pr-status

name	pr-status
description	Check PR latest review status.
user-invocable	true
allowed-tools	["Bash"]

pr-status

Report a PR's review status honestly: based on the most recent review, with its findings actually parsed — never on an earlier cached "verdict." A newer review may have landed since (from the @claude bot, a human, or a re-trigger), and it may carry findings the old one missed.

When this fires

• "what's the status of PR #N", "is #N ready to merge", "is the review clean".
• Before you would state, anywhere, that a PR is merge-ready, clean, or ready to merge. GitHub's mergeable field is conflict existence, not a merge license. mergeStateStatus: CLEAN is conflict-free (GitHub mergeable) plus passing commit status, not a review verdict. Do not describe a PR as merge-ready, ready to merge, or "green and merge-ready" unless it is fully clean, including an authentic clean review verdict on HEAD.
• Any other question about a live PR this session is driving, even when the user never said "status": "why didn't you wait", "did you fix it", "why haven't you responded to that comment". Fetch the latest review and CI before answering. The review may have landed while you were answering a different process question. (gha#511, 2026-08-18: the session answered a CI-wait question from chat and never opened the PR thread, so a Needs more work review sat unanswered.)

Commands below are annotated with their abstract operation token (e.g. VIEW_PR, PR_CHECKS) — resolve to your model's tool via tool-mappings.md instead of the gh command shown if this session doesn't have gh.

Verify the PR is still open first

Before checking CI or review, confirm the PR hasn't merged or closed since you last looked, and fetch its branch name and draft state:

gh pr view <N> --json state,title,isDraft,headRefName --jq '"\(.state): \(.title) (draft: \(.isDraft), branch: \(.headRefName))"'   # VIEW_PR

• OPEN → proceed with CI and review checks below.
• MERGED → stop; trigger post-merge instead of reporting CI details.
• CLOSED → stop; report the actual state to the user.

A PR can merge between a "status?" call and a follow-up "status?" in the same session. Running gh pr checks on a merged PR returns stale data and delays noticing the merge happened.

CI green ≠ review clean

gh pr checks <N> (PR_CHECKS) going green is about CI state, not the review verdict. A PR can have all checks passing and still have unaddressed review findings. Always parse the latest review body for findings — don't infer "clean" from green checks.

Read the LATEST review, checked for currency

gh pr view "<N>" --json comments,commits,headRefOid,author,reviewRequests \
  --jq '{
    author: .author.login,
    reviewRequests: [.reviewRequests[].login],
    review: ([.comments[] | select(.author.login | startswith("claude"))] | last | {url: .url, body: .body, createdAt: .createdAt}),
    lastCommitDate: (.commits[-1].committedDate),
    headRefOid: .headRefOid
  }'

This call fetches more than READ_PR_COMMENTS maps to -- tool-mappings.md maps that token to a comments-only MCP call (pull_request_read(method=get_comments)), which returns neither commits nor headRefOid. In a remote/MCP session without gh, fetch those two fields with a separate call (e.g. pull_request_read(method=get) for headRefOid, plus the commits list) rather than assuming the token mapping covers this expanded query.

If .review.createdAt is earlier than .lastCommitDate, the review predates the latest push -- treat it as stale, not current, regardless of what its body says (both are ISO 8601 UTC timestamps, so a plain string comparison works). This timing check alone is not proof of currency: committedDate is the commit's local committer timestamp, not when GitHub received the push, so a commit authored earlier but pushed later can pass the timing check while still being newer than the review. When the review body names the commit it reviewed (the @claude bot commonly writes "commit <sha>"), cross-check that SHA's prefix against .headRefOid. Require a SHA match to call it clean -- when the review passes the timing check but names no SHA (or the mentioned SHA doesn't match), report unverified, not clean, every time, not just when the gap "looks small" (there's no reliable way to judge that from the timing check alone).

The reviewer's bot login varies by API and setup:

• gh pr view reports it as claude
• the REST API (gh api .../comments) reports it as claude[bot]
• some setups post reviews as github-actions[bot]

startswith("claude") matches the @claude bot across both gh pr view and gh api. If your reviewer posts under a different login (e.g. github-actions[bot]), broaden the filter — otherwise the --jq returns null and you silently false-pass a PR with open findings. (Structured MCP GitHub tools like mcp__github_ci__get_ci_status are an alternative where the gh JSON parsing gets fragile.)

Check for a genuine external verdict, not just self-review

The @claude comment above isn't the whole picture. Per fully-clean.md, reporting a PR clean/ready requires a genuine posted verdict at the current head from an external reviewer, whenever one is reachable -- a self-review (posted when @claude was skipped or stubbed) is a fallback, never a substitute, once an external reviewer becomes available again. Formal reviews (e.g. Copilot) don't show up in the comments query above at all -- they're a separate review object.

This is a status query -- inspect an existing Copilot or human review, don't request one. Requesting a review is a mutation: it triggers a review job, consumes reviewer quota, and can collide with an active ardi loop driving the same PR. Use the read-only half of ardi's step 2 -- fetch the matched review's body + inline comments at the current commit_id and require a zero-findings verdict -- but skip the POST /requested_reviewers call. If no genuine Copilot verdict exists at the current head, check for a human's formal review at the head (next subsection) before reporting no verdict at head; only when neither exists, report that and offer to run ardi (which can request one); don't request it yourself here. Green CI plus a clean self-review is not sufficient on its own if an external reviewer is reachable.

A human's formal review at the current head is an external verdict too

The Copilot query above matches only Copilot's own login, so it cannot see a review a real person submitted through GitHub's review UI. Without this check, a PR a human already reviewed at the current head gets reported no verdict at head -- the gap #668 tracks. Fetch the formal reviews and keep the non-bot ones at the current head (READ_PR_REVIEWS):

set -o pipefail
head="$(gh pr view "<N>" --json headRefOid -q .headRefOid)"   # VIEW_PR
gh api "repos/<owner>/<repo>/pulls/<N>/reviews" --paginate \
  | jq -s --arg h "$head" \
  '[.[][] | select(.user.type == "User" and .commit_id == $h
                   and .state != "DISMISSED")]
   | group_by(.user.login)
   | map(sort_by(.submitted_at) | last
         | {id, login: .user.login, state, submitted_at})'

Exclude DISMISSED reviews before reading anything. GitHub's dismiss action flips the review's own state in place rather than adding a new review, and it retracts neither the review's body nor its inline threads -- so without the exclusion, a dismissed review is still its reviewer's latest at the head, and a substance read would report findings an explicit dismissal already resolved. Dismissal is one of the two resolution paths the Check for a blocking human CHANGES_REQUESTED section below already documents; this filter keeps the two checks consistent about it.

Reduce per reviewer, not across all reviewers at once. The group_by(.user.login) mirrors the CHANGES_REQUESTED check below, and for the same reason: two humans can review the same head, and a bare | last over the combined list would return only the chronologically latest review -- so a later clean "LGTM" from one reviewer would silently drop an earlier reviewer's body-only findings (inline findings would still surface through the thread count, but a finding stated only in a review body has no thread to catch it). The command returns each human reviewer's latest review at the head; read every one of them.

Filter on .user.type == "User", not on a login list. A bot's REST user object carries type: "Bot", so the type field separates humans from every bot without maintaining a login blocklist (measured 2026-08-15 on this repo: copilot-pull-request-reviewer[bot]'s review objects report type: "Bot", and the human reviewers' report type: "User").

Judge each matched review by its substance, not its state. Keying on state == "APPROVED" reads as the obvious check and on this repo would never fire: 106 of 106 formal reviews across 60 merged PRs are COMMENTED, zero APPROVED, humans included (measured 2026-07-30 on #668), because reviews here are posted as comments rather than through GitHub's approve flow. A check that can never fire is worse than none -- it reports no verdict at head on PRs a human did review, and the failure is invisible, since a check that never fires looks the same as a repo with no human reviews. So read each matched review the way the Copilot step reads its own: fetch its body plus its inline comments at that commit_id, and apply the same zero-findings bar. Clean-by-substance human reviews at the current head, with no matched review carrying findings, are a genuine external verdict; any matched review with findings is open items to report, whatever the other reviewers said. An APPROVED state, where a repo's convention does produce one, still qualifies -- it just cannot be the key. A CHANGES_REQUESTED stays blocking via the Check for a blocking human CHANGES_REQUESTED section below, regardless of anything this subsection finds.

Parse for findings before declaring clean

Read the full latest review body and scan for any "Findings", "Issues", "Remaining", "Non-blocking", "Minor", "Could improve", "Consider", etc. section. The bar for reporting clean: "Looks good" / "no findings" / "approved" with zero follow-on bullets under any heading. A posted rebuttal the reviewer is still disputing is open, not clean — a rebuttal counts only once it convinced the reviewer (they dropped the item).

Check for a blocking human CHANGES_REQUESTED

A bot's clean verdict does not clear a human's formal review state. A CHANGES_REQUESTED review submitted via GitHub's review UI is invisible to the comments query above -- it's a separate object, and often has an empty top-level body with the actual finding in an inline comment (READ_PR_REVIEWS -- abstract operation token; resolve to your model's tool via tool-mappings.md):

gh pr view "<N>" --json reviews \
  --jq '[.reviews[] | select(.author.login != null and (.state == "APPROVED" or .state == "CHANGES_REQUESTED" or .state == "DISMISSED"))] | group_by(.author.login) | map(sort_by(.submittedAt) | last) | .[] | select(.state == "CHANGES_REQUESTED") | "\(.author.login) \(.submittedAt)"'

--json reviews returns the full review history, not one entry per reviewer, and a reviewer's decisive state persists across neutral comments -- GitHub only clears CHANGES_REQUESTED when that same reviewer later APPROVEDs, or via an explicit dismissal; a neutral COMMENTED review in between does not clear it (verified against GitHub's own docs and community reports). Filtering to only APPROVED/ CHANGES_REQUESTED/DISMISSED states before reducing to each author's latest review is required -- reducing over all states first would let a later COMMENTED round hide an earlier CHANGES_REQUESTED (verified with a synthetic fixture: a CHANGES_REQUESTED followed by a COMMENTED from the same author incorrectly produced no output under the naive reduction, and correctly still blocked once state-filtered first). DISMISSED must stay in the filter, not just APPROVED/CHANGES_REQUESTED -- an explicit dismissal is itself one of the two ways a CHANGES_REQUESTED clears, so dropping it before the per-author reduction lets an older CHANGES_REQUESTED win over its own later dismissal (verified with a second synthetic fixture: CHANGES_REQUESTED then DISMISSED from the same author incorrectly still returned the author as blocking under an APPROVED/CHANGES_REQUESTED-only filter, and correctly returned nothing once DISMISSED was included in the filter). The final select(.state == "CHANGES_REQUESTED") still excludes DISMISSED reviews from the blocking list -- only its presence in the pre-reduction filter matters.

If this returns anything, the PR is blocking regardless of what any bot says -- only the human (or an explicit dismissal) resolves it. Report it as open and name the reviewer; don't let a later "Ready for merge" bot comment paper over it.

A PR is only fully clean / ready to merge when at least one of the @claude comment or an external reviewer's verdict (see Check for a genuine external verdict above) is clean at the current head -- the canonical rule needs one genuine external verdict, not both; if a reachable external reviewer hasn't posted a current-head verdict at all, that's no verdict at head, not automatically a fail, but it means the @claude comment alone has to carry the "clean" claim and no human CHANGES_REQUESTED review is outstanding and all CI workflows are green and every inline review thread is resolved (the only open conversation being the final all-clear and your reply to it — see Check thread-resolution state below). Do not report "ready to merge with one minor nit noted" / "harmless as-is" / "can address if you want" — that hedging just pushes triage back to the user. If there are open items, report them as open (and offer to run ardi to clear them).

Check thread-resolution state

A clean review body isn't the whole bar — unresolved inline threads count as open too. Count the unresolved ones via GraphQL:

gh api graphql -f query='query {
  repository(owner:"<owner>", name:"<repo>") {
    pullRequest(number:<N>) {
      reviewThreads(first:100) {
        totalCount
        nodes { isResolved }
      }
    }
  }
}' --jq '.data.repository.pullRequest.reviewThreads as $rt |
  ($rt.nodes | map(select(.isResolved | not)) | length) as $open |
  if $rt.totalCount > ($rt.nodes | length)
  then "\($open)+ open (totalCount \($rt.totalCount); cap reached — may undercount)"
  else "\($open)"
  end'

Interpret the output as:

• 0 — all threads resolved; clean on this dimension.
• A plain non-zero number (e.g. 3) — that many threads are unresolved.
• A +-suffixed string (e.g. 0+ open (totalCount 150; cap reached — may undercount)) — the 100-thread cap was hit. Cannot confirm clean, even if the visible count is 0; treat as unresolved until the cap is lifted or the PR is confirmed clean another way.

(The resolve mutation lives in the ard skill, step 4b.)

Check if the branch is behind main

Compare remote-tracking refs to see if main has moved ahead of the branch:

git fetch origin main <headRefName> -q && git rev-list --count origin/<headRefName>..origin/main

• 0 -- up to date with main.
• >0 -- behind main by that many commits (offer sync-pr-branch).

Output

Render a Review Summary Table for the PR:

PR	Author	AI Review Verdict	CI State	Reviewers Requested	Next Step
#	<author>	✅ Clean (Round N)	🟢 All Green	the repository owner	Ready for human review

• PR --- markdown link [#<N>](https://github.com/<owner>/<repo>/pull/<N>).
• Author --- author login.
• AI Review Verdict --- hyperlinked directly to the latest review comment URL (e.g. [✅ Clean (Round N)](https://github.com/...#issuecomment-...)). Verified current with the latest commit (.createdAt >= .lastCommitDate and matching commit SHA). If the review predates the latest push, display [⏳ In-Flight / Stale](url). If no SHA is named, display [⚠️ Unverified](url).
• CI State --- 🟢 All Green / ❌ Failing (<name>) / ⏳ Pending (<name>).
• Reviewers Requested --- evaluates human review status per copilot-review-before-human.md. If human review has requested changes, flag ❌ Changes requested by <login>. For self-authored PRs, note *Self-authored*. When AI review is clean and CI is green, list requested reviewers (e.g. the repository owner) or flag ⚠️ None (Request human review). When AI review is clean but CI is failing or pending, display - (CI in progress / failing). When AI review is in-flight or unclean, display - (AI review in progress).
• Next Step --- computed deterministically using the full state matrix:
  • If isDraft: Draft (Work in progress).
  • If human CHANGES_REQUESTED is pending: Blocked on human changes (<login>) (overrides everything below).
  • If branch is behind main: Resolve conflicts / Sync main (<N> commits behind).
  • If CI is failing: Fix CI (<failing-check>).
  • If unaddressed review threads remain: Resolve inline threads (<N> open).
  • If AI review or External review has open findings: Drive to clean (ARDI).
  • If AI review is running: In-flight AI review.
  • If CI is pending: Wait for CI (<pending-check>).
  • If neither AI review nor External review has a verified clean verdict at head: Confirm review (no verified verdict at head).

State, plainly: the latest review's verdict, who/what posted it, and the list of any open findings (or "none"). If you read null, say the filter didn't match a reviewer login — don't report it as clean. A PR without a clean review verdict on the latest commit is not merge-ready. Do not describe it as merge-ready. Report it as blocked on review. Extended operational rationale and empirical measurement histories live in pr-status-all.rationale.md.