agentleFS
rajasdevelrajasdevel/outcomebound/skills/review-findings/SKILL.md

review-findings

Skill0 starsChanged 15 days agoApache-2.0View on GitHub

Use when acting on review findings, or writing a review the project requires. Check each finding as a claim and give it a disposition.

2 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 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.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Check each finding against the code at the revision you hold before you act on it: run the case it describes, or read the lines it names.

    line 16

≈ “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

Review findings

You are done when the change holds against every finding that was right, and each finding has one disposition and the evidence for it. A finding is a claim about the code, not an instruction: a reviewer can read an older revision, lack a fact that you have, or be wrong. A person's instruction in this session is an instruction, not a finding; this skill is for what a review says about the code.

Acting on findings

Check each finding against the code at the revision you hold before you act on it: run the case it describes, or read the lines it names. Where the finding gives a counterexample you can run, leave its disposition open during the fix. After the last relevant change, complete that check and read its result before writing fixed or PASS; report an unavailable check UNVERIFIED. Then give it one disposition:

  • fixed, and where: the commit, file or test that holds the change. Where the finding names a kind of defect, such as a missing check or a repeated pattern, look for its other instances first: fix each, or name those left as deferred.
  • rejected, and why: the evidence that the claim is wrong, such as the case you ran or the line that answers it.
  • deferred, and where it is recorded: the ticket or design row that holds it.

accepted is no disposition. A finding you agree with is fixed or deferred, and until then it is open. Where a finding is unclear, ask whoever wrote it what they mean and go on with the others; do not guess its meaning and act on that. Where they cannot be asked, record it deferred with the question, where the ticket or the design holds it. Write each disposition where the project keeps reviews, else in your reply to the reviewer or your report; do not make a review file for it.

A reviewer you dispatch reads again only where a fix changes a claim its finding rests on or a result a user can observe, and then reads the changed claims and the context needed to check them. A finding that changes neither takes its disposition in this round and starts no other. Where a round finds another case of a defect an earlier round fixed, check whether the two share a cause: where they do, fix that cause within the task's bounds; where they do not, each finding takes its own disposition. Where the choice left is the person's, tell them how many rounds ran and what the last ones found, and hold only the work that waits on their answer. For a review file (the form is below), outcomebound review since <file> shows what changed in Git since the revision its Reviewed: line names, and which findings cite a changed path. A cited path that changed does not make a finding stale, nor an unchanged one hold. It writes nothing.

Writing a review

A review states the revision it read. Each finding names the claim, where it holds and the evidence: the command and what it printed, or the lines read, so that a reader can check it without you. Try a finding's counterexample before you report it; call a finding you did not check UNVERIFIED. Findings that share one cause go under that cause; where a finding is one case of a rule, check that rule across the surface the change affects and report the cases you find together. Where the change rewrites text that carries a requirement, a bound or a verdict, compare the old and new text against what it carries. Separate a finding that stops the accepted outcome or makes its evidence wrong (a test that cannot fail, a wrong expected value) from an improvement the outcome does not need; mark the second optional; whoever acts on it may reject it with that reason. Give the change one verdict, first: FAIL while a finding of the first kind stands; UNVERIFIED where you did not measure the property the change exists to protect, such as a time, a limit or a rollback path; else PASS. It is your judgment of the change, not of your own scope, and not outcomebound review check's verdict on dispositions. End with what you did not read or run, and nothing else. Leave the Disposition: line to whoever acts on the finding: a finding is open until then, and outcomebound review check reads it as FAIL.

The review file

Where the project keeps reviews as files, a file has this form: a Reviewed: line near its top, and one heading for each finding, with its id in letters, digits and -, and under it one Disposition: line.

Reviewed: <ref>

### <id> · <title>
Disposition: fixed — <where>

### <id> · <title>
Disposition: rejected — <reason>

### <id> · <title>
Disposition: deferred — <where>

Text between headings is notes. outcomebound review check <file> fails a finding with no disposition, an unknown disposition or an empty reason. It does not show that a fix is right, that a rejection is right, or that the place named holds what it says; that is yours to check.

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.