[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6NJYTP8ZCCG2BFKT6NCT2ZV":3,"skill-same:01M6NJYTP8ZCCG2BFKT6NCT2ZV":60},{"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":33,"linkOnly":34,"markdownContent":35,"markdownBody":36,"frontMatter":37,"refName":39,"commitSha":40,"skillDirectory":41,"forksCount":42,"repositoryHtmlUrl":43,"repositoryHomepageUrl":6,"ownerHtmlUrl":44,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":45,"install":46,"resources":59},"01M6NJYTP8ZCCG2BFKT6NCT2ZV","internal-docs-prune",null,"Prune internal Claude-facing docs (CLAUDE.md, nested CLAUDE.md files, NOTES.md, .claude\u002Frules\u002F*.md) — spawns internal-docs-reviewer, auto-applies redundant\u002Fstale findings without content loss, and escalates judgment findings (interactive) or defers them (unattended)","plwhite\u002Fclaudeplugins","plwhite","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F8448749?v=4","01M6NJY6F3ZZDJYK81DHP6JVJN","Peter White","01M6NJY6KSDFD9W61Y3Q85WR86",3,"MIT","https:\u002F\u002Fgithub.com\u002Fplwhite\u002Fclaudeplugins\u002Fblob\u002Fmain\u002Fdevproc\u002Fskills\u002Finternal-docs-prune\u002FSKILL.md",[],[19,20,21,22,23,24,25,26,27,28,29,30,31,32],"docs","prune","internal docs prune","notes","content","Rules","reviewer","interactive","auto","stale","files","nested","judgment","unattended","2026-11-10 10:12:30",false,"---\nname: internal-docs-prune\ndescription: Prune internal Claude-facing docs (CLAUDE.md, nested CLAUDE.md files, NOTES.md, .claude\u002Frules\u002F*.md) — spawns internal-docs-reviewer, auto-applies redundant\u002Fstale findings without content loss, and escalates judgment findings (interactive) or defers them (unattended)\nargument-hint: [unattended]\n---\n\nPrune this repository's internal, Claude-facing documentation — the working\nmemory Claude itself maintains across sessions, as distinct from user-facing\n`docs\u002F`. This spawns the read-only `internal-docs-reviewer` agent, then\napplies its findings by gating class.\n\nThe invocation is: $ARGUMENTS\n\n## Step 1 — Determine mode\n\nCheck $ARGUMENTS for an unattended token: `unattended` or `--unattended`\n(case-insensitive, anywhere in the string). If present, run in **unattended\nmode**. Otherwise run in **interactive mode** — the default.\n\nThe mode only changes how `judgment` findings are handled (Step 5). Both\nmodes apply `redundant` and `stale` findings identically (Step 4).\n\n## Step 2 — Identify scope\n\nThe in-scope surface is exactly what `internal-docs-reviewer` reviews: the\nroot `CLAUDE.md`, any nested `CLAUDE.md` files, `NOTES.md`, and\n`.claude\u002Frules\u002F*.md`. Do not pre-filter this list yourself beyond finding the\ncandidate files (e.g. `Glob **\u002FCLAUDE.md`) — the agent decides what is\nactually reviewable within it.\n\nThis skill never edits anything outside this surface, with one narrow\nexception: a `move` finding's named destination, which is always one of\n`features\u002FCOMPLETED.md`, a named file under `features\u002Fplans\u002F`, or a named\nfile under `.claude\u002Frules\u002F` (the agent's own permitted destinations — see the\n`move` action in its findings contract). Report-only or user-facing files\n(`docs\u002F`, `README.md`,`CONTRIBUTING.md`) are never touched, even if a finding\nsomehow named one - treat that as a defect in the agent's output (see Step 3)\nrather than acting on it.\n\n## Step 3 — Spawn the reviewer and collect findings\n\nCall @internal-docs-reviewer with the workspace root (or the specific\nin-scope files from Step 2) to review.\n\nParse its output into a list of findings, each carrying: Gating class\n(`redundant` | `stale` | `judgment`), file path, Anchor (exact verbatim\ntext), Issue, Action (`delete` | `move` → destination | `condense`), and\nReplacement\u002FDestination text where the action requires one.\n\nA finding missing a required field (see the agent's findings contract — every\nfinding needs exactly one gating class and action, plus a\ndestination\u002Freplacement wherever the action requires one) is a defect in the\nagent's output, not something to guess at: skip it, note it in the summary\n(Step 6) as \"could not apply — incomplete finding\", and do not act on it.\n\nA finding whose gating class and action are an **invalid combination** is\nlikewise a defect — do not act on it. The agent's contract constrains the\npairing: a `redundant` finding's action must be `delete` (anything else would\nduplicate content already at its canonical home), and a `judgment` finding's\naction must be `condense` or `move`, never `delete` (the skill never silently\ndeletes content a human should decide about). If a `redundant` finding carries\n`move`\u002F`condense`, or a `judgment` finding carries `delete`, skip it and note\nit in the summary as \"could not apply — invalid class\u002Faction combination\". This\nmeans `redundant`\u002F`stale` reaching Step 4 and `judgment` reaching Step 5 are\nalready known to carry a valid action.\n\nIf the agent reports no findings, skip to Step 6 and report a clean run.\n\n## Step 4 — Apply `redundant` and `stale` findings (both modes)\n\nThese are auto-applied in both interactive and unattended mode — no\nconfirmation needed. Apply them file by file, largest\u002Fmost-impactful first\n(the order the agent returns them in), and **re-verify each anchor is still\npresent, verbatim, in the file immediately before acting on it** — an earlier\nfinding applied to the same file can shift or remove text a later finding\nalso depends on. If an anchor is no longer found, skip that finding and note\nit in the summary as \"anchor not found — file changed since review,\nre-run to confirm.\"\n\nFor each finding, by Action:\n\n- **`delete`** — remove the anchor text entirely. Nothing replaces it.\n- **`condense`** — replace the anchor text with the finding's exact\n  Replacement text.\n- **`move`** — this is the highest-consequence action (content leaving its\n  only home) and the ordering below is not optional:\n  1. Read the destination file (create it only if the agent's Destination\n     names a file that legitimately does not exist yet — this should be rare,\n     since the three permitted destinations are established files).\n  2. Write the Replacement\u002FDestination text into the destination, in a\n     position consistent with that destination's own style (e.g. alongside\n     other entries in `features\u002FCOMPLETED.md`, appended to the relevant\n     section of a plan file).\n  3. **Verify the destination now contains that text** — re-read the file\n     and confirm the exact text is present — before touching the source.\n  4. Only once verified present at the destination, remove the anchor text\n     from the source file.\n\n  If step 3's verification fails for any reason, stop — do not proceed to\n  step 4 for that finding. Report it in the summary as blocked, with the\n  source left untouched. **A `move` must never remove content from the\n  source without first confirming it landed at the destination** — this is\n  the one failure mode (\"silent loss\") this skill exists to prevent, and no\n  other consideration in this skill overrides it.\n\n## Step 5 — Handle `judgment` findings, by mode\n\nBefore handling any `judgment` finding, **load the settled-judgment memory**\n(see `## Project memory` below) and filter the agent's `judgment` findings\nagainst it: if a finding's Anchor matches a settled call recorded there **and\nthe anchor text is unchanged** since that decision, the user already ruled on\nit — drop it silently (do not re-surface or re-defer it). This filtering\napplies in **both** modes. A finding whose anchor text has changed since the\nrecorded decision is treated as new — the earlier ruling no longer applies to\nit.\n\n- **Unattended mode:** take no action on any surviving `judgment` finding.\n  Collect them for the summary (Step 6) as deferred, with their file, Anchor,\n  Issue, and proposed Action, so the calling session can review and act on\n  them later. (Unattended mode records nothing to memory — it settles\n  nothing.)\n\n- **Interactive mode (default):** present each `judgment` finding to the user\n  in turn — file, Anchor, Issue, and the agent's proposed Action\u002FReplacement.\n  Ask what to do with it. **The default, if the user simply confirms or gives\n  no other direction, is the agent's own proposed Action — its `condense`\n  text, or, for a `judgment` + `move` finding, the proposed move — but never\n  `delete`.** Concretely:\n  - Proposed Action `condense` → apply the finding's Replacement text.\n  - Proposed Action `move` → apply the move, following Step 4's `move`\n    procedure in full (see the re-verification note below). (A `judgment`\n    finding may legitimately be a `move`; action and gating class vary\n    independently.)\n\n  The user may instead choose to delete it, pick a different action, edit the\n  replacement or destination, or leave it as-is. Apply whatever they decide\n  before moving to the next finding.\n\n  Whatever action is applied, **re-verify the anchor is still present,\n  verbatim, immediately before acting** (exactly as Step 4 requires) — an\n  earlier `judgment` finding just applied to the same file can shift or remove\n  text a later one depends on. For a `move`, follow Step 4's full four-step\n  sequence — read the destination, write the text, verify it landed, and only\n  then remove the source — including that initial read-the-destination step.\n\n  The agent's contract forbids it from *proposing* `delete` for a `judgment`\n  finding (enforced at Step 3), and this skill will not silently `delete` one\n  either — a deletion here happens only because the user explicitly asks for\n  it.\n\n  **Record each resolved `judgment` finding to memory** (see `## Project\n  memory`) as you settle it — whether the user kept it as-is, condensed it,\n  moved it, or deleted it — so a later run does not re-ask about the same\n  call. Record the anchor, the decision, and the date.\n\n## Step 6 — Summary\n\nReport, grouped by outcome:\n\n- **Deleted** — file and anchor (or a short description) for each.\n- **Moved** — file, anchor, and destination for each, confirming the\n  write-then-remove ordering held.\n- **Condensed** — file and anchor for each.\n- **Retained** — `judgment` findings left untouched: in interactive mode,\n  ones the user chose to keep as-is; in unattended mode, all of them, listed\n  for later review with enough detail (file, Anchor, Issue) that a future\n  session or `\u002Finternal-docs-prune` run can act on them without re-reviewing\n  from scratch.\n- **Not applied** — any finding skipped for a mechanical reason (anchor not\n  found, incomplete finding, failed destination verification), separate from\n  the categories above since these are not decisions, they are problems.\n\n## Step 7 — Commit prefix\n\nThis skill does not commit on its own initiative — committing stays\nuser-gated, per repository convention. If the user asks to commit the\nchanges this run made (or asks you to prepare a commit message), use the\n`docs(prune):` prefix, e.g. `docs(prune): remove duplicated status entries\nand stale webhook note`, so prune commits are discoverable by prefix in\nhistory. If multiple distinct changes happened (e.g. a mechanical\nauto-applied pass and a separate user-directed judgment resolution), it's\nfine to suggest they be split into more than one `docs(prune):` commit —\npropose the split rather than bundling silently.\n\n## Project memory\n\nThis skill — not the agent — owns the record of settled `judgment` calls, so\nrepeat runs don't re-litigate borderline entries the user has already ruled\non. The agent is stateless and re-reports every borderline call each run; this\nskill is the component that holds the user dialogue and therefore the only one\nthat knows how each call was decided.\n\nKeep a `MEMORY.md` index at\n`$CLAUDE_PROJECT_DIR\u002F.claude\u002Fagent-memory\u002Fdevproc-internal-docs-reviewer\u002F`\n(namespaced under the reviewer agent's name with the `devproc-` plugin prefix,\nso this skill's memory stays distinct from other devproc agents' and skills').\nAlways resolve the\npath against `$CLAUDE_PROJECT_DIR` (the workspace root), never the current\nworking directory, so memory lands in the same place regardless of where the\nskill is invoked from. Create the directory and file if they do not yet exist;\nthe first run that settles a `judgment` call creates them.\n\nPer settled `judgment` call, record:\n- The anchor — file path plus the verbatim anchor text at the time of the\n  decision (so a later run can tell whether the text has since changed).\n- The decision made (kept as-is, condensed to *X*, moved to *Y*, or deleted).\n- The date.\n\nRead this index at the start of Step 5 and use it to suppress already-decided\n`judgment` findings (Step 5's filter). Write to it only when a `judgment` call\nis actually settled — which happens only in interactive mode, since unattended\nmode defers rather than decides. Never record `redundant` or `stale` findings\nhere: those are re-verified against the live repository on every run\nregardless of past decisions (a duplication or contradiction can newly arise\nor newly resolve between runs), so memory must never suppress them.\n\n## Idempotency\n\nA second run over files this skill has just pruned should find no new\n`redundant` or `stale` findings — those are exactly what Step 4 clears. A\nresidual `judgment` finding is expected and fine: those are never\nauto-resolved, so one a user chose to keep as-is (interactive) or one that\nwas deferred (unattended) will still be *raised by the agent* — but if it was\nsettled in interactive mode, the Step 5 memory filter suppresses it, so it\ndoes not re-bother the user; a deferred (unattended) one, never settled, will\nre-surface, which is exactly what \"escalated, not silently resolved\" means. If\na second run does surface a `redundant` or `stale` finding, that is either\nnewly introduced since the first run (something else changed the file) or a\nsign the previous apply pass didn't fully take — investigate rather than\nassuming it's expected.\n","Prune this repository's internal, Claude-facing documentation — the working\nmemory Claude itself maintains across sessions, as distinct from user-facing\n`docs\u002F`. This spawns the read-only `internal-docs-reviewer` agent, then\napplies its findings by gating class.\n\nThe invocation is: $ARGUMENTS\n\n## Step 1 — Determine mode\n\nCheck $ARGUMENTS for an unattended token: `unattended` or `--unattended`\n(case-insensitive, anywhere in the string). If present, run in **unattended\nmode**. Otherwise run in **interactive mode** — the default.\n\nThe mode only changes how `judgment` findings are handled (Step 5). Both\nmodes apply `redundant` and `stale` findings identically (Step 4).\n\n## Step 2 — Identify scope\n\nThe in-scope surface is exactly what `internal-docs-reviewer` reviews: the\nroot `CLAUDE.md`, any nested `CLAUDE.md` files, `NOTES.md`, and\n`.claude\u002Frules\u002F*.md`. Do not pre-filter this list yourself beyond finding the\ncandidate files (e.g. `Glob **\u002FCLAUDE.md`) — the agent decides what is\nactually reviewable within it.\n\nThis skill never edits anything outside this surface, with one narrow\nexception: a `move` finding's named destination, which is always one of\n`features\u002FCOMPLETED.md`, a named file under `features\u002Fplans\u002F`, or a named\nfile under `.claude\u002Frules\u002F` (the agent's own permitted destinations — see the\n`move` action in its findings contract). Report-only or user-facing files\n(`docs\u002F`, `README.md`,`CONTRIBUTING.md`) are never touched, even if a finding\nsomehow named one - treat that as a defect in the agent's output (see Step 3)\nrather than acting on it.\n\n## Step 3 — Spawn the reviewer and collect findings\n\nCall @internal-docs-reviewer with the workspace root (or the specific\nin-scope files from Step 2) to review.\n\nParse its output into a list of findings, each carrying: Gating class\n(`redundant` | `stale` | `judgment`), file path, Anchor (exact verbatim\ntext), Issue, Action (`delete` | `move` → destination | `condense`), and\nReplacement\u002FDestination text where the action requires one.\n\nA finding missing a required field (see the agent's findings contract — every\nfinding needs exactly one gating class and action, plus a\ndestination\u002Freplacement wherever the action requires one) is a defect in the\nagent's output, not something to guess at: skip it, note it in the summary\n(Step 6) as \"could not apply — incomplete finding\", and do not act on it.\n\nA finding whose gating class and action are an **invalid combination** is\nlikewise a defect — do not act on it. The agent's contract constrains the\npairing: a `redundant` finding's action must be `delete` (anything else would\nduplicate content already at its canonical home), and a `judgment` finding's\naction must be `condense` or `move`, never `delete` (the skill never silently\ndeletes content a human should decide about). If a `redundant` finding carries\n`move`\u002F`condense`, or a `judgment` finding carries `delete`, skip it and note\nit in the summary as \"could not apply — invalid class\u002Faction combination\". This\nmeans `redundant`\u002F`stale` reaching Step 4 and `judgment` reaching Step 5 are\nalready known to carry a valid action.\n\nIf the agent reports no findings, skip to Step 6 and report a clean run.\n\n## Step 4 — Apply `redundant` and `stale` findings (both modes)\n\nThese are auto-applied in both interactive and unattended mode — no\nconfirmation needed. Apply them file by file, largest\u002Fmost-impactful first\n(the order the agent returns them in), and **re-verify each anchor is still\npresent, verbatim, in the file immediately before acting on it** — an earlier\nfinding applied to the same file can shift or remove text a later finding\nalso depends on. If an anchor is no longer found, skip that finding and note\nit in the summary as \"anchor not found — file changed since review,\nre-run to confirm.\"\n\nFor each finding, by Action:\n\n- **`delete`** — remove the anchor text entirely. Nothing replaces it.\n- **`condense`** — replace the anchor text with the finding's exact\n  Replacement text.\n- **`move`** — this is the highest-consequence action (content leaving its\n  only home) and the ordering below is not optional:\n  1. Read the destination file (create it only if the agent's Destination\n     names a file that legitimately does not exist yet — this should be rare,\n     since the three permitted destinations are established files).\n  2. Write the Replacement\u002FDestination text into the destination, in a\n     position consistent with that destination's own style (e.g. alongside\n     other entries in `features\u002FCOMPLETED.md`, appended to the relevant\n     section of a plan file).\n  3. **Verify the destination now contains that text** — re-read the file\n     and confirm the exact text is present — before touching the source.\n  4. Only once verified present at the destination, remove the anchor text\n     from the source file.\n\n  If step 3's verification fails for any reason, stop — do not proceed to\n  step 4 for that finding. Report it in the summary as blocked, with the\n  source left untouched. **A `move` must never remove content from the\n  source without first confirming it landed at the destination** — this is\n  the one failure mode (\"silent loss\") this skill exists to prevent, and no\n  other consideration in this skill overrides it.\n\n## Step 5 — Handle `judgment` findings, by mode\n\nBefore handling any `judgment` finding, **load the settled-judgment memory**\n(see `## Project memory` below) and filter the agent's `judgment` findings\nagainst it: if a finding's Anchor matches a settled call recorded there **and\nthe anchor text is unchanged** since that decision, the user already ruled on\nit — drop it silently (do not re-surface or re-defer it). This filtering\napplies in **both** modes. A finding whose anchor text has changed since the\nrecorded decision is treated as new — the earlier ruling no longer applies to\nit.\n\n- **Unattended mode:** take no action on any surviving `judgment` finding.\n  Collect them for the summary (Step 6) as deferred, with their file, Anchor,\n  Issue, and proposed Action, so the calling session can review and act on\n  them later. (Unattended mode records nothing to memory — it settles\n  nothing.)\n\n- **Interactive mode (default):** present each `judgment` finding to the user\n  in turn — file, Anchor, Issue, and the agent's proposed Action\u002FReplacement.\n  Ask what to do with it. **The default, if the user simply confirms or gives\n  no other direction, is the agent's own proposed Action — its `condense`\n  text, or, for a `judgment` + `move` finding, the proposed move — but never\n  `delete`.** Concretely:\n  - Proposed Action `condense` → apply the finding's Replacement text.\n  - Proposed Action `move` → apply the move, following Step 4's `move`\n    procedure in full (see the re-verification note below). (A `judgment`\n    finding may legitimately be a `move`; action and gating class vary\n    independently.)\n\n  The user may instead choose to delete it, pick a different action, edit the\n  replacement or destination, or leave it as-is. Apply whatever they decide\n  before moving to the next finding.\n\n  Whatever action is applied, **re-verify the anchor is still present,\n  verbatim, immediately before acting** (exactly as Step 4 requires) — an\n  earlier `judgment` finding just applied to the same file can shift or remove\n  text a later one depends on. For a `move`, follow Step 4's full four-step\n  sequence — read the destination, write the text, verify it landed, and only\n  then remove the source — including that initial read-the-destination step.\n\n  The agent's contract forbids it from *proposing* `delete` for a `judgment`\n  finding (enforced at Step 3), and this skill will not silently `delete` one\n  either — a deletion here happens only because the user explicitly asks for\n  it.\n\n  **Record each resolved `judgment` finding to memory** (see `## Project\n  memory`) as you settle it — whether the user kept it as-is, condensed it,\n  moved it, or deleted it — so a later run does not re-ask about the same\n  call. Record the anchor, the decision, and the date.\n\n## Step 6 — Summary\n\nReport, grouped by outcome:\n\n- **Deleted** — file and anchor (or a short description) for each.\n- **Moved** — file, anchor, and destination for each, confirming the\n  write-then-remove ordering held.\n- **Condensed** — file and anchor for each.\n- **Retained** — `judgment` findings left untouched: in interactive mode,\n  ones the user chose to keep as-is; in unattended mode, all of them, listed\n  for later review with enough detail (file, Anchor, Issue) that a future\n  session or `\u002Finternal-docs-prune` run can act on them without re-reviewing\n  from scratch.\n- **Not applied** — any finding skipped for a mechanical reason (anchor not\n  found, incomplete finding, failed destination verification), separate from\n  the categories above since these are not decisions, they are problems.\n\n## Step 7 — Commit prefix\n\nThis skill does not commit on its own initiative — committing stays\nuser-gated, per repository convention. If the user asks to commit the\nchanges this run made (or asks you to prepare a commit message), use the\n`docs(prune):` prefix, e.g. `docs(prune): remove duplicated status entries\nand stale webhook note`, so prune commits are discoverable by prefix in\nhistory. If multiple distinct changes happened (e.g. a mechanical\nauto-applied pass and a separate user-directed judgment resolution), it's\nfine to suggest they be split into more than one `docs(prune):` commit —\npropose the split rather than bundling silently.\n\n## Project memory\n\nThis skill — not the agent — owns the record of settled `judgment` calls, so\nrepeat runs don't re-litigate borderline entries the user has already ruled\non. The agent is stateless and re-reports every borderline call each run; this\nskill is the component that holds the user dialogue and therefore the only one\nthat knows how each call was decided.\n\nKeep a `MEMORY.md` index at\n`$CLAUDE_PROJECT_DIR\u002F.claude\u002Fagent-memory\u002Fdevproc-internal-docs-reviewer\u002F`\n(namespaced under the reviewer agent's name with the `devproc-` plugin prefix,\nso this skill's memory stays distinct from other devproc agents' and skills').\nAlways resolve the\npath against `$CLAUDE_PROJECT_DIR` (the workspace root), never the current\nworking directory, so memory lands in the same place regardless of where the\nskill is invoked from. Create the directory and file if they do not yet exist;\nthe first run that settles a `judgment` call creates them.\n\nPer settled `judgment` call, record:\n- The anchor — file path plus the verbatim anchor text at the time of the\n  decision (so a later run can tell whether the text has since changed).\n- The decision made (kept as-is, condensed to *X*, moved to *Y*, or deleted).\n- The date.\n\nRead this index at the start of Step 5 and use it to suppress already-decided\n`judgment` findings (Step 5's filter). Write to it only when a `judgment` call\nis actually settled — which happens only in interactive mode, since unattended\nmode defers rather than decides. Never record `redundant` or `stale` findings\nhere: those are re-verified against the live repository on every run\nregardless of past decisions (a duplication or contradiction can newly arise\nor newly resolve between runs), so memory must never suppress them.\n\n## Idempotency\n\nA second run over files this skill has just pruned should find no new\n`redundant` or `stale` findings — those are exactly what Step 4 clears. A\nresidual `judgment` finding is expected and fine: those are never\nauto-resolved, so one a user chose to keep as-is (interactive) or one that\nwas deferred (unattended) will still be *raised by the agent* — but if it was\nsettled in interactive mode, the Step 5 memory filter suppresses it, so it\ndoes not re-bother the user; a deferred (unattended) one, never settled, will\nre-surface, which is exactly what \"escalated, not silently resolved\" means. If\na second run does surface a `redundant` or `stale` finding, that is either\nnewly introduced since the first run (something else changed the file) or a\nsign the previous apply pass didn't fully take — investigate rather than\nassuming it's expected.\n",{"name":5,"description":7,"argument-hint":38},[32],"main","3af5e96e991385c47f6d792f323ba4d44b4a7caa","devproc\u002Fskills\u002Finternal-docs-prune",0,"https:\u002F\u002Fgithub.com\u002Fplwhite\u002Fclaudeplugins","https:\u002F\u002Fgithub.com\u002Fplwhite","2026-11-01 21:12:04",{"source":8,"skillName":5,"commands":47,"prompt":57,"zipAvailable":58},[48,51,54],{"tool":49,"command":50},"npx","npx skills add plwhite\u002Fclaudeplugins --skill internal-docs-prune",{"tool":52,"command":53},"bunx","bunx skills add plwhite\u002Fclaudeplugins --skill internal-docs-prune",{"tool":55,"command":56},"pnpm","pnpm dlx skills add plwhite\u002Fclaudeplugins --skill internal-docs-prune","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fplwhite\u002Fclaudeplugins\u002Fblob\u002Fmain\u002Fdevproc\u002Fskills\u002Finternal-docs-prune\u002FSKILL.md",true,[],[61,72,77,86,93,120],{"skillNo":62,"name":63,"nameZh":6,"description":64,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":65,"categories":66,"keywords":67,"publishedAt":33},"01M6NJZNMTFQ12V28AKWC6K6VA","greet","Greet a person by name using the greeter tool","https:\u002F\u002Fgithub.com\u002Fplwhite\u002Fclaudeplugins\u002Fblob\u002Fmain\u002Ftests\u002Fdevproc\u002Freview-docs\u002Ffixture\u002Fchange\u002Fskills\u002Fgreet\u002FSKILL.md",[],[63,68,69,70,71],"tool","greeter","name","person",{"skillNo":73,"name":63,"nameZh":6,"description":64,"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":33},"01M6NJZHFSET5D7Y55Q5392JGR","https:\u002F\u002Fgithub.com\u002Fplwhite\u002Fclaudeplugins\u002Fblob\u002Fmain\u002Ftests\u002Fdevproc\u002Freview-docs\u002Ffixture\u002Fbaseline\u002Fskills\u002Fgreet\u002FSKILL.md",[],[63,71,70,69,68],{"skillNo":78,"name":79,"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":80,"categories":81,"keywords":82,"publishedAt":33},"01M6NJZDKAYGVR4Y6TPSMFNARD","review-full","https:\u002F\u002Fgithub.com\u002Fplwhite\u002Fclaudeplugins\u002Fblob\u002Fmain\u002Fdevproc\u002Fskills\u002Freview-full\u002FSKILL.md",[],[83,84,85],"review","full","review full",{"skillNo":87,"name":88,"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":89,"categories":90,"keywords":91,"publishedAt":33},"01M6NJZACEJV6ET66CSGT3JZHF","review-docs","https:\u002F\u002Fgithub.com\u002Fplwhite\u002Fclaudeplugins\u002Fblob\u002Fmain\u002Fdevproc\u002Fskills\u002Freview-docs\u002FSKILL.md",[],[83,19,92],"review docs",{"skillNo":94,"name":95,"nameZh":6,"description":96,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":97,"categories":98,"keywords":104,"publishedAt":33},"01M6NJZ359VNFGGC5YYCDB5V7P","review-component","Code review scoped to a described component or area — resolves the description to files, then runs simplicity, general, and nitty agents; auto-applies code-level findings until convergence; escalates architectural findings for user confirmation","https:\u002F\u002Fgithub.com\u002Fplwhite\u002Fclaudeplugins\u002Fblob\u002Fmain\u002Fdevproc\u002Fskills\u002Freview-component\u002FSKILL.md",[99],{"categoryNo":100,"categoryName":101,"taxonomyType":102,"taxonomyTypeDesc":103},"01JBSEEDCAT000000000000007","代码审查","SCENARIO","使用场景",[105,106,83,107,108,109,110,111,29,27,112,113,114,115,116,117,118,119],"review component","component","scoped","for","convergence","module","general","simplicity","user","confirmation","src","payments","code","auth","code review",{"skillNo":121,"name":122,"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":123,"categories":124,"keywords":125,"publishedAt":33},"01M6NJYYMDJ6QGSTNXVV3GWGRW","review-branch","https:\u002F\u002Fgithub.com\u002Fplwhite\u002Fclaudeplugins\u002Fblob\u002Fmain\u002Fdevproc\u002Fskills\u002Freview-branch\u002FSKILL.md",[],[126,83,127],"branch","review branch"]