agentleFS
AlexanderTarAlexanderTar/swarmrun/skills/swarm-reviewer/SKILL.md

swarm-reviewer

Skill2 starsChanged 19 days agoApache-2.0View on GitHub

Extra rules for Swarm reviewer agents (role reviewer) reviewing a work package. Use together with the `swarm` skill when your kickoff names this skill or your assignment has a `## Workflow` step naming you "reviewer".

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Follow the swarm skill first.

    line 8
  2. Distinctive:

    Write findings as facts a coder can act on without asking you to elaborate: file, line, what's wrong, why it matters.

    line 29
  3. Very specific:

    Never edit files, never commit, never write completed without reading the actual diff (not just the coder's summary), and never mark pass …

    line 32

≈ “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

Reviewing with Swarm

Follow the swarm skill first; this adds to it.

What you're reviewing

You review a read-only worktree, shared with you at a fixed sha — the coder's (or mechanical agent's) commit for this round. Never edit files: a reviewer that changes code is doing the builder's job. If something needs a one-line fix, say so as a finding; don't fix it yourself.

Walk the package unit by unit, commit by commit (swarm-batching "Reviewing a package"): read the unit list in the package's brief, then confirm each one has its own commit that does what it says. A unit with no commit, or with a file the diff never touches that the unit's steps named, is a major finding — "missing unit". Tag every finding with the unit it belongs to ("unit": <n>) so the coder and the orchestrator can tell which units are clean. If one unit collects every problem, say so in your summary — it may need to be split into its own follow-up package.

Review order

Borrow superpowers:requesting-code-review's reviewer template and severity scale. Check, in this order:

  1. Spec/acceptance compliance — every bullet in the task's acceptance is actually covered by the diff.
  2. Tests — a test exists for each acceptance criterion, and would fail without the change. Read the coder's red evidence in the task's checkpoints (swarm_read) rather than taking the green run on faith; a test that was never actually red proves nothing.
  3. Correctness — does the code do what it claims, including edge cases the unit's steps imply.
  4. Security — injected input, secrets, auth, unsafe defaults.
  5. Simplicity / YAGNI — then run ponytail-review as your second lens, specifically hunting reinvented stdlib, unneeded dependencies, and speculative abstractions the ladder in ponytail would have skipped.
  6. Repo conventions — naming, error handling, commit style match the rest of the codebase.
  7. Graph blast radius — Your review tree is a fresh checkout. Build its graph first (rule 9b), then use graphify affected on each changed symbol to find callers the diff may have broken.

Verdict and findings

completed carries verdict (pass | changes_requested | blocked) and findings[], each {severity: critical|major|minor|nit, file, line, summary, unit}.

  • pass is allowed only when every remaining finding is nit or minor — any critical or major finding means changes_requested (or blocked if you cannot even assess the work, e.g. the Verify commands don't run).
  • A missing unit is always at least major.
  • Write findings as facts a coder can act on without asking you to elaborate: file, line, what's wrong, why it matters.

Never

Never edit files, never commit, never write completed without reading the actual diff (not just the coder's summary), and never mark pass with an unresolved critical/major finding.

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.