[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M74PHBW7DMXDTN7J548T7CMD":3,"skill-same:01M74PHBW7DMXDTN7J548T7CMD":49},{"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":26,"commitSha":27,"skillDirectory":28,"forksCount":29,"repositoryHtmlUrl":30,"repositoryHomepageUrl":31,"ownerHtmlUrl":32,"ownerBlogUrl":33,"ownerDescription":6,"updatedAt":34,"install":35,"resources":48},"01M74PHBW7DMXDTN7J548T7CMD","refine-pr-review",null,"Triage a fetched PR review with the user, comment by comment, drafting each reply and producing a REQUIREMENTS file for the accepted code changes. Takes the PR-REVIEW file produced by fetch-pr-review, or a PR URL. Invoke manually only.","eai-org\u002Fagent-toolkit","eai-org","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F301886321?v=4","01M74PFC3HF57T58MQ14P212BQ","Engineering in the age of AI","01M74PFC46MSKTT9SW5Q437FSD",85,"MIT","https:\u002F\u002Fgithub.com\u002Feai-org\u002Fagent-toolkit\u002Fblob\u002Fmain\u002Fskills\u002Frefine-pr-review\u002FSKILL.md",[],[],"2026-11-11 15:11:14",false,"---\nname: refine-pr-review\ndescription: Triage a fetched PR review with the user, comment by comment, drafting each reply and producing a REQUIREMENTS file for the accepted code changes. Takes the PR-REVIEW file produced by fetch-pr-review, or a PR URL. Invoke manually only.\nlicense: MIT\nmetadata:\n  version: \"1.12\"\n---\n\n# Refine PR review\n\nThe **Refine** phase for review feedback: turn a `.PR-REVIEW.md` (from \u002Ffetch-pr-review; ask if\nthe path is ambiguous) into user-settled per-comment verdicts, drafted replies, and a REQUIREMENTS\nfile covering the accepted code changes. Triage and replies are settled here; the plan and execute\nphases see only code changes.\n\nA PR URL instead: when the harness has subagents and fetch-pr-review is available\n(`..\u002Ffetch-pr-review\u002FSKILL.md` beside this skill), a subagent fetches it first per its\n[delegate.md](..\u002Ffetch-pr-review\u002Fdelegate.md) and the file it returns is the input. Not available\n→ say so and ask for the fetched file; available but no subagents → hand off\n`\u002Ffetch-pr-review \u003Curl>` to a fresh session and stop.\n\nA trailing `head=\u003Csha>` token (fetch-pr-review's refresh hand-off) is the head a refresh just\nconfirmed: Phase 1's moved-head check uses it in place of the file's `Head` line.\n\nIts own `.PR-REVIEW.REQUIREMENTS.md` instead, once a blocking answer is in: run Phase 1's checkout\nand remote-head check first — a checkout differing from the ANSWERS `Head`, or a dirty tree, is\nreported and the user decides, never a branch switch; a moved PR head is reported with the\nre-fetch offer and the user decides whether to finish this round first. Verdicts, replies and\nLessons stand as its `.PR-REVIEW.ANSWERS.md` records them. Apply the answer: every reply whose\ndescribed change it alters is re-drafted under Phase 1's voice rule, confirmed, and updated in\nplace; a verdict changes only as an explicit, user-confirmed exception to Phase 2's \"Scope is\nsettled in Phase 1\"; a lesson whose source rows are no longer all accepted is put to the user; no\nrow affected → the ANSWERS file is not rewritten. Then Phase 2 on that file, then Next step.\n\n## Verdicts\n\n- **address**: the change enters the requirements in full; the reply stays a canned one-liner\n  (\"fixed\", \"good catch, thanks\").\n- **partial**: part of the change is accepted into the requirements, and the reply explains what\n  was done and what was kept and why.\n- **push back**: no code change; the reply carries the reason.\n- **no action**: no code change; a reply is optional (a question answered, an outdated concern the\n  current code settles, an approval with no ask, the author's own comment) — when written, it is\n  drafted and shown like any other.\n\nBot findings take `bot: included` (reply \"fixed\"), `bot: partial` (Notes say which part; the\nreply names what was fixed and why the rest was not) or `bot: ignored` (reply: the reason, one\nshort factual sentence, no thanks or hedging).\n\n## Naming a comment\n\nNever refer to a comment by index or position, anywhere — questions, batch lists, hand-off text.\nQuote its opening words instead, truncated to the shortest unique prefix — the comment \"the null\ncheck here misses …\". Always verbatim, never translated or corrected, so the quote can be searched\non the page. Only the ANSWERS table's own row numbers stay, for in-file cross-references.\n\n## Phase 1 — triage every comment with the user\n\nBefore the first code check, compare the checkout with the PR-REVIEW head: a different commit or\nbranch, or a dirty tree, is reported and the user decides; never switch branches yourself. A PR\nhead that moved since the fetch (the remote compared with the file's `Head`, or with the\n`head=\u003Csha>` token when given) → offer a re-fetch before triage; a refresh (no new activity) keeps\nthe file, the checkout still compared with the current head, and a refresh the URL path's fetch\njust reported needs no re-fetch. The URL path gets the same check — the subagent's fetch is seconds\nold, the checkout is still the user's.\n\nOn a `-N` file, read the earlier rounds' ANSWERS first: a comment re-raising one already triaged\nkeeps its verdict and reply unless it brings evidence the earlier one lacked, and its Notes say\n`kept from round K, row M`; a kept row's work and lessons stay with round K, never counted as\naccepted below. An earlier round without ANSWERS was never triaged: its comments join this triage,\neach row's Notes saying `joined from round K`, its status read from the latest `## Since round`\nblock listing it. A `-N` file with no touched thread (a Since block only) → say there is nothing\nto triage, list the status changes, write no ANSWERS or REQUIREMENTS file, print no launch command,\nstop. This round's ANSWERS file or REQUIREMENTS file already exists (and the REQUIREMENTS file is\nnot the input) → one question before triage, covering both: **update in place** (recommended —\nexisting rows stand, their verdicts and replies kept unless the user changes them or a comment\nbrings new evidence; only new or changed comments are triaged) or **start over** (every comment\nre-triaged, both files replaced at write time). Never a second ANSWERS sheet; nothing overwritten\nwithout this answer.\n\nRead the code behind each comment first: verify its premise against the checkout, never judging\nfrom the quoted hunk alone. Then form a recommendation — verdict, reasoning, and a draft reply for\nevery verdict but address: partial, push back, and a `no action` reply that answers or explains, a\ncourtesy reply on an outdated thread included. Before the first such draft in a round (even one\nwith no partial or push-back row), actually invoke `use-conversational-language` (Developer\nconversations, author replies) and follow it; reciting its rules from memory does not count.\nNot installed: plain prose, no dashes and no AI tells. Every reply other than address's canned\none-liner is shown to the user before it is written. Nothing is decided silently, and every\nquestion to the user is worded via `explain-in-simple-language` when available, the draft reply it\ncarries excepted:\n\n- **Contentious ones one at a time**: every comment recommended partial or push back, or\n  uncertain, is its own question carrying recommendation, reasoning, and draft reply.\n- **Obvious ones as one batch**: then list the comments recommended address or no action, each\n  `no action` item with its draft reply, for a single confirmation; the user can pull any item out\n  for individual discussion.\n\nPer comment type:\n\n- Open\u002Factive human threads → full triage.\n- **Outdated** → code check first; when the concern is moot in the current code, recommend \"no\n  action, optional courtesy reply\".\n- **Unclear** (a status the fetch could not settle) → ask the user its status first, then triage\n  it as that status.\n- **Resolved** → no questions; table row marked \"resolved earlier\"; closed as won't-fix → Notes\n  say so.\n- Review verdict summaries and general comments → each actionable ask inside them is triaged like\n  a thread. An `(author)` comment → no action unless it carries an ask worth tracking.\n- Bot comments → present the actionable findings as one batch include-or-ignore question, each\n  ignore or partial recommendation carrying its draft reply; a bot comment with no finding (a\n  passing quality gate, a change summary) gets no reply. A bot thread flagged `human reply` is\n  triaged as a human thread, the reply addressed to that person.\n\nPhase 1 is complete only when every comment in the PR-REVIEW file has a user-confirmed verdict.\nWrite the ANSWERS file now, before Phase 2.\n\n## Phase 2 — requirements via \u002Frefine-ticket\n\nTreat the accepted work (address, the accepted part of each partial and bot: partial, and included\nbot findings) as a settled-scope ticket and follow **\u002Frefine-ticket** (load and follow its\nSKILL.md): code verification, blocking vs non-blocking, the REQUIREMENTS structure, location and\nnaming — all as written, with these overrides:\n\n- Scope is settled in Phase 1 — never re-ask whether to address a comment; grilling covers only\n  residual requirement-level gaps.\n- The PR-REVIEW file is the local input, so the output lands next to it per its naming rule\n  (`1234-task.PR-REVIEW.md` → `1234-task.PR-REVIEW.REQUIREMENTS.md`); later review rounds carry\n  their suffix through (`…PR-REVIEW-2.REQUIREMENTS.md`).\n- The existing-file question was asked before triage: never ask again at write time; update in\n  place or replace as answered.\n- A requirement's source tag is the comment's opening words and its `path:line` (threadless: its\n  Location label), e.g. `(\"Validate both new option arguments\", install.sh:88)` — never a row\n  number.\n- Nothing about replies or verdicts enters the REQUIREMENTS file.\n- Ignore its hand-off commands, the blocking-question one included — this skill owns the hand-off.\n\nWhen nothing is accepted (no address, partial, bot: included or bot: partial), skip this phase: no\nREQUIREMENTS file, nothing to plan or execute.\n\n## The ANSWERS file\n\n`\u003Cinput name>.ANSWERS.md` next to the input (`1234-task.PR-REVIEW.md` →\n`1234-task.PR-REVIEW.ANSWERS.md`) — the user's copy-paste sheet and the durable record of what was\npushed back. Header: `# Answers: \u003CPR title>`, then `> **Source** [\u003CPR id>](\u003Curl>)`,\n`> **Head** \u003Csha>` (the head the verdicts were settled against), `> **Triaged** YYYY-MM-DD`; round\nand requirements file follow from the file name. Then one table row per comment, resolved and bot\nrows included for completeness:\n\n| # | Original comment | Location | Verdict | Reply | Notes |\n\n- **Original comment** — its opening words verbatim, truncated to the shortest unique prefix.\n- **Location** — the thread's permalink from the PR-REVIEW file, with `path:line` as the readable\n  label; a reply goes inside its thread, never as a top-level comment. A review summary or general\n  comment has no thread: labelled `review summary` or `general comment`, linked to its permalink\n  when the fetch has one, its reply a top-level comment quoting the ask.\n- **Verdict** — `address` \u002F `partial` \u002F `push back` \u002F `no action`, or `resolved earlier` \u002F\n  `bot: included` \u002F `bot: partial` \u002F `bot: ignored`.\n- **Reply** — canned one-liner for address and bot: included. Post-fix stance: written as if the\n  accepted changes are already made (\"did X, kept Y as is because …\") — the sheet is pasted after\n  the fixes land. A reply too long for a cell goes below the table under a `## Reply \u003Crow #>`\n  heading that opens with the original comment quoted in full as a blockquote, so each reply is\n  checkable against its question without scrolling back to the table.\n- **Notes** — the reasoning in a few words, for future review rounds.\n\n## Lessons\n\nDuring Phase 1, note accepted comments (address, the accepted part of partial, bot: included, bot:\npartial) whose lesson generalizes — would a fresh agent on another task here plausibly repeat the\nmistake? A one-off slip fails; a convention or recurring pattern passes. After Phase 2, present\nthem as one batch for the user to strike or add to, each phrased as the general rule, not the\nincident (\"Mock external HTTP in unit tests\", not \"the null check on line 42\"). Append the\nsurvivors as a `## Lessons` section at the end of the ANSWERS file, one line each with its source\nrow numbers; none → no section.\n\n## Boundaries\n\n- **Never touch the PR**: no posting, replying, resolving, or voting — file in, files out.\n- No implementation, no plan in this session: it writes only the ANSWERS and REQUIREMENTS files;\n  the subagents of Next step write their own.\n\n## Next step\n\nState the output paths (project-relative). Remind the user the ANSWERS file is theirs to post\nmanually once the fixes land. No REQUIREMENTS file → no launch command, only that reminder (and\nthe self-improve subagent or line below when Lessons were written). While the REQUIREMENTS file\nholds a **blocking** Open questions entry, print no execute or plan command and no subagent offer,\nonly who must answer it and the command for once the answer is in:\n`\u002Frefine-pr-review \u003Crequirements-file>`. Otherwise hand off by the size of the accepted change set,\nleading with your recommendation but giving each option as its own fenced block under the label\nquoted below, holding a single copy-pasteable launch command (vendor-agnostic — `claude` below is\nonly the example). The recommendation names only options printed below: without\ncreate-implementation-plan, a substantial change set is recommended to harden first (when\navailable), else to execute directly.\n\n### Subagents first\n\nWhen the harness has subagents and at least one subagent below applies, lead with one sentence\noffering to spawn them now, in this session, with `autoaccept`, naming what each would do, that\nit applies its own recommended choices without showing diffs, asking only on a choice that would\nreshape the change, and that the self-improve one waits for the execute one (\"Recommended: I spawn\ntwo subagents now, one applying \u003Crequirements-file> and running validation, then, once it is done,\none persisting the two lessons into the project's docs; both apply their own picks unasked and stop\nonly on a choice that reshapes the change. Yes, or pick one.\"); nothing is spawned before the yes,\nand the fenced blocks then follow under \"or in a fresh session:\". On the yes, spawn each as one\nsubagent briefed with nothing else from the session:\n\n- **Execute** — applies when the change set is the small mechanical kind and the REQUIREMENTS file\n  has no Open questions entry, blocking or not. Brief: the REQUIREMENTS path; the branch and head\n  checked out now, which it must find unchanged, else `blocked`; read-only git; apply the file,\n  then run the project's validation (lint, tests, build); inside the accepted scope take your own\n  recommended choice, and stop before any choice that would reshape the change (a design decision,\n  scope beyond the file) or a validation failure outside that scope. Report only `done` with the\n  files changed and the validation result, or `blocked` with what was already applied, the\n  decision, its candidate answers and your pick with why; no hand-off text.\n- **Self-improve** — applies whenever Lessons were written and `\u002Fself-improve` is available, per\n  its [delegate.md](..\u002Fself-improve\u002Fdelegate.md) on the ANSWERS file's `## Lessons` section, every\n  lesson in one brief, whatever the change set's size. Spawned after the execute subagent has\n  finished — its `done`, or the user's answer to its `blocked` stopping it — when one was spawned,\n  else at once; the yes to the offer covers that later spawn. A REQUIREMENTS file whose changes no\n  execute subagent finished applying (none spawned, or one stopped) means they go to a separate\n  session:\n  say, when announcing this spawn, that the execute session is to start after its report, since\n  both may edit the same file.\n\n`blocked` → put the decision to the user as one question, recommending the subagent's pick, and\nrelaunch with the answer added to the brief — the execute subagent on the tree as it left it, told\nwhat it already applied, or, the user declining the relaunch, stopped there with nothing more\napplied and the execute block re-shown with the report's applied list and the open decision added\nto its prompt; the self-improve one per its delegate.md. Show each report as returned, then the\nANSWERS reminder; fidelity is the subagent's job — never re-run validation or re-read its edits to\ncheck.\n\n\"Apply the changes directly, for a small mechanical change set:\"\n\n```\nclaude --name execute-review-\u003Cslug> \"Apply \u003Crequirements-file>, then run the project's validation (lint, tests, build)\"\n```\n\n`\u002Fcreate-implementation-plan` available — \"or plan first, for a substantial change set (cross-file\nrework, design decisions):\"\n\n```\nclaude --name create-plan-\u003Cslug> \"\u002Fcreate-implementation-plan \u003Crequirements-file>\"\n```\n\n`\u002Fharden-artifact` available — \"or harden the requirements first, recommended for a substantial\nchange set:\"\n\n```\nclaude --name harden-requirements-\u003Cslug> \"\u002Fharden-artifact \u003Crequirements-file>\"\n```\n\nLessons written and `\u002Fself-improve` available — \"Persist the lessons:\", or, under an execute\ncommand above, \"Persist the lessons after the execute session:\", since a mechanical lesson may edit\na lint config or CI job the execution also touches:\n\n```\nclaude --name self-improve-\u003Cslug> \"\u002Fself-improve Persist the lessons listed under Lessons in \u003Canswers-file>\"\n```\n\nThen offer the alternative: one line offering to clear the current session (vendor-agnostic —\n`\u002Fclear` is only the example) and run the next phase here, then the same labelled blocks with the\nbare prompts, under the same gates.\n","# Refine PR review\n\nThe **Refine** phase for review feedback: turn a `.PR-REVIEW.md` (from \u002Ffetch-pr-review; ask if\nthe path is ambiguous) into user-settled per-comment verdicts, drafted replies, and a REQUIREMENTS\nfile covering the accepted code changes. Triage and replies are settled here; the plan and execute\nphases see only code changes.\n\nA PR URL instead: when the harness has subagents and fetch-pr-review is available\n(`..\u002Ffetch-pr-review\u002FSKILL.md` beside this skill), a subagent fetches it first per its\n[delegate.md](..\u002Ffetch-pr-review\u002Fdelegate.md) and the file it returns is the input. Not available\n→ say so and ask for the fetched file; available but no subagents → hand off\n`\u002Ffetch-pr-review \u003Curl>` to a fresh session and stop.\n\nA trailing `head=\u003Csha>` token (fetch-pr-review's refresh hand-off) is the head a refresh just\nconfirmed: Phase 1's moved-head check uses it in place of the file's `Head` line.\n\nIts own `.PR-REVIEW.REQUIREMENTS.md` instead, once a blocking answer is in: run Phase 1's checkout\nand remote-head check first — a checkout differing from the ANSWERS `Head`, or a dirty tree, is\nreported and the user decides, never a branch switch; a moved PR head is reported with the\nre-fetch offer and the user decides whether to finish this round first. Verdicts, replies and\nLessons stand as its `.PR-REVIEW.ANSWERS.md` records them. Apply the answer: every reply whose\ndescribed change it alters is re-drafted under Phase 1's voice rule, confirmed, and updated in\nplace; a verdict changes only as an explicit, user-confirmed exception to Phase 2's \"Scope is\nsettled in Phase 1\"; a lesson whose source rows are no longer all accepted is put to the user; no\nrow affected → the ANSWERS file is not rewritten. Then Phase 2 on that file, then Next step.\n\n## Verdicts\n\n- **address**: the change enters the requirements in full; the reply stays a canned one-liner\n  (\"fixed\", \"good catch, thanks\").\n- **partial**: part of the change is accepted into the requirements, and the reply explains what\n  was done and what was kept and why.\n- **push back**: no code change; the reply carries the reason.\n- **no action**: no code change; a reply is optional (a question answered, an outdated concern the\n  current code settles, an approval with no ask, the author's own comment) — when written, it is\n  drafted and shown like any other.\n\nBot findings take `bot: included` (reply \"fixed\"), `bot: partial` (Notes say which part; the\nreply names what was fixed and why the rest was not) or `bot: ignored` (reply: the reason, one\nshort factual sentence, no thanks or hedging).\n\n## Naming a comment\n\nNever refer to a comment by index or position, anywhere — questions, batch lists, hand-off text.\nQuote its opening words instead, truncated to the shortest unique prefix — the comment \"the null\ncheck here misses …\". Always verbatim, never translated or corrected, so the quote can be searched\non the page. Only the ANSWERS table's own row numbers stay, for in-file cross-references.\n\n## Phase 1 — triage every comment with the user\n\nBefore the first code check, compare the checkout with the PR-REVIEW head: a different commit or\nbranch, or a dirty tree, is reported and the user decides; never switch branches yourself. A PR\nhead that moved since the fetch (the remote compared with the file's `Head`, or with the\n`head=\u003Csha>` token when given) → offer a re-fetch before triage; a refresh (no new activity) keeps\nthe file, the checkout still compared with the current head, and a refresh the URL path's fetch\njust reported needs no re-fetch. The URL path gets the same check — the subagent's fetch is seconds\nold, the checkout is still the user's.\n\nOn a `-N` file, read the earlier rounds' ANSWERS first: a comment re-raising one already triaged\nkeeps its verdict and reply unless it brings evidence the earlier one lacked, and its Notes say\n`kept from round K, row M`; a kept row's work and lessons stay with round K, never counted as\naccepted below. An earlier round without ANSWERS was never triaged: its comments join this triage,\neach row's Notes saying `joined from round K`, its status read from the latest `## Since round`\nblock listing it. A `-N` file with no touched thread (a Since block only) → say there is nothing\nto triage, list the status changes, write no ANSWERS or REQUIREMENTS file, print no launch command,\nstop. This round's ANSWERS file or REQUIREMENTS file already exists (and the REQUIREMENTS file is\nnot the input) → one question before triage, covering both: **update in place** (recommended —\nexisting rows stand, their verdicts and replies kept unless the user changes them or a comment\nbrings new evidence; only new or changed comments are triaged) or **start over** (every comment\nre-triaged, both files replaced at write time). Never a second ANSWERS sheet; nothing overwritten\nwithout this answer.\n\nRead the code behind each comment first: verify its premise against the checkout, never judging\nfrom the quoted hunk alone. Then form a recommendation — verdict, reasoning, and a draft reply for\nevery verdict but address: partial, push back, and a `no action` reply that answers or explains, a\ncourtesy reply on an outdated thread included. Before the first such draft in a round (even one\nwith no partial or push-back row), actually invoke `use-conversational-language` (Developer\nconversations, author replies) and follow it; reciting its rules from memory does not count.\nNot installed: plain prose, no dashes and no AI tells. Every reply other than address's canned\none-liner is shown to the user before it is written. Nothing is decided silently, and every\nquestion to the user is worded via `explain-in-simple-language` when available, the draft reply it\ncarries excepted:\n\n- **Contentious ones one at a time**: every comment recommended partial or push back, or\n  uncertain, is its own question carrying recommendation, reasoning, and draft reply.\n- **Obvious ones as one batch**: then list the comments recommended address or no action, each\n  `no action` item with its draft reply, for a single confirmation; the user can pull any item out\n  for individual discussion.\n\nPer comment type:\n\n- Open\u002Factive human threads → full triage.\n- **Outdated** → code check first; when the concern is moot in the current code, recommend \"no\n  action, optional courtesy reply\".\n- **Unclear** (a status the fetch could not settle) → ask the user its status first, then triage\n  it as that status.\n- **Resolved** → no questions; table row marked \"resolved earlier\"; closed as won't-fix → Notes\n  say so.\n- Review verdict summaries and general comments → each actionable ask inside them is triaged like\n  a thread. An `(author)` comment → no action unless it carries an ask worth tracking.\n- Bot comments → present the actionable findings as one batch include-or-ignore question, each\n  ignore or partial recommendation carrying its draft reply; a bot comment with no finding (a\n  passing quality gate, a change summary) gets no reply. A bot thread flagged `human reply` is\n  triaged as a human thread, the reply addressed to that person.\n\nPhase 1 is complete only when every comment in the PR-REVIEW file has a user-confirmed verdict.\nWrite the ANSWERS file now, before Phase 2.\n\n## Phase 2 — requirements via \u002Frefine-ticket\n\nTreat the accepted work (address, the accepted part of each partial and bot: partial, and included\nbot findings) as a settled-scope ticket and follow **\u002Frefine-ticket** (load and follow its\nSKILL.md): code verification, blocking vs non-blocking, the REQUIREMENTS structure, location and\nnaming — all as written, with these overrides:\n\n- Scope is settled in Phase 1 — never re-ask whether to address a comment; grilling covers only\n  residual requirement-level gaps.\n- The PR-REVIEW file is the local input, so the output lands next to it per its naming rule\n  (`1234-task.PR-REVIEW.md` → `1234-task.PR-REVIEW.REQUIREMENTS.md`); later review rounds carry\n  their suffix through (`…PR-REVIEW-2.REQUIREMENTS.md`).\n- The existing-file question was asked before triage: never ask again at write time; update in\n  place or replace as answered.\n- A requirement's source tag is the comment's opening words and its `path:line` (threadless: its\n  Location label), e.g. `(\"Validate both new option arguments\", install.sh:88)` — never a row\n  number.\n- Nothing about replies or verdicts enters the REQUIREMENTS file.\n- Ignore its hand-off commands, the blocking-question one included — this skill owns the hand-off.\n\nWhen nothing is accepted (no address, partial, bot: included or bot: partial), skip this phase: no\nREQUIREMENTS file, nothing to plan or execute.\n\n## The ANSWERS file\n\n`\u003Cinput name>.ANSWERS.md` next to the input (`1234-task.PR-REVIEW.md` →\n`1234-task.PR-REVIEW.ANSWERS.md`) — the user's copy-paste sheet and the durable record of what was\npushed back. Header: `# Answers: \u003CPR title>`, then `> **Source** [\u003CPR id>](\u003Curl>)`,\n`> **Head** \u003Csha>` (the head the verdicts were settled against), `> **Triaged** YYYY-MM-DD`; round\nand requirements file follow from the file name. Then one table row per comment, resolved and bot\nrows included for completeness:\n\n| # | Original comment | Location | Verdict | Reply | Notes |\n\n- **Original comment** — its opening words verbatim, truncated to the shortest unique prefix.\n- **Location** — the thread's permalink from the PR-REVIEW file, with `path:line` as the readable\n  label; a reply goes inside its thread, never as a top-level comment. A review summary or general\n  comment has no thread: labelled `review summary` or `general comment`, linked to its permalink\n  when the fetch has one, its reply a top-level comment quoting the ask.\n- **Verdict** — `address` \u002F `partial` \u002F `push back` \u002F `no action`, or `resolved earlier` \u002F\n  `bot: included` \u002F `bot: partial` \u002F `bot: ignored`.\n- **Reply** — canned one-liner for address and bot: included. Post-fix stance: written as if the\n  accepted changes are already made (\"did X, kept Y as is because …\") — the sheet is pasted after\n  the fixes land. A reply too long for a cell goes below the table under a `## Reply \u003Crow #>`\n  heading that opens with the original comment quoted in full as a blockquote, so each reply is\n  checkable against its question without scrolling back to the table.\n- **Notes** — the reasoning in a few words, for future review rounds.\n\n## Lessons\n\nDuring Phase 1, note accepted comments (address, the accepted part of partial, bot: included, bot:\npartial) whose lesson generalizes — would a fresh agent on another task here plausibly repeat the\nmistake? A one-off slip fails; a convention or recurring pattern passes. After Phase 2, present\nthem as one batch for the user to strike or add to, each phrased as the general rule, not the\nincident (\"Mock external HTTP in unit tests\", not \"the null check on line 42\"). Append the\nsurvivors as a `## Lessons` section at the end of the ANSWERS file, one line each with its source\nrow numbers; none → no section.\n\n## Boundaries\n\n- **Never touch the PR**: no posting, replying, resolving, or voting — file in, files out.\n- No implementation, no plan in this session: it writes only the ANSWERS and REQUIREMENTS files;\n  the subagents of Next step write their own.\n\n## Next step\n\nState the output paths (project-relative). Remind the user the ANSWERS file is theirs to post\nmanually once the fixes land. No REQUIREMENTS file → no launch command, only that reminder (and\nthe self-improve subagent or line below when Lessons were written). While the REQUIREMENTS file\nholds a **blocking** Open questions entry, print no execute or plan command and no subagent offer,\nonly who must answer it and the command for once the answer is in:\n`\u002Frefine-pr-review \u003Crequirements-file>`. Otherwise hand off by the size of the accepted change set,\nleading with your recommendation but giving each option as its own fenced block under the label\nquoted below, holding a single copy-pasteable launch command (vendor-agnostic — `claude` below is\nonly the example). The recommendation names only options printed below: without\ncreate-implementation-plan, a substantial change set is recommended to harden first (when\navailable), else to execute directly.\n\n### Subagents first\n\nWhen the harness has subagents and at least one subagent below applies, lead with one sentence\noffering to spawn them now, in this session, with `autoaccept`, naming what each would do, that\nit applies its own recommended choices without showing diffs, asking only on a choice that would\nreshape the change, and that the self-improve one waits for the execute one (\"Recommended: I spawn\ntwo subagents now, one applying \u003Crequirements-file> and running validation, then, once it is done,\none persisting the two lessons into the project's docs; both apply their own picks unasked and stop\nonly on a choice that reshapes the change. Yes, or pick one.\"); nothing is spawned before the yes,\nand the fenced blocks then follow under \"or in a fresh session:\". On the yes, spawn each as one\nsubagent briefed with nothing else from the session:\n\n- **Execute** — applies when the change set is the small mechanical kind and the REQUIREMENTS file\n  has no Open questions entry, blocking or not. Brief: the REQUIREMENTS path; the branch and head\n  checked out now, which it must find unchanged, else `blocked`; read-only git; apply the file,\n  then run the project's validation (lint, tests, build); inside the accepted scope take your own\n  recommended choice, and stop before any choice that would reshape the change (a design decision,\n  scope beyond the file) or a validation failure outside that scope. Report only `done` with the\n  files changed and the validation result, or `blocked` with what was already applied, the\n  decision, its candidate answers and your pick with why; no hand-off text.\n- **Self-improve** — applies whenever Lessons were written and `\u002Fself-improve` is available, per\n  its [delegate.md](..\u002Fself-improve\u002Fdelegate.md) on the ANSWERS file's `## Lessons` section, every\n  lesson in one brief, whatever the change set's size. Spawned after the execute subagent has\n  finished — its `done`, or the user's answer to its `blocked` stopping it — when one was spawned,\n  else at once; the yes to the offer covers that later spawn. A REQUIREMENTS file whose changes no\n  execute subagent finished applying (none spawned, or one stopped) means they go to a separate\n  session:\n  say, when announcing this spawn, that the execute session is to start after its report, since\n  both may edit the same file.\n\n`blocked` → put the decision to the user as one question, recommending the subagent's pick, and\nrelaunch with the answer added to the brief — the execute subagent on the tree as it left it, told\nwhat it already applied, or, the user declining the relaunch, stopped there with nothing more\napplied and the execute block re-shown with the report's applied list and the open decision added\nto its prompt; the self-improve one per its delegate.md. Show each report as returned, then the\nANSWERS reminder; fidelity is the subagent's job — never re-run validation or re-read its edits to\ncheck.\n\n\"Apply the changes directly, for a small mechanical change set:\"\n\n```\nclaude --name execute-review-\u003Cslug> \"Apply \u003Crequirements-file>, then run the project's validation (lint, tests, build)\"\n```\n\n`\u002Fcreate-implementation-plan` available — \"or plan first, for a substantial change set (cross-file\nrework, design decisions):\"\n\n```\nclaude --name create-plan-\u003Cslug> \"\u002Fcreate-implementation-plan \u003Crequirements-file>\"\n```\n\n`\u002Fharden-artifact` available — \"or harden the requirements first, recommended for a substantial\nchange set:\"\n\n```\nclaude --name harden-requirements-\u003Cslug> \"\u002Fharden-artifact \u003Crequirements-file>\"\n```\n\nLessons written and `\u002Fself-improve` available — \"Persist the lessons:\", or, under an execute\ncommand above, \"Persist the lessons after the execute session:\", since a mechanical lesson may edit\na lint config or CI job the execution also touches:\n\n```\nclaude --name self-improve-\u003Cslug> \"\u002Fself-improve Persist the lessons listed under Lessons in \u003Canswers-file>\"\n```\n\nThen offer the alternative: one line offering to clear the current session (vendor-agnostic —\n`\u002Fclear` is only the example) and run the next phase here, then the same labelled blocks with the\nbare prompts, under the same gates.\n",{"name":5,"license":15,"metadata":24,"description":7},{"version":25},"1.12","main","d79feee1f4817fe025b16c6b910364c6f1d5f472","skills\u002Frefine-pr-review",19,"https:\u002F\u002Fgithub.com\u002Feai-org\u002Fagent-toolkit","https:\u002F\u002Feai-org.github.io\u002Fagent-toolkit-docs","https:\u002F\u002Fgithub.com\u002Feai-org","https:\u002F\u002Fmedium.com\u002Fengineering-in-the-age-of-ai","2026-11-14 22:36:21",{"source":8,"skillName":5,"commands":36,"prompt":46,"zipAvailable":47},[37,40,43],{"tool":38,"command":39},"npx","npx skills add eai-org\u002Fagent-toolkit --skill refine-pr-review",{"tool":41,"command":42},"bunx","bunx skills add eai-org\u002Fagent-toolkit --skill refine-pr-review",{"tool":44,"command":45},"pnpm","pnpm dlx skills add eai-org\u002Fagent-toolkit --skill refine-pr-review","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Feai-org\u002Fagent-toolkit\u002Fblob\u002Fmain\u002Fskills\u002Frefine-pr-review\u002FSKILL.md",true,[],[50,56,63,70,77,84],{"skillNo":51,"name":52,"nameZh":6,"description":6,"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},"01M74PKAVPTTY5PSSAWHCRVY9T","no-frontmatter","https:\u002F\u002Fgithub.com\u002Feai-org\u002Fagent-toolkit\u002Fblob\u002Fmain\u002Ftest\u002Fcompact-skill-creator\u002Ffixtures\u002Fno-frontmatter\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},"01M74PK8J99W7V3669Y4JG2JKV","other-name","Fixture skill whose name differs from its folder.","https:\u002F\u002Fgithub.com\u002Feai-org\u002Fagent-toolkit\u002Fblob\u002Fmain\u002Ftest\u002Fcompact-skill-creator\u002Ffixtures\u002Fname-mismatch\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},"01M74PK6T9DHVRCNHTEV2ZA4Y3","md-off","A fixture skill whose folder turns the line length off below a root that sets 99.","https:\u002F\u002Fgithub.com\u002Feai-org\u002Fagent-toolkit\u002Fblob\u002Fmain\u002Ftest\u002Fcompact-skill-creator\u002Ffixtures\u002Fmd-off\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},"01M74PK4H4N3D290V4C08M4DGY","long-line","Fixture skill with one prose line of 100 characters, over the fixtures' limit of 99.","https:\u002F\u002Fgithub.com\u002Feai-org\u002Fagent-toolkit\u002Fblob\u002Fmain\u002Ftest\u002Fcompact-skill-creator\u002Ffixtures\u002Flong-line\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},"01M74PK2DJ2MCR5KFPMYVQWNJG","flag-without-flow","Fixture skill with the flag and no type.","https:\u002F\u002Fgithub.com\u002Feai-org\u002Fagent-toolkit\u002Fblob\u002Fmain\u002Ftest\u002Fcompact-skill-creator\u002Ffixtures\u002Fflag-without-flow\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},"01M74PK03XGEBQ986FDAE46WCM","double-backtick","A fixture skill with double-backtick code spans holding a link template and an overrun.","https:\u002F\u002Fgithub.com\u002Feai-org\u002Fagent-toolkit\u002Fblob\u002Fmain\u002Ftest\u002Fcompact-skill-creator\u002Ffixtures\u002Fdouble-backtick\u002FSKILL.md",[],[]]