pr-comment-review
Audit and remediate an open GitHub PR's review feedback: verify Copilot and Codex actually reviewed, drive every unresolved thread to a fix-or-justify reply, correct PR title/description drift against the real diff, and run a one-time self-review pass. Trigger on "have all PR comments been addressed", "did copilot and codex review this", "reply to the PR comments", "address the PR feedback", "is this PR ready to merge", "clean up this PR before merging", "make sure nothing got missed on this PR" — and on generic "review comments" mentions, on single-bot mentions, and when prepping a new PR for review before comments exist. Not a fresh review of an unpushed working diff; that's `/code-review`.
anchildress1 · Static checks on file
Installation
Read the source and dependencies before running a command. Copying does not install or execute anything.
npx skills add anchildress1/awesome-github-copilot --skill pr-comment-reviewAbout this skill
status: polish
name: pr-comment-review
description: >
Audit and remediate an open GitHub PR's review feedback: verify Copilot and Codex actually
reviewed, drive every unresolved thread to a fix-or-justify reply, correct PR title/description
drift against the real diff, and run a one-time self-review pass. Trigger on "have all PR
comments been addressed", "did copilot and codex review this", "reply to the PR comments",
"address the PR feedback", "is this PR ready to merge", "clean up this PR before merging",
"make sure nothing got missed on this PR" — and on generic "review comments" mentions, on
single-bot mentions, and when prepping a new PR for review before comments exist. Not a fresh
review of an unpushed working diff; that's /code-review.
argument-hint: "optional: PR number or URL (defaults to the PR for the current branch)"
allowed-tools: Bash, Read, Edit, Write, Grep, Glob, Agent
disable-model-invocation: false
user-invocable: true
Remediation workflow for an existing PR. Bot threads rot: unresolved thread, PR merges, the "minor" finding was the real one. Every step below exists to close that gap.
GitHub MCP tools are preferred when loaded
(ToolSearch("select:mcp__plugin_github_github__pull_request_read")); gh api equivalents are
given inline.
1. Resolve the PR
Given a number/URL, use it. Otherwise:
gh pr view --json number,title,body,url,isDraft,baseRefName,headRefName
owner/repo come from whichever identified the PR. A supplied URL carries its own
owner/repo — parse them from it, because the PR may live in a repo the current checkout isn't
(same number, different repo, and every read and reply would land on a stranger's PR). Fall back
to git remote -v (origin) only for a bare number or the current-branch case.
No open PR for the branch → say so and stop. Don't guess.
2. Confirm both required bots reviewed
- Copilot:
copilot-pull-request-reviewer[bot] - Codex:
chatgpt-codex-connector[bot]
pull_request_read(method="get_reviews" | "get_comments", owner, repo, pullNumber)
# gh api --paginate repos/{owner}/{repo}/pulls/{pr}/reviews | .../issues/{pr}/comments
Read every page before concluding anyone is absent. gh api returns one page by default, so a
bot that reviewed after 30 other events looks missing without --paginate; the MCP tool pages
via perPage + the after cursor from pageInfo.
A missing bot is a diagnosis, not a shrug. Diagnose per bot — they have independent configurations, so one explanation rarely covers both:
-
Draft PR — check
isDraft. This only explains Copilot if the repo has not enabled Review draft pull requests in its ruleset; with that on, a draft is no explanation at all and you need a different cause. -
Pending — last push minutes ago; report the timestamp instead of "absent".
-
No prior review history — check whether the bot has ever commented on any PR here:
gh api -X GET search/issues -f q="repo:{owner}/{repo} commenter:app/{bot-slug}" --jq '.total_count'-X GETis required —-falone makesghPOST, and search answers with a bare404that reads like a missing repo.{bot-slug}is the login without the[bot]suffix.Report a
0as "no prior comments in this repo" — never as "not installed". A freshly installed app, or one enabled before its first eligible PR, returns exactly the same0, and calling that an installation problem sends the user to fix something that isn't broken.There is no user-token API that answers "is this app installed" (
repos/{owner}/{repo}/installationneeds a GitHub App JWT and returns401togh authcredentials). A recent review on any PR here is the practical proof — a bot that answered yesterday is installed today.You can also check whether this bot was requested, but filter by who was requested — a bare count of
review_requestedevents counts requests aimed at humans and at the other bot:gh api --paginate repos/{owner}/{repo}/issues/{pr}/timeline \ --jq '[.[] | select(.event=="review_requested") | .requested_reviewer.login // .requested_team.name] 'Read the result carefully in both directions:
- Requested and silent, while it answered other PRs in minutes → installed and not delivering. A bot-side failure, worth saying plainly.
- Never requested proves nothing on its own. Bots that review automatically never appear
here at all — Copilot shows up because a ruleset requests it, while Codex triggers on PR
open / ready-for-review / an
@codex reviewcomment and so leaves no request event.
Before declaring any automatic reviewer broken, re-trigger it manually and time the response. Posting
@codex reviewis the supported path, and a bot that answers a manual trigger in minutes after missing the automatic one has a trigger problem, not an install problem — a much more useful thing to tell the user. Send them to Settings → GitHub Apps only when nothing else explains it. -
Other — state the evidence. Never "reason unclear"; if inconclusive, say what you checked and what came back, so the user doesn't re-derive it.
3. Drive every unresolved thread to a conclusion
pull_request_read(method="get_review_comments", owner, repo, pullNumber)
Returns review_threads[], each with is_resolved / is_outdated / is_collapsed and its
comments — snake_case in the payload, even though the tool description spells them camelCase.
Page with perPage + after until pageInfo.hasNextPage is false; an unresolved thread on page
two counts exactly as much as one on page one.
The gh api fallback can't do this: REST /pulls/{pr}/comments returns flat comment records with
no thread resolution state. If you're on the fallback path, get it from GraphQL:
gh api graphql -f query='query($o:String!,$r:String!,$n:Int!){repository(owner:$o,name:$r){
pullReque
Read the complete source ↗