[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7356XRBFNN4QVGXQ1S6MTFA":3,"skill-same:01M7356XRBFNN4QVGXQ1S6MTFA":48},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":27,"commitSha":28,"skillDirectory":29,"forksCount":30,"repositoryHtmlUrl":31,"repositoryHomepageUrl":32,"ownerHtmlUrl":33,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":34,"install":35,"resources":47},"01M7356XRBFNN4QVGXQ1S6MTFA","pr-status",null,"Check PR latest review status.","Morrison-Lab\u002Fai-config","Morrison-Lab","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F104235520?v=4","01M734G0451S2G35MN4HFQK7XP","Morrison Lab","01M734G059BYFDFQA8YTPDGPMH",4,"MIT","https:\u002F\u002Fgithub.com\u002FMorrison-Lab\u002Fai-config\u002Fblob\u002Fmain\u002Fskills\u002Fpr-status\u002FSKILL.md",[],[],"2026-11-11 15:11:14",false,"---\nname: pr-status\ndescription: \"Check PR latest review status.\"\nuser-invocable: true\nallowed-tools:\n  - Bash\n---\n\n# pr-status\n\nReport a PR's review status honestly: based on the **most recent** review, with\nits findings actually parsed — never on an earlier cached \"verdict.\" A newer\nreview may have landed since (from the `@claude` bot, a human, or a\nre-trigger), and it may carry findings the old one missed.\n\n## When this fires\n\n- \"what's the status of PR #N\", \"is #N ready to merge\", \"is the review clean\".\n- Before you would state, anywhere, that a PR is merge-ready, clean, or ready to merge.\n  GitHub's `mergeable` field is conflict existence, not a merge license.\n  `mergeStateStatus: CLEAN` is conflict-free (GitHub `mergeable`) plus passing commit status, not a review verdict.\n  Do not describe a PR as merge-ready, ready to merge, or \"green and merge-ready\"\n  unless it is fully clean,\n  including an authentic clean review verdict on HEAD.\n- Any other question about a live PR this session is driving, even when\n  the user never said \"status\": \"why didn't you wait\", \"did you fix it\",\n  \"why haven't you responded to that comment\".\n  Fetch the latest review **and** CI before answering.\n  The review may have landed while you were answering a different process\n  question.\n  (gha#511, 2026-08-18: the session answered a CI-wait question from chat\n  and never opened the PR thread, so a Needs more work review sat\n  unanswered.)\n\nCommands below are annotated with their abstract operation token (e.g.\n`VIEW_PR`, `PR_CHECKS`) — resolve to your model's tool via\n[`tool-mappings.md`](..\u002F..\u002Ftool-mappings.md) instead of the `gh` command shown\nif this session doesn't have `gh`.\n\n## Check the transcript for stalled turns or dropped balls\n\nBefore querying the forge, inspect recent session turns in the conversation transcript:\n- Did a push, review request, or CI check error out without being retried?\n- Did a PR monitoring loop, subagent, or scheduled timer fail or stop unexpectedly?\n- Did an earlier turn leave findings unaddressed or a merge unperformed after a clean verdict?\n\nIf the session got stuck, frozen, or dropped the ball on the PR lifecycle, diagnose the stall and resume the next concrete step immediately in the same turn.\n\n## Verify the PR is still open first\n\nBefore checking CI or review, confirm the PR hasn't merged or closed since you last looked, and fetch its branch name and draft state:\n\n```bash\ngh pr view \u003CN> --json state,title,isDraft,headRefName --jq '\"\\(.state): \\(.title) (draft: \\(.isDraft), branch: \\(.headRefName))\"'   # VIEW_PR\n```\n\n- `OPEN` → proceed with CI and review checks below.\n- `MERGED` → stop; trigger `post-merge` instead of reporting CI details.\n- `CLOSED` → stop; report the actual state to the user.\n\nA PR can merge between a \"status?\" call and a follow-up \"status?\" in the\nsame session. Running `gh pr checks` on a merged PR returns stale data and\ndelays noticing the merge happened.\n\n## CI green ≠ review clean\n\n`gh pr checks \u003CN>` (`PR_CHECKS`) going green is about **CI state**, not the review verdict. A\nPR can have all checks passing and still have unaddressed review findings.\nAlways parse the latest **review body** for findings — don't infer \"clean\"\nfrom green checks.\n\n## Read the LATEST review, checked for currency\n\n```bash\ngh pr view \"\u003CN>\" --json comments,commits,headRefOid,author,reviewRequests \\\n  --jq '{\n    author: .author.login,\n    reviewRequests: [.reviewRequests[].login],\n    review: ([.comments[] | select((.author.login \u002F\u002F \"\") | startswith(\"claude\"))] | last | {url: .url, body: .body, createdAt: .createdAt}),\n    lastCommitDate: (.commits[-1].committedDate),\n    headRefOid: .headRefOid\n  }'\n```\n\n**This call fetches more than `READ_PR_COMMENTS` maps to** --\n`tool-mappings.md` maps that token to a comments-only MCP call\n(`pull_request_read(method=get_comments)`), which returns neither `commits`\nnor `headRefOid`. In a remote\u002FMCP session without `gh`, fetch those two\nfields with a separate call (e.g. `pull_request_read(method=get)` for\n`headRefOid`, plus the commits list) rather than assuming the token mapping\ncovers this expanded query.\n\n**If `.review.createdAt` is earlier than `.lastCommitDate`, the review\npredates the latest push** -- treat it as stale, not current, regardless of\nwhat its body says (both are ISO 8601 UTC timestamps, so a plain string\ncomparison works). This timing check alone is **not proof of currency**:\n`committedDate` is the commit's local committer timestamp, not when GitHub\nreceived the push, so a commit authored earlier but pushed later can pass\nthe timing check while still being newer than the review. When the review\nbody names the commit it reviewed (the `@claude` bot commonly writes\n\"commit `\u003Csha>`\"), cross-check that SHA's prefix against `.headRefOid`.\n**Require a SHA match to call it `clean`** -- when the review passes the\ntiming check but names no SHA (or the mentioned SHA doesn't match), report\n**`unverified`**, not clean, every time, not just when the gap \"looks\nsmall\" (there's no reliable way to judge that from the timing check alone).\n\nThe reviewer's bot login **varies by API and setup**:\n\n- `gh pr view` reports it as `claude`\n- the REST API (`gh api ...\u002Fcomments`) reports it as `claude[bot]`\n- some setups post reviews as `github-actions[bot]`\n\n`startswith(\"claude\")` matches the @claude bot across both `gh pr view` and\n`gh api`. If your reviewer posts under a different login (e.g.\n`github-actions[bot]`), **broaden the filter** — otherwise the `--jq` returns\n`null` and you silently false-pass a PR with open findings. (Structured MCP\nGitHub tools like `mcp__github_ci__get_ci_status` are an alternative where the\n`gh` JSON parsing gets fragile.)\n\n## Check for a genuine external verdict, not just self-review\n\nThe `@claude` comment above isn't the whole picture. Per\n[`fully-clean.md`](..\u002F..\u002Fshared\u002Fworkflow\u002Ffully-clean.md), reporting a PR\nclean\u002Fready requires a **genuine posted verdict at the current head from an\nexternal reviewer, whenever one is reachable** -- a self-review (posted when\n`@claude` was skipped or stubbed) is a fallback, never a substitute, once an\nexternal reviewer becomes available again. Formal reviews (e.g. Copilot)\ndon't show up in the comments query above at all -- they're a separate\nreview object.\n\n**This is a status query -- inspect an existing Copilot or human review,\ndon't request one.**\nRequesting a review is a mutation: it triggers a review job,\nconsumes reviewer quota, and can collide with an active `ardi` loop driving\nthe same PR. Use the read-only half of\n[`ardi`'s step 2](..\u002Fardi\u002FSKILL.md) -- fetch the matched review's body +\ninline comments at the current `commit_id` and require a zero-findings\nverdict -- but skip the `POST \u002Frequested_reviewers` call. If no genuine\nCopilot verdict exists at the current head, check for a human's formal\nreview at the head (next subsection) before reporting `no verdict at head`;\nonly when neither exists, report that and\noffer to run `ardi` (which can request one); don't request it yourself here.\nGreen CI plus a clean self-review is not sufficient on its own if an\nexternal reviewer is reachable.\n\n### A human's formal review at the current head is an external verdict too\n\nThe Copilot query above matches only Copilot's own login,\nso it cannot see a review a real person submitted through GitHub's review UI.\nWithout this check, a PR a human already reviewed at the current head\ngets reported `no verdict at head` --\nthe gap #668 tracks.\nFetch the formal reviews and keep the non-bot ones at the current head\n(`READ_PR_REVIEWS`):\n\n```bash\nset -o pipefail\nhead=\"$(gh pr view \"\u003CN>\" --json headRefOid -q .headRefOid)\"   # VIEW_PR\ngh api \"repos\u002F\u003Cowner>\u002F\u003Crepo>\u002Fpulls\u002F\u003CN>\u002Freviews\" --paginate \\\n  | jq -s --arg h \"$head\" \\\n  '[.[][] | select(.user.type == \"User\" and .commit_id == $h\n                   and .state != \"DISMISSED\")]\n   | group_by(.user.login)\n   | map(sort_by(.submitted_at) | last\n         | {id, login: .user.login, state, submitted_at})'\n```\n\n**Exclude `DISMISSED` reviews before reading anything.**\nGitHub's dismiss action flips the review's own `state` in place\nrather than adding a new review,\nand it retracts neither the review's body nor its inline threads --\nso without the exclusion,\na dismissed review is still its reviewer's latest at the head,\nand a substance read would report findings\nan explicit dismissal already resolved.\nDismissal is one of the two resolution paths\nthe *Check for a blocking human CHANGES_REQUESTED* section below\nalready documents;\nthis filter keeps the two checks consistent about it.\n\n**Reduce per reviewer, not across all reviewers at once.**\nThe `group_by(.user.login)` mirrors the `CHANGES_REQUESTED` check below,\nand for the same reason:\ntwo humans can review the same head,\nand a bare `| last` over the combined list\nwould return only the chronologically latest review --\nso a later clean \"LGTM\" from one reviewer\nwould silently drop an earlier reviewer's body-only findings\n(inline findings would still surface through the thread count,\nbut a finding stated only in a review body has no thread to catch it).\nThe command returns each human reviewer's latest review at the head;\nread **every** one of them.\n\n**Filter on `.user.type == \"User\"`, not on a login list.**\nA bot's REST user object carries `type: \"Bot\"`,\nso the type field separates humans from every bot\nwithout maintaining a login blocklist\n(measured 2026-08-15 on this repo:\n`copilot-pull-request-reviewer[bot]`'s review objects report `type: \"Bot\"`,\nand the human reviewers' report `type: \"User\"`).\n\n**Judge each matched review by its substance, not its `state`.**\nKeying on `state == \"APPROVED\"` reads as the obvious check\nand on this repo would never fire:\n106 of 106 formal reviews across 60 merged PRs are `COMMENTED`,\nzero `APPROVED`, humans included\n(measured 2026-07-30 on #668),\nbecause reviews here are posted as comments\nrather than through GitHub's approve flow.\nA check that can never fire is worse than none --\nit reports `no verdict at head` on PRs a human did review,\nand the failure is invisible,\nsince a check that never fires looks the same as a repo with no human\nreviews.\nSo read each matched review the way the Copilot step reads its own:\nfetch its body plus its inline comments at that `commit_id`,\nand apply the same zero-findings bar.\nClean-by-substance human reviews at the current head,\nwith no matched review carrying findings,\nare a genuine external verdict;\nany matched review with findings is open items to report,\nwhatever the other reviewers said.\nAn `APPROVED` state, where a repo's convention does produce one,\nstill qualifies -- it just cannot be the key.\nA `CHANGES_REQUESTED` stays blocking\nvia the *Check for a blocking human CHANGES_REQUESTED* section below,\nregardless of anything this subsection finds.\n\n## Parse for findings before declaring clean\n\nRead the full latest review body and scan for any \"Findings\", \"Issues\",\n\"Remaining\", \"Non-blocking\", \"Minor\", \"Could improve\", \"Consider\", etc.\nsection. The bar for reporting **clean**: \"Looks good\" \u002F \"no findings\" \u002F\n\"approved\" with **zero** follow-on bullets under any heading. A posted rebuttal\nthe reviewer is still disputing is **open**, not clean — a rebuttal counts only\nonce it convinced the reviewer (they dropped the item).\n\nA collapsed suppression block carries findings too, and none of those keywords finds one.\nMatch it case-insensitively on `suppressed` in a `\u003Csummary>` element or in an ATX heading inside a collapsed `\u003Cdetails>` region --- not on either literal phrase, not on `\u003Csummary>` alone, and not anywhere in the body.\n[`fully-clean.md`](..\u002F..\u002Fshared\u002Fworkflow\u002Ffully-clean.md)'s *A clean overview can hide a collapsed findings block* carries that matcher and the measurements behind it.\n\n- **Do:** run the heading match over every review body, whatever its overview says.\n- **Don't:** treat the keyword list above as the whole scan --- a suppression block's heading contains none of those words.\n\n## Check for a blocking human CHANGES_REQUESTED\n\n**List every review first: this section's query is deliberately narrow, and its narrowness is not the clean bar.**\nThat query answers one question --- whether a GitHub review *state* is blocking the merge button --- and that question is indifferent to a bot's `COMMENTED` review carrying real findings in its body, which the forge never blocks and which vetoes merge under [`fully-clean`](..\u002F..\u002Fshared\u002Fworkflow\u002Ffully-clean.md) all the same.\nThe blind spot is about *where the finding sits*, not about who posted it or which state they picked, so take the unfiltered listing of every review's state, author, and body before narrowing to the state question below:\n\n```bash\ngh pr view \"\u003CN>\" --json reviews \\\n  --jq '.reviews[] | [.state, .author.login, .submittedAt,\n          ((.body \u002F\u002F \"\") | split(\"\\n\") | map(select(length > 0)) | .[0] \u002F\u002F \"(empty body)\")] | @tsv'\n```\n\nOne line per review keeps a long PR's review history readable, and the last field is the body's first non-empty line, or `(empty body)` when the review has no body at all.\nThat line identifies a review rather than settling it, so read the full body of every review the listing names --- not only those whose first line looks unclean, since \"generated no new comments\" is the overview a suppression block hides behind --- and read a collapsed suppression block inside it as findings (see *Parse for findings before declaring clean* above for the matcher).\n\n- **Do:** list every review's state and body, whoever posted it, before running the state-filtered query below.\n- **Do:** read the full body of every review the listing names, whatever its first line says.\n- **Don't:** read an empty `CHANGES_REQUESTED` result as evidence that no review carries findings.\n- **Don't:** gate the full-body read on a review's first line reading unclean --- a plainly clean first line is the false-clean shape.\n\nA bot's clean verdict does **not** clear a human's formal review state. A\n`CHANGES_REQUESTED` review submitted via GitHub's review UI is invisible to\nthe `--json comments` query in *Read the LATEST review* above -- it's a\nseparate object, and often has an\n**empty** top-level body with the actual finding in an inline comment\n(`READ_PR_REVIEWS` -- abstract operation token; resolve to your model's\ntool via [`tool-mappings.md`](..\u002F..\u002Ftool-mappings.md)):\n\n```bash\ngh pr view \"\u003CN>\" --json reviews \\\n  --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)\"'\n```\n\n**`--json reviews` returns the full review history, not one entry per\nreviewer, and a reviewer's *decisive* state persists across neutral\ncomments** -- GitHub only clears `CHANGES_REQUESTED` when that same\nreviewer later `APPROVED`s, or via an explicit dismissal; a neutral\n`COMMENTED` review in between does **not** clear it (verified against\nGitHub's own docs and community reports). Filtering to only `APPROVED`\u002F\n`CHANGES_REQUESTED`\u002F`DISMISSED` states *before* reducing to each author's\nlatest review is required -- reducing over *all* states first would let a\nlater `COMMENTED` round hide an earlier `CHANGES_REQUESTED` (verified with\na synthetic fixture: a `CHANGES_REQUESTED` followed by a `COMMENTED` from\nthe same author incorrectly produced no output under the naive reduction,\nand correctly still blocked once state-filtered first). **`DISMISSED` must\nstay in the filter, not just `APPROVED`\u002F`CHANGES_REQUESTED`** -- an\nexplicit dismissal is itself one of the two ways a `CHANGES_REQUESTED`\nclears, so dropping it before the per-author reduction lets an older\n`CHANGES_REQUESTED` win over its own later dismissal (verified with a\nsecond synthetic fixture: `CHANGES_REQUESTED` then `DISMISSED` from the\nsame author incorrectly still returned the author as blocking under an\n`APPROVED`\u002F`CHANGES_REQUESTED`-only filter, and correctly returned nothing\nonce `DISMISSED` was included in the filter). The final `select(.state ==\n\"CHANGES_REQUESTED\")` still excludes `DISMISSED` reviews from the\nblocking list -- only its *presence* in the pre-reduction filter matters.\n\nIf this returns anything, the PR is **blocking** regardless of what any bot\nsays -- only the human (or an explicit dismissal) resolves it. Report it as\nopen and name the reviewer; don't let a later \"Ready for merge\" bot comment\npaper over it.\n\nA PR is only **fully clean \u002F ready to merge** when **at least one** of the\n`@claude` comment or an external reviewer's verdict (see *Check for a\ngenuine external verdict* above) is clean at the current head -- the\ncanonical rule needs one genuine external verdict, not both; if a reachable\nexternal reviewer hasn't posted a current-head verdict at all, that's `no\nverdict at head`, not automatically a fail, but it means the `@claude`\ncomment alone has to carry the \"clean\" claim *and* no human\n`CHANGES_REQUESTED` review is outstanding *and* all CI\nworkflows are green *and* every inline review thread is resolved (the\nonly open conversation being the final all-clear and your reply to it — see\n*Check thread-resolution state* below). Do **not** report \"ready to merge with\none minor nit noted\" \u002F \"harmless as-is\" \u002F \"can address if you want\" — that\nhedging just pushes triage back to the user. If there are open items, report\nthem as open (and offer to run `ardi` to clear them).\n\n## Check thread-resolution state\n\nA clean review *body* isn't the whole bar — unresolved inline threads count as\nopen too. Count the unresolved ones via GraphQL:\n\n```bash\ngh api graphql -f query='query {\n  repository(owner:\"\u003Cowner>\", name:\"\u003Crepo>\") {\n    pullRequest(number:\u003CN>) {\n      reviewThreads(first:100) {\n        totalCount\n        nodes { isResolved }\n      }\n    }\n  }\n}' --jq '.data.repository.pullRequest.reviewThreads as $rt |\n  ($rt.nodes | map(select(.isResolved | not)) | length) as $open |\n  if $rt.totalCount > ($rt.nodes | length)\n  then \"\\($open)+ open (totalCount \\($rt.totalCount); cap reached — may undercount)\"\n  else \"\\($open)\"\n  end'\n```\n\nInterpret the output as:\n\n- `0` — all threads resolved; clean on this dimension.\n- A plain non-zero number (e.g. `3`) — that many threads are unresolved.\n- 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.\n\n(The resolve mutation lives in the `ard` skill, step 4b.)\n\n## Check if the branch is behind main\n\nCompare remote-tracking refs to see if `main` has moved ahead of the branch:\n\n```bash\ngit fetch origin main \u003CheadRefName> -q && git rev-list --count origin\u002F\u003CheadRefName>..origin\u002Fmain\n```\n\n- `0` -- up to date with main.\n- `>0` -- behind main by that many commits (offer `sync-pr-branch`).\n\n## Output\n\nRender a **Review Summary Table** for the PR:\n\n| PR | Author | AI Review Verdict | CI State | Reviewers Requested | Next Step |\n|:---|:---|:---:|:---:|:---:|:---|\n| [#\u003CN>](https:\u002F\u002Fgithub.com\u002F\u003Cowner>\u002F\u003Crepo>\u002Fpull\u002F\u003CN>) | `\u003Cauthor>` | [✅ Clean (Round N)](url) | 🟢 All Green | `the repository owner` | Ready for human review |\n\n- **PR** --- markdown link `[#\u003CN>](https:\u002F\u002Fgithub.com\u002F\u003Cowner>\u002F\u003Crepo>\u002Fpull\u002F\u003CN>)`.\n- **Author** --- author login.\n- **AI Review Verdict** --- hyperlinked directly to the latest review comment URL (e.g. `[✅ Clean (Round N)](https:\u002F\u002Fgithub.com\u002F...#issuecomment-...)`).\n  Verified current with the latest commit (`.createdAt >= .lastCommitDate` and matching commit SHA).\n  If the review predates the latest push, display `[⏳ In-Flight \u002F Stale](url)`.\n  If no SHA is named, display `[⚠️ Unverified](url)`.\n- **CI State** --- `🟢 All Green` \u002F `❌ Failing (\u003Cname>)` \u002F `⏳ Pending (\u003Cname>)`.\n- **Reviewers Requested** --- evaluates human review status per [`copilot-review-before-human.md`](..\u002F..\u002Fshared\u002Fvendored\u002Fcopilot-review-before-human.md).\n  If human review has requested changes, flag `❌ Changes requested by \u003Clogin>`.\n  For self-authored PRs, note `*Self-authored*`.\n  When AI review is clean and CI is green, list requested reviewers (e.g. `the repository owner`) or flag `⚠️ None (Request human review)`.\n  When AI review is clean but CI is failing or pending, display `- (CI in progress \u002F failing)`.\n  When AI review is in-flight or unclean, display `- (AI review in progress)`.\n- **Next Step** --- computed deterministically using the full state matrix:\n  - If `isDraft`: `Draft (Work in progress)`.\n  - If human `CHANGES_REQUESTED` is pending: `Blocked on human changes (\u003Clogin>)` (overrides everything below).\n  - If branch is behind main: `Resolve conflicts \u002F Sync main (\u003CN> commits behind)`.\n  - If CI is failing: `Fix CI (\u003Cfailing-check>)`.\n  - If unaddressed review threads remain: `Resolve inline threads (\u003CN> open)`.\n  - If AI review or External review has open findings: `Drive to clean (ARDI)`.\n  - If AI review is running: `In-flight AI review`.\n  - If CI is pending: `Wait for CI (\u003Cpending-check>)`.\n  - If neither AI review nor External review has a verified clean verdict at head: `Confirm review (no verified verdict at head)`.\n  - If fully clean (no human blocks, at least one verified clean review at head with 0 open findings across all reviews, CI green, 0 open threads, up to date with main):\n    - If `Author` is `the repository owner` (self-authored): `Ready for self-merge`.\n    - If `Author` is external and human review is requested (`the repository owner`): `Ready for human review`.\n    - If `Author` is external and human review is not yet requested: `Request human review`.\n\nState, plainly: the latest review's verdict, who\u002Fwhat posted it, and the list of any open findings (or \"none\").\nIf you read `null`, say the filter didn't match a reviewer login — don't report it as clean.\nA PR without a clean review verdict on the latest commit is not merge-ready.\nDo not describe it as merge-ready.\nReport it as blocked on review.\nExtended operational rationale and empirical measurement histories live in [`pr-status-all.rationale.md`](..\u002Fpr-status-all\u002Fpr-status-all.rationale.md).\n","# pr-status\n\nReport a PR's review status honestly: based on the **most recent** review, with\nits findings actually parsed — never on an earlier cached \"verdict.\" A newer\nreview may have landed since (from the `@claude` bot, a human, or a\nre-trigger), and it may carry findings the old one missed.\n\n## When this fires\n\n- \"what's the status of PR #N\", \"is #N ready to merge\", \"is the review clean\".\n- Before you would state, anywhere, that a PR is merge-ready, clean, or ready to merge.\n  GitHub's `mergeable` field is conflict existence, not a merge license.\n  `mergeStateStatus: CLEAN` is conflict-free (GitHub `mergeable`) plus passing commit status, not a review verdict.\n  Do not describe a PR as merge-ready, ready to merge, or \"green and merge-ready\"\n  unless it is fully clean,\n  including an authentic clean review verdict on HEAD.\n- Any other question about a live PR this session is driving, even when\n  the user never said \"status\": \"why didn't you wait\", \"did you fix it\",\n  \"why haven't you responded to that comment\".\n  Fetch the latest review **and** CI before answering.\n  The review may have landed while you were answering a different process\n  question.\n  (gha#511, 2026-08-18: the session answered a CI-wait question from chat\n  and never opened the PR thread, so a Needs more work review sat\n  unanswered.)\n\nCommands below are annotated with their abstract operation token (e.g.\n`VIEW_PR`, `PR_CHECKS`) — resolve to your model's tool via\n[`tool-mappings.md`](..\u002F..\u002Ftool-mappings.md) instead of the `gh` command shown\nif this session doesn't have `gh`.\n\n## Check the transcript for stalled turns or dropped balls\n\nBefore querying the forge, inspect recent session turns in the conversation transcript:\n- Did a push, review request, or CI check error out without being retried?\n- Did a PR monitoring loop, subagent, or scheduled timer fail or stop unexpectedly?\n- Did an earlier turn leave findings unaddressed or a merge unperformed after a clean verdict?\n\nIf the session got stuck, frozen, or dropped the ball on the PR lifecycle, diagnose the stall and resume the next concrete step immediately in the same turn.\n\n## Verify the PR is still open first\n\nBefore checking CI or review, confirm the PR hasn't merged or closed since you last looked, and fetch its branch name and draft state:\n\n```bash\ngh pr view \u003CN> --json state,title,isDraft,headRefName --jq '\"\\(.state): \\(.title) (draft: \\(.isDraft), branch: \\(.headRefName))\"'   # VIEW_PR\n```\n\n- `OPEN` → proceed with CI and review checks below.\n- `MERGED` → stop; trigger `post-merge` instead of reporting CI details.\n- `CLOSED` → stop; report the actual state to the user.\n\nA PR can merge between a \"status?\" call and a follow-up \"status?\" in the\nsame session. Running `gh pr checks` on a merged PR returns stale data and\ndelays noticing the merge happened.\n\n## CI green ≠ review clean\n\n`gh pr checks \u003CN>` (`PR_CHECKS`) going green is about **CI state**, not the review verdict. A\nPR can have all checks passing and still have unaddressed review findings.\nAlways parse the latest **review body** for findings — don't infer \"clean\"\nfrom green checks.\n\n## Read the LATEST review, checked for currency\n\n```bash\ngh pr view \"\u003CN>\" --json comments,commits,headRefOid,author,reviewRequests \\\n  --jq '{\n    author: .author.login,\n    reviewRequests: [.reviewRequests[].login],\n    review: ([.comments[] | select((.author.login \u002F\u002F \"\") | startswith(\"claude\"))] | last | {url: .url, body: .body, createdAt: .createdAt}),\n    lastCommitDate: (.commits[-1].committedDate),\n    headRefOid: .headRefOid\n  }'\n```\n\n**This call fetches more than `READ_PR_COMMENTS` maps to** --\n`tool-mappings.md` maps that token to a comments-only MCP call\n(`pull_request_read(method=get_comments)`), which returns neither `commits`\nnor `headRefOid`. In a remote\u002FMCP session without `gh`, fetch those two\nfields with a separate call (e.g. `pull_request_read(method=get)` for\n`headRefOid`, plus the commits list) rather than assuming the token mapping\ncovers this expanded query.\n\n**If `.review.createdAt` is earlier than `.lastCommitDate`, the review\npredates the latest push** -- treat it as stale, not current, regardless of\nwhat its body says (both are ISO 8601 UTC timestamps, so a plain string\ncomparison works). This timing check alone is **not proof of currency**:\n`committedDate` is the commit's local committer timestamp, not when GitHub\nreceived the push, so a commit authored earlier but pushed later can pass\nthe timing check while still being newer than the review. When the review\nbody names the commit it reviewed (the `@claude` bot commonly writes\n\"commit `\u003Csha>`\"), cross-check that SHA's prefix against `.headRefOid`.\n**Require a SHA match to call it `clean`** -- when the review passes the\ntiming check but names no SHA (or the mentioned SHA doesn't match), report\n**`unverified`**, not clean, every time, not just when the gap \"looks\nsmall\" (there's no reliable way to judge that from the timing check alone).\n\nThe reviewer's bot login **varies by API and setup**:\n\n- `gh pr view` reports it as `claude`\n- the REST API (`gh api ...\u002Fcomments`) reports it as `claude[bot]`\n- some setups post reviews as `github-actions[bot]`\n\n`startswith(\"claude\")` matches the @claude bot across both `gh pr view` and\n`gh api`. If your reviewer posts under a different login (e.g.\n`github-actions[bot]`), **broaden the filter** — otherwise the `--jq` returns\n`null` and you silently false-pass a PR with open findings. (Structured MCP\nGitHub tools like `mcp__github_ci__get_ci_status` are an alternative where the\n`gh` JSON parsing gets fragile.)\n\n## Check for a genuine external verdict, not just self-review\n\nThe `@claude` comment above isn't the whole picture. Per\n[`fully-clean.md`](..\u002F..\u002Fshared\u002Fworkflow\u002Ffully-clean.md), reporting a PR\nclean\u002Fready requires a **genuine posted verdict at the current head from an\nexternal reviewer, whenever one is reachable** -- a self-review (posted when\n`@claude` was skipped or stubbed) is a fallback, never a substitute, once an\nexternal reviewer becomes available again. Formal reviews (e.g. Copilot)\ndon't show up in the comments query above at all -- they're a separate\nreview object.\n\n**This is a status query -- inspect an existing Copilot or human review,\ndon't request one.**\nRequesting a review is a mutation: it triggers a review job,\nconsumes reviewer quota, and can collide with an active `ardi` loop driving\nthe same PR. Use the read-only half of\n[`ardi`'s step 2](..\u002Fardi\u002FSKILL.md) -- fetch the matched review's body +\ninline comments at the current `commit_id` and require a zero-findings\nverdict -- but skip the `POST \u002Frequested_reviewers` call. If no genuine\nCopilot verdict exists at the current head, check for a human's formal\nreview at the head (next subsection) before reporting `no verdict at head`;\nonly when neither exists, report that and\noffer to run `ardi` (which can request one); don't request it yourself here.\nGreen CI plus a clean self-review is not sufficient on its own if an\nexternal reviewer is reachable.\n\n### A human's formal review at the current head is an external verdict too\n\nThe Copilot query above matches only Copilot's own login,\nso it cannot see a review a real person submitted through GitHub's review UI.\nWithout this check, a PR a human already reviewed at the current head\ngets reported `no verdict at head` --\nthe gap #668 tracks.\nFetch the formal reviews and keep the non-bot ones at the current head\n(`READ_PR_REVIEWS`):\n\n```bash\nset -o pipefail\nhead=\"$(gh pr view \"\u003CN>\" --json headRefOid -q .headRefOid)\"   # VIEW_PR\ngh api \"repos\u002F\u003Cowner>\u002F\u003Crepo>\u002Fpulls\u002F\u003CN>\u002Freviews\" --paginate \\\n  | jq -s --arg h \"$head\" \\\n  '[.[][] | select(.user.type == \"User\" and .commit_id == $h\n                   and .state != \"DISMISSED\")]\n   | group_by(.user.login)\n   | map(sort_by(.submitted_at) | last\n         | {id, login: .user.login, state, submitted_at})'\n```\n\n**Exclude `DISMISSED` reviews before reading anything.**\nGitHub's dismiss action flips the review's own `state` in place\nrather than adding a new review,\nand it retracts neither the review's body nor its inline threads --\nso without the exclusion,\na dismissed review is still its reviewer's latest at the head,\nand a substance read would report findings\nan explicit dismissal already resolved.\nDismissal is one of the two resolution paths\nthe *Check for a blocking human CHANGES_REQUESTED* section below\nalready documents;\nthis filter keeps the two checks consistent about it.\n\n**Reduce per reviewer, not across all reviewers at once.**\nThe `group_by(.user.login)` mirrors the `CHANGES_REQUESTED` check below,\nand for the same reason:\ntwo humans can review the same head,\nand a bare `| last` over the combined list\nwould return only the chronologically latest review --\nso a later clean \"LGTM\" from one reviewer\nwould silently drop an earlier reviewer's body-only findings\n(inline findings would still surface through the thread count,\nbut a finding stated only in a review body has no thread to catch it).\nThe command returns each human reviewer's latest review at the head;\nread **every** one of them.\n\n**Filter on `.user.type == \"User\"`, not on a login list.**\nA bot's REST user object carries `type: \"Bot\"`,\nso the type field separates humans from every bot\nwithout maintaining a login blocklist\n(measured 2026-08-15 on this repo:\n`copilot-pull-request-reviewer[bot]`'s review objects report `type: \"Bot\"`,\nand the human reviewers' report `type: \"User\"`).\n\n**Judge each matched review by its substance, not its `state`.**\nKeying on `state == \"APPROVED\"` reads as the obvious check\nand on this repo would never fire:\n106 of 106 formal reviews across 60 merged PRs are `COMMENTED`,\nzero `APPROVED`, humans included\n(measured 2026-07-30 on #668),\nbecause reviews here are posted as comments\nrather than through GitHub's approve flow.\nA check that can never fire is worse than none --\nit reports `no verdict at head` on PRs a human did review,\nand the failure is invisible,\nsince a check that never fires looks the same as a repo with no human\nreviews.\nSo read each matched review the way the Copilot step reads its own:\nfetch its body plus its inline comments at that `commit_id`,\nand apply the same zero-findings bar.\nClean-by-substance human reviews at the current head,\nwith no matched review carrying findings,\nare a genuine external verdict;\nany matched review with findings is open items to report,\nwhatever the other reviewers said.\nAn `APPROVED` state, where a repo's convention does produce one,\nstill qualifies -- it just cannot be the key.\nA `CHANGES_REQUESTED` stays blocking\nvia the *Check for a blocking human CHANGES_REQUESTED* section below,\nregardless of anything this subsection finds.\n\n## Parse for findings before declaring clean\n\nRead the full latest review body and scan for any \"Findings\", \"Issues\",\n\"Remaining\", \"Non-blocking\", \"Minor\", \"Could improve\", \"Consider\", etc.\nsection. The bar for reporting **clean**: \"Looks good\" \u002F \"no findings\" \u002F\n\"approved\" with **zero** follow-on bullets under any heading. A posted rebuttal\nthe reviewer is still disputing is **open**, not clean — a rebuttal counts only\nonce it convinced the reviewer (they dropped the item).\n\nA collapsed suppression block carries findings too, and none of those keywords finds one.\nMatch it case-insensitively on `suppressed` in a `\u003Csummary>` element or in an ATX heading inside a collapsed `\u003Cdetails>` region --- not on either literal phrase, not on `\u003Csummary>` alone, and not anywhere in the body.\n[`fully-clean.md`](..\u002F..\u002Fshared\u002Fworkflow\u002Ffully-clean.md)'s *A clean overview can hide a collapsed findings block* carries that matcher and the measurements behind it.\n\n- **Do:** run the heading match over every review body, whatever its overview says.\n- **Don't:** treat the keyword list above as the whole scan --- a suppression block's heading contains none of those words.\n\n## Check for a blocking human CHANGES_REQUESTED\n\n**List every review first: this section's query is deliberately narrow, and its narrowness is not the clean bar.**\nThat query answers one question --- whether a GitHub review *state* is blocking the merge button --- and that question is indifferent to a bot's `COMMENTED` review carrying real findings in its body, which the forge never blocks and which vetoes merge under [`fully-clean`](..\u002F..\u002Fshared\u002Fworkflow\u002Ffully-clean.md) all the same.\nThe blind spot is about *where the finding sits*, not about who posted it or which state they picked, so take the unfiltered listing of every review's state, author, and body before narrowing to the state question below:\n\n```bash\ngh pr view \"\u003CN>\" --json reviews \\\n  --jq '.reviews[] | [.state, .author.login, .submittedAt,\n          ((.body \u002F\u002F \"\") | split(\"\\n\") | map(select(length > 0)) | .[0] \u002F\u002F \"(empty body)\")] | @tsv'\n```\n\nOne line per review keeps a long PR's review history readable, and the last field is the body's first non-empty line, or `(empty body)` when the review has no body at all.\nThat line identifies a review rather than settling it, so read the full body of every review the listing names --- not only those whose first line looks unclean, since \"generated no new comments\" is the overview a suppression block hides behind --- and read a collapsed suppression block inside it as findings (see *Parse for findings before declaring clean* above for the matcher).\n\n- **Do:** list every review's state and body, whoever posted it, before running the state-filtered query below.\n- **Do:** read the full body of every review the listing names, whatever its first line says.\n- **Don't:** read an empty `CHANGES_REQUESTED` result as evidence that no review carries findings.\n- **Don't:** gate the full-body read on a review's first line reading unclean --- a plainly clean first line is the false-clean shape.\n\nA bot's clean verdict does **not** clear a human's formal review state. A\n`CHANGES_REQUESTED` review submitted via GitHub's review UI is invisible to\nthe `--json comments` query in *Read the LATEST review* above -- it's a\nseparate object, and often has an\n**empty** top-level body with the actual finding in an inline comment\n(`READ_PR_REVIEWS` -- abstract operation token; resolve to your model's\ntool via [`tool-mappings.md`](..\u002F..\u002Ftool-mappings.md)):\n\n```bash\ngh pr view \"\u003CN>\" --json reviews \\\n  --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)\"'\n```\n\n**`--json reviews` returns the full review history, not one entry per\nreviewer, and a reviewer's *decisive* state persists across neutral\ncomments** -- GitHub only clears `CHANGES_REQUESTED` when that same\nreviewer later `APPROVED`s, or via an explicit dismissal; a neutral\n`COMMENTED` review in between does **not** clear it (verified against\nGitHub's own docs and community reports). Filtering to only `APPROVED`\u002F\n`CHANGES_REQUESTED`\u002F`DISMISSED` states *before* reducing to each author's\nlatest review is required -- reducing over *all* states first would let a\nlater `COMMENTED` round hide an earlier `CHANGES_REQUESTED` (verified with\na synthetic fixture: a `CHANGES_REQUESTED` followed by a `COMMENTED` from\nthe same author incorrectly produced no output under the naive reduction,\nand correctly still blocked once state-filtered first). **`DISMISSED` must\nstay in the filter, not just `APPROVED`\u002F`CHANGES_REQUESTED`** -- an\nexplicit dismissal is itself one of the two ways a `CHANGES_REQUESTED`\nclears, so dropping it before the per-author reduction lets an older\n`CHANGES_REQUESTED` win over its own later dismissal (verified with a\nsecond synthetic fixture: `CHANGES_REQUESTED` then `DISMISSED` from the\nsame author incorrectly still returned the author as blocking under an\n`APPROVED`\u002F`CHANGES_REQUESTED`-only filter, and correctly returned nothing\nonce `DISMISSED` was included in the filter). The final `select(.state ==\n\"CHANGES_REQUESTED\")` still excludes `DISMISSED` reviews from the\nblocking list -- only its *presence* in the pre-reduction filter matters.\n\nIf this returns anything, the PR is **blocking** regardless of what any bot\nsays -- only the human (or an explicit dismissal) resolves it. Report it as\nopen and name the reviewer; don't let a later \"Ready for merge\" bot comment\npaper over it.\n\nA PR is only **fully clean \u002F ready to merge** when **at least one** of the\n`@claude` comment or an external reviewer's verdict (see *Check for a\ngenuine external verdict* above) is clean at the current head -- the\ncanonical rule needs one genuine external verdict, not both; if a reachable\nexternal reviewer hasn't posted a current-head verdict at all, that's `no\nverdict at head`, not automatically a fail, but it means the `@claude`\ncomment alone has to carry the \"clean\" claim *and* no human\n`CHANGES_REQUESTED` review is outstanding *and* all CI\nworkflows are green *and* every inline review thread is resolved (the\nonly open conversation being the final all-clear and your reply to it — see\n*Check thread-resolution state* below). Do **not** report \"ready to merge with\none minor nit noted\" \u002F \"harmless as-is\" \u002F \"can address if you want\" — that\nhedging just pushes triage back to the user. If there are open items, report\nthem as open (and offer to run `ardi` to clear them).\n\n## Check thread-resolution state\n\nA clean review *body* isn't the whole bar — unresolved inline threads count as\nopen too. Count the unresolved ones via GraphQL:\n\n```bash\ngh api graphql -f query='query {\n  repository(owner:\"\u003Cowner>\", name:\"\u003Crepo>\") {\n    pullRequest(number:\u003CN>) {\n      reviewThreads(first:100) {\n        totalCount\n        nodes { isResolved }\n      }\n    }\n  }\n}' --jq '.data.repository.pullRequest.reviewThreads as $rt |\n  ($rt.nodes | map(select(.isResolved | not)) | length) as $open |\n  if $rt.totalCount > ($rt.nodes | length)\n  then \"\\($open)+ open (totalCount \\($rt.totalCount); cap reached — may undercount)\"\n  else \"\\($open)\"\n  end'\n```\n\nInterpret the output as:\n\n- `0` — all threads resolved; clean on this dimension.\n- A plain non-zero number (e.g. `3`) — that many threads are unresolved.\n- 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.\n\n(The resolve mutation lives in the `ard` skill, step 4b.)\n\n## Check if the branch is behind main\n\nCompare remote-tracking refs to see if `main` has moved ahead of the branch:\n\n```bash\ngit fetch origin main \u003CheadRefName> -q && git rev-list --count origin\u002F\u003CheadRefName>..origin\u002Fmain\n```\n\n- `0` -- up to date with main.\n- `>0` -- behind main by that many commits (offer `sync-pr-branch`).\n\n## Output\n\nRender a **Review Summary Table** for the PR:\n\n| PR | Author | AI Review Verdict | CI State | Reviewers Requested | Next Step |\n|:---|:---|:---:|:---:|:---:|:---|\n| [#\u003CN>](https:\u002F\u002Fgithub.com\u002F\u003Cowner>\u002F\u003Crepo>\u002Fpull\u002F\u003CN>) | `\u003Cauthor>` | [✅ Clean (Round N)](url) | 🟢 All Green | `the repository owner` | Ready for human review |\n\n- **PR** --- markdown link `[#\u003CN>](https:\u002F\u002Fgithub.com\u002F\u003Cowner>\u002F\u003Crepo>\u002Fpull\u002F\u003CN>)`.\n- **Author** --- author login.\n- **AI Review Verdict** --- hyperlinked directly to the latest review comment URL (e.g. `[✅ Clean (Round N)](https:\u002F\u002Fgithub.com\u002F...#issuecomment-...)`).\n  Verified current with the latest commit (`.createdAt >= .lastCommitDate` and matching commit SHA).\n  If the review predates the latest push, display `[⏳ In-Flight \u002F Stale](url)`.\n  If no SHA is named, display `[⚠️ Unverified](url)`.\n- **CI State** --- `🟢 All Green` \u002F `❌ Failing (\u003Cname>)` \u002F `⏳ Pending (\u003Cname>)`.\n- **Reviewers Requested** --- evaluates human review status per [`copilot-review-before-human.md`](..\u002F..\u002Fshared\u002Fvendored\u002Fcopilot-review-before-human.md).\n  If human review has requested changes, flag `❌ Changes requested by \u003Clogin>`.\n  For self-authored PRs, note `*Self-authored*`.\n  When AI review is clean and CI is green, list requested reviewers (e.g. `the repository owner`) or flag `⚠️ None (Request human review)`.\n  When AI review is clean but CI is failing or pending, display `- (CI in progress \u002F failing)`.\n  When AI review is in-flight or unclean, display `- (AI review in progress)`.\n- **Next Step** --- computed deterministically using the full state matrix:\n  - If `isDraft`: `Draft (Work in progress)`.\n  - If human `CHANGES_REQUESTED` is pending: `Blocked on human changes (\u003Clogin>)` (overrides everything below).\n  - If branch is behind main: `Resolve conflicts \u002F Sync main (\u003CN> commits behind)`.\n  - If CI is failing: `Fix CI (\u003Cfailing-check>)`.\n  - If unaddressed review threads remain: `Resolve inline threads (\u003CN> open)`.\n  - If AI review or External review has open findings: `Drive to clean (ARDI)`.\n  - If AI review is running: `In-flight AI review`.\n  - If CI is pending: `Wait for CI (\u003Cpending-check>)`.\n  - If neither AI review nor External review has a verified clean verdict at head: `Confirm review (no verified verdict at head)`.\n  - If fully clean (no human blocks, at least one verified clean review at head with 0 open findings across all reviews, CI green, 0 open threads, up to date with main):\n    - If `Author` is `the repository owner` (self-authored): `Ready for self-merge`.\n    - If `Author` is external and human review is requested (`the repository owner`): `Ready for human review`.\n    - If `Author` is external and human review is not yet requested: `Request human review`.\n\nState, plainly: the latest review's verdict, who\u002Fwhat posted it, and the list of any open findings (or \"none\").\nIf you read `null`, say the filter didn't match a reviewer login — don't report it as clean.\nA PR without a clean review verdict on the latest commit is not merge-ready.\nDo not describe it as merge-ready.\nReport it as blocked on review.\nExtended operational rationale and empirical measurement histories live in [`pr-status-all.rationale.md`](..\u002Fpr-status-all\u002Fpr-status-all.rationale.md).\n",{"name":5,"description":7,"allowed-tools":24,"user-invocable":26},[25],"Bash",true,"main","0f720481f8ce3b3d9fef485f6e230d8b3035748e","skills\u002Fpr-status",1,"https:\u002F\u002Fgithub.com\u002FMorrison-Lab\u002Fai-config","https:\u002F\u002Fmorrison-lab.github.io\u002Fai-config\u002F","https:\u002F\u002Fgithub.com\u002FMorrison-Lab","2026-11-15 04:39:15",{"source":8,"skillName":5,"commands":36,"prompt":46,"zipAvailable":26},[37,40,43],{"tool":38,"command":39},"npx","npx skills add Morrison-Lab\u002Fai-config --skill pr-status",{"tool":41,"command":42},"bunx","bunx skills add Morrison-Lab\u002Fai-config --skill pr-status",{"tool":44,"command":45},"pnpm","pnpm dlx skills add Morrison-Lab\u002Fai-config --skill pr-status","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FMorrison-Lab\u002Fai-config\u002Fblob\u002Fmain\u002Fskills\u002Fpr-status\u002FSKILL.md",[],[49,56,63,70,77,84],{"skillNo":50,"name":51,"nameZh":6,"description":52,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":53,"categories":54,"keywords":55,"publishedAt":19},"01M735BVYCGR5VEC65AFDVSNKW","wrap-up","End-of-session verification and UMS.","https:\u002F\u002Fgithub.com\u002FMorrison-Lab\u002Fai-config\u002Fblob\u002Fmain\u002Fskills\u002Fwrap-up\u002FSKILL.md",[],[],{"skillNo":57,"name":58,"nameZh":6,"description":59,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":60,"categories":61,"keywords":62,"publishedAt":19},"01M735BTH469T1CQ8G2Q9GENCR","workaround-watcher","Watch upstream issue for fix.","https:\u002F\u002Fgithub.com\u002FMorrison-Lab\u002Fai-config\u002Fblob\u002Fmain\u002Fskills\u002Fworkaround-watcher\u002FSKILL.md",[],[],{"skillNo":64,"name":65,"nameZh":6,"description":66,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":67,"categories":68,"keywords":69,"publishedAt":19},"01M735BRKZWFANSQ0GM380B4TD","wn","→ whats-next.","https:\u002F\u002Fgithub.com\u002FMorrison-Lab\u002Fai-config\u002Fblob\u002Fmain\u002Fskills\u002Fwn\u002FSKILL.md",[],[],{"skillNo":71,"name":72,"nameZh":6,"description":73,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":19},"01M735BPE9V830B3B6NGXQ56J6","whats-next","Recommend concrete next step or action.","https:\u002F\u002Fgithub.com\u002FMorrison-Lab\u002Fai-config\u002Fblob\u002Fmain\u002Fskills\u002Fwhats-next\u002FSKILL.md",[],[],{"skillNo":78,"name":79,"nameZh":6,"description":80,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":81,"categories":82,"keywords":83,"publishedAt":19},"01M735BMHQ60EAD5B7QAWXC5A0","wait-for-results","Wait for pipeline\u002FCI results.","https:\u002F\u002Fgithub.com\u002FMorrison-Lab\u002Fai-config\u002Fblob\u002Fmain\u002Fskills\u002Fwait-for-results\u002FSKILL.md",[],[],{"skillNo":85,"name":86,"nameZh":6,"description":87,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":88,"categories":89,"keywords":90,"publishedAt":19},"01M735BKGZHPN1PEMJ7YBD5VJ8","wai","→ send-notes.","https:\u002F\u002Fgithub.com\u002FMorrison-Lab\u002Fai-config\u002Fblob\u002Fmain\u002Fskills\u002Fwai\u002FSKILL.md",[],[]]