agentleFS
HyperVonHyperVon/agent-guidance-kit/skills/review-feedback-resolution/SKILL.md

review-feedback-resolution

Skill1 starsChanged 15 days agoApache-2.0View on GitHub

Resolve incoming review feedback (PR comments, code/architecture/security review findings, or audits) against repository evidence: disposition each item and apply only accepted fixes. Use when a review has returned comments to act on. Do not use to find defects (use code-review, security-review, or architecture-review); this skill receives findings, it does not generate them.

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 6 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this skill after a review has returned comments — from a PR, a code review, an architecture review, a security review, or an audit.

    line 24
  2. Distinctive:

    Before judging any comment, record the review point so dispositions stay anchored to what was actually reviewed:

    line 30
  3. Very specific:

    After assigning independent dispositions, evaluate all accepted items together before implementation.

    line 64
  4. Very specific:

    When one suggestion conflicts with that authority, retain the supported item and mark the other rejected-with-evidence.

    line 70
  5. Very specific:

    When the controlling contract or intended behavior is genuinely unclear, mark the conflicting items needs-clarification and do not …

    line 72
  6. Distinctive:

    Do not invent a hybrid compromise that satisfies neither contract.

    line 74
  7. Very specific:

    Keep valid but out-of-scope architectural work deferred.

    line 75
  8. Distinctive:

    run the appropriate verification (the applicable tests named in the frozen context) and record the result.

    line 90
  9. Distinctive:

    Do not assume reviewers are correct: Verify the claim against the frozen context and current source.

    line 115
  10. Distinctive:

    Do not lose rejected items: Every comment gets a disposition, even "no change".

    line 119

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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 Feedback Resolution

Authority and boundary

This skill receives review output and decides what to do with it. It does not replace code-review, which finds defects and produces findings. Keep the two roles separate:

SkillOwns
code-review / security-review / architecture-reviewFinding defects and producing review findings
review-feedback-resolutionReceiving findings, evaluating each, resolving accepted items, rejecting unsupported ones

Use this skill after a review has returned comments — from a PR, a code review, an architecture review, a security review, or an audit. It is the resolution step, not the discovery step.

Freeze the review context

Before judging any comment, record the review point so dispositions stay anchored to what was actually reviewed:

  • revision — the commit/SHA the review targeted.
  • branch/diff — the exact branch and diff range under review.
  • relevant contracts — public interfaces, specs, schemas, or local rules the comment appeals to.
  • applicable tests — which checks should pass after a change.

For a GitHub PR, record the PR's head SHA and the diff range at review time (e.g. gh pr view <n> --json headRefOid,baseRefOid). If the PR has been force-pushed or updated since the review, note that the frozen context no longer matches the current head and re-freeze before dispositioning.

Re-reading the diff at a different revision than the reviewer saw produces wrong dispositions. Confirm the working tree matches the reviewed revision before acting.

Disposition every comment independently

Evaluate each comment on its own evidence. Do not let one obviously-wrong comment poison the rest, and do not accept a batch just because most of it is sound. Assign exactly one disposition per item:

DispositionMeaning
acceptedThe suggestion is valid and in scope; make the smallest safe change.
rejected-with-evidenceThe suggestion is wrong, redundant, or out of scope; explain why using repository evidence.
already-resolvedThe concern is already handled in the current code or an in-flight change.
duplicateIt repeats another comment; resolve once and link the pair.
needs-clarificationThe comment is ambiguous or missing the evidence needed to act; ask a precise question.
deferredValid but out of scope for this change; record it as a tracked follow-up.

Reconcile dispositions as a set

After assigning independent dispositions, evaluate all accepted items together before implementation. Two individually plausible suggestions may still require incompatible changes.

  • Identify the source-of-truth contract, repository policy, or authorized decision that controls the conflict.
  • When one suggestion conflicts with that authority, retain the supported item and mark the other rejected-with-evidence.
  • When the controlling contract or intended behavior is genuinely unclear, mark the conflicting items needs-clarification and do not implement either one.
  • Do not invent a hybrid compromise that satisfies neither contract.
  • Keep valid but out-of-scope architectural work deferred; recording it in the report does not authorize creating an issue or another remote follow-up artifact.

Maintain three separated streams and do not blur them:

  • review feedback — what the reviewer said and your disposition of it;
  • implementation decision — the change you choose to make;
  • final verification — the checks that prove the change is safe.

Resolve accepted items

For each accepted item:

  • make the smallest safe change that addresses the concern;
  • preserve scope boundaries — do not widen the change to unrelated cleanup;
  • run the appropriate verification (the applicable tests named in the frozen context) and record the result.

For each rejected-with-evidence item, state the precise reason with a repository anchor (file/line, contract, or test) so the rejection is auditable, not an opinion.

Side effects and approval boundary

Assessment mode is read-only by default. Do not modify code, reply to reviewers, update branches, or apply accepted fixes unless the user explicitly requests implementation of selected accepted items. Resolving whether feedback is valid and implementing the resolution are separate steps:

  1. evaluate the review feedback against the frozen context;
  2. decide whether each item is valid and what disposition it receives;
  3. implement accepted changes only as a separate, explicitly authorized action.

Preserve the ownership model: code-review finds defects, this skill evaluates incoming feedback, and implementation is a distinct authorized step — not an automatic consequence of resolving a comment.

Anti-patterns

  • Do not blindly obey reviewers. A review comment is an input, not an order.
  • Do not assume reviewers are correct. Verify the claim against the frozen context and current source.
  • Do not expand a localized issue into an architecture rewrite without evidence. A naming nit does not justify restructuring a module.
  • Do not lose rejected items. Every comment gets a disposition, even "no change"; silence is not a resolution.

Output

Return a resolution report with:

  • the frozen review context (revision, branch/diff, contracts, tests);
  • a per-comment table: comment, disposition, evidence, action taken or reason;
  • the verification result for every applied change;
  • a list of deferred or needs-clarification items still open.

Completion means every comment has a disposition and every accepted change is verified. It does not mean every comment was applied.

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.