Git review pr comments
Skill ainova-systems/intelligence-dev-packs/packs/core/skills/git-review-pr-comments
Triage PR review comments: fix, discuss, or decline with reason - every thread answeredFrom its SKILL.md
npx -y skills add ainova-systems/intelligence-dev-packs --skill git-review-pr-commentsAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
- runs commandsInstructs the agent to run 2 commands, including `gh pr list --head <branch> --state open --json number --jq '.[0].number'` and 1 more.
SKILL.md
2.7 KB, 607 tokens by cl100k_base, as published. Nobody here has run it
Handle PR Review Comments
Drain reviewer feedback: every thread ends with a fix commit or a reasoned reply. Silence is never a response.
Steps
- Resolve the PR for the current branch:
gh pr list --head <branch> --state open --json number --jq '.[0].number'. - Fetch open threads: GraphQL
reviewThreads(first: 100) { nodes { id isResolved comments(first: 1) { nodes { path line body author { login } } } } }- keepisResolved == false(already-resolved threads were handled on a prior run; skipping them is what makes re-running safe); plusgh pr view <pr> --json reviews,commentsfor conversation-level notes. - For each thread, read the cited file and line, then VERIFY the claim against the code before trusting it - automated reviewers (bots) often cite a rationale that is stale or does not match the project's conventions. Grep for the actual precedent (sibling code, the relevant rule) instead of mirroring the suggestion verbatim.
- Classify each: fix (reviewer is right, or the change is cheaper than the debate) / discuss (real tradeoff - answer with the reasoning, no code yet) / decline with reason (conflicts with a project rule or an accepted ADR - cite it).
- When fixing a real issue, grep for the same class of issue across the tree and fix the siblings in the same commit - a reviewer flags one instance, not the whole class. If the comment exposes an unclear or wrong documented rule, fix the rule in the same change (
dev-context-engineering). Apply fixes grouped into logical commits viagit-commit-push(gates run before push). - Reply to every handled thread, then mark it resolved (
resolveReviewThreadmutation by thread id) - the resolved flag is the only state that survives a re-run; a reply without resolve re-appears next run. Reply content: the commit reference for fixes, the reasoning for discuss/decline. Match the reviewer's tone; keep replies short. - Out-of-scope asks: agree in the reply, file a follow-up item, link it - do not grow the PR.
- Report counts per category plus anything needing an owner decision.
Verify
- Zero unanswered threads; every handled thread replied AND resolved; every fix commit pushed; gates green.
Scope / hand-off
- CI babysitting and the outcome label -
git-finalize-pr; merging -git-merge-pr.
CRITICAL
- Never resolve a thread without a reply; never reply without resolving (it re-appears next run).
- Verify a reviewer's claim against the code before accepting it - a confident but wrong bot comment is still wrong.
- Conflicting reviewer asks - surface to both, never silently pick one.
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.