agentleFS
oktawianwybieralskioktawianwybieralski/coagent8/skills/coagent8-review/SKILL.md

coagent8-review

Skill2 starsChanged 11 days agoMITView on GitHub

Independent review of Git changes by another local coding agent through the CoAgent8 review tool. Use when the user asks to review uncommitted or staged changes, a commit, a branch or a range, wants a check before committing or opening a pull request, or names CoAgent8 review.

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 1 is distinctive, and 3 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.

  • 3

    Very specific

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

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

    Put the review focus in instructions: the goal of the change, acceptance criteria and risky areas, naming files instead of pasting code.

    line 12
  2. Very specific:

    Leave backend and model unset unless the user named one.

    line 16
  3. Very specific:

    Read the result. The reviewer's VERDICT: in the body is its judgement.

    line 24
  4. Distinctive:

    After two further rounds without progress, stop and ask the user.

    line 39

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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

CoAgent8 review

Text the user typed with this command names the scope or the review focus.

  1. Choose scope: uncommitted (default), staged, a commit, a branch or a range a...b. Check with Git that the scope contains the changes you mean.
  2. Put the review focus in instructions: the goal of the change, acceptance criteria and risky areas, naming files instead of pasting code; the reviewer reads the diff from Git. Do not paste earlier review transcripts and do not ask the reviewer to answer READY.
  3. Leave backend and model unset unless the user named one; auto prefers a different CLI than yours. For a user's price limit or range (e.g. "X-Y" means output_price_below: Y), pass output_price_below and leave model unset; the server picks the qualifying default or the most recently released qualifying model. The author of a change cannot supply its own independent approval. A model coagent8-doctor tags ask runs only after the user approves the server's form: tell the user before the call. Never call a model tagged deny.
  4. Read the result. The reviewer's VERDICT: in the body is its judgement; the server's Verdict: READY or Verdict: BLOCKED (reason) in the footer is the gate decision.
  5. A read-only reviewer cannot run typecheck or tests. Pass them in checks (such as npm test): the server runs them on exactly the reviewed change before the turn, gives the reviewer their results, and a failure blocks. They run with the user's rights, outside any sandbox, so the user approves each on a form or lists it in allowedChecks; you cannot approve them. Without checks, run them yourself and report their results separately. Repeating the same review does not change the verdict.
  6. Fix the findings you accept, then review again. Pass the session_handle from the footer so the reviewer compares with its previous round. Any code change invalidates the previous verdict; do not repeat a review of unchanged code that completed with full coverage: the server then starts no turn and returns the earlier verdict (an incomplete or failed review can be retried). After two further rounds without progress, stop and ask the user.

A large change is reviewed in parts by the server, then in one synthesis turn; the result lists each part with its session. A change too large even in parts fails with INPUT_LIMIT before any turn. Then narrow the scope: put a mechanical commit first and review the range after it, which still ends at the checked-out state, and verify the mechanical part by rerunning its command. Say which part no reviewer saw. A range ending elsewhere is refused.

Report the server verdict, each finding with its location and your disposition (fixed, rejected with the reason, or deferred), and the checks you ran.

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.