diff against a quality bar. Keep your real bar in a reference file so this stays short:

  • Correctness, error handling, resource cleanup.
  • Readability over cleverness; naming; dead code.
  • Tests exist for the new behaviour.

See references/quality-checklist.md for the full list.

Stage 3 — Adversarial (the pass that earns its keep)

Stop reviewing as the author; start attacking as an adversary. Ask:

  • What input breaks this? Empty, huge, malformed, concurrent, malicious?
  • What does this assume that isn't guaranteed?
  • If this is security- or money-adjacent, how would I exploit it?

Scope gate: you can skip Stage 3 for tiny, non-sensitive changes (say, ≤2 files and a few dozen lines that don't touch auth, payments, or data integrity). Everything else gets it.

Output

Report findings ranked by severity (blocking / important / nit) and tagged with a confidence (high / medium / low), each with a file:line and a concrete fix. Severity sets what gates the merge: blocking must be fixed, important should be, nit is optional author's-discretion — say so, so a polish suggestion isn't mistaken for a merge blocker. Mark a low-confidence finding "verify", not "fix" — the adversarial pass throws plausible-but-wrong hits; labelling them honestly keeps real issues from drowning in speculation. Don't soften a real problem to be polite.

References

  • references/quality-checklist.md — the Stage 2 bar

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.