[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6NQNHTK6ZBN8J2Y1S26RYM4":3,"skill-same:01M6NQNHTK6ZBN8J2Y1S26RYM4":112},{"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":23,"publishedAt":44,"linkOnly":45,"markdownContent":46,"markdownBody":47,"frontMatter":48,"refName":80,"commitSha":81,"skillDirectory":82,"forksCount":14,"repositoryHtmlUrl":83,"repositoryHomepageUrl":84,"ownerHtmlUrl":85,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":86,"install":87,"resources":100},"01M6NQNHTK6ZBN8J2Y1S26RYM4","skill-reviewer",null,"Audit an existing processkit skill against the 5 Skill Killers, the Agent Skills standard, and Anthropic's troubleshooting playbook, producing a categorized findings report and draft Gotchas.","projectious-work\u002Fprocesskit","projectious-work","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F263889726?v=4","01M6N97TJJ118WQ35CHRV8RD0E","projectious.work","01M6N97TPWJPF0BMKYQ4PBY5W8",0,"MIT","https:\u002F\u002Fgithub.com\u002Fprojectious-work\u002Fprocesskit\u002Fblob\u002Fmain\u002Fsrc\u002Fcontext\u002Fskills\u002Fprocesskit\u002Fskill-reviewer\u002FSKILL.md",[18],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43],"skill reviewer","reviewer","agent skills standard","review","Existing","canonical","event","events","creation","full","project","wed","builder","log","work","Rules","history","playbook","audit","run","2026-11-10 10:12:30",false,"---\nname: skill-reviewer\ndescription: >\n  Audit an existing processkit skill against the 5 Skill Killers,\n  the Agent Skills standard, and Anthropic's troubleshooting playbook,\n  producing a categorized findings report and draft Gotchas.\nmetadata:\n  processkit:\n    apiVersion: processkit.projectious.work\u002Fv2\n    id: SKILL-skill-reviewer\n    version: \"1.0.0\"\n    created: 2026-04-08T00:00:00Z\n    category: processkit\n    layer: 3\n    uses:\n      - skill: skill-builder\n        purpose: \"Reference the canonical skill format and the 5 Skill Killers checklist that skill-builder enforces during creation. skill-reviewer applies the same rules to existing skills.\"\n      - skill: index-management\n        purpose: \"Locate the skill being reviewed and surface related skills it should reference under uses:.\"\n      - skill: event-log\n        purpose: \"Log skill.reviewed events so the project's audit trail captures review history.\"\n    provides:\n      primitives: []\n      mcp_tools: []\n      assets: []\n      processes: [skill-review, gotchas-generation]\n    commands:\n      - name: pk-skill-audit\n        args: \"skill-name\"\n        description: \"Run a full 13-category review of the named skill\"\n---\n\n# Skill Reviewer\n\n## Intro\n\n`skill-reviewer` audits an existing processkit skill against the\ncanonical format, the 5 Skill Killers, the Agent Skills standard, and\nAnthropic's troubleshooting playbook. It produces a categorized findings\nreport (must-fix \u002F should-fix \u002F nit) and, crucially, **generates draft\nGotchas content for any skill whose Gotchas section is missing, empty,\nor generic** — the killer feature that lets one skill-reviewer pass\npopulate the Gotchas section of every skill in the catalog.\n\nskill-reviewer is the brownfield counterpart to `skill-builder`:\nskill-builder is for greenfield (creating new skills), skill-reviewer\nis for everything else (auditing, fixing, retrofitting).\n\n## Overview\n\n### When to use skill-reviewer\n\n- The user says *\"review the X skill\"*, *\"check skills\u002FY\"*, *\"is this\n  skill good\"*, *\"audit skills\u002F\"*, *\"lint the X skill\"*, or *\"generate\n  gotchas for X\"*.\n- A new skill has just been created and you want a second-pass quality\n  check before committing.\n- The Agent Skills standard or the format spec changed, and existing\n  skills need a sweep.\n- Bulk operation: **populate Gotchas across the whole catalog** by\n  running skill-reviewer over every skill in `src\u002Fskills\u002F`.\n\n### The review checklist (13 categories)\n\nFor each skill, check all 13 categories. Surface findings under one\nof the three severity buckets at the end.\n\n#### 1. Frontmatter compliance (Agent Skills standard)\n\n- Top-level `name` exists, kebab-case, matches the directory name.\n- Top-level `name` does NOT contain \"claude\" or \"anthropic\" (reserved\n  by Anthropic).\n- Top-level `description` exists and is \u003C 1024 characters.\n- Top-level `description` includes BOTH what AND when (with literal\n  trigger phrases the user would say).\n- NO XML angle brackets (`\u003C` \u002F `>`) anywhere in frontmatter.\n- `metadata.processkit.apiVersion` is present and equals\n  `processkit.projectious.work\u002Fv1`.\n- `metadata.processkit.id` is present and equals `SKILL-\u003Cname>`.\n- `metadata.processkit.version` is a valid semver string.\n- `metadata.processkit.created` is a valid ISO 8601 timestamp.\n- `metadata.processkit.category` is one of the registered values\n  (`process`, `language`, `framework`, `infrastructure`, `architecture`,\n  `design`, `data`, `ai`, `api`, `security`, `observability`,\n  `database`, `performance`, `meta`).\n- For `category: process`, `metadata.processkit.layer` is an integer\n  0-4. For other categories, `layer` is `null` or absent.\n- Every `uses:` entry is an OBJECT with a `skill` field AND a `purpose`\n  field. Bare strings or missing-purpose entries are must-fix.\n- Every skill name listed under `uses:` actually exists in\n  `src\u002Fskills\u002F`.\n\n#### 2. Section structure\n\n- The body has these four section headings IN THIS ORDER:\n  `## Intro`, `## Overview`, `## Gotchas`, `## Full reference`.\n- No headings starting with `## Level 1`, `## Level 2`, or `## Level 3`\n  (those were dropped in v0.6.0).\n- The Intro is 1-3 sentences. Longer is should-fix.\n- The Gotchas section is REQUIRED (must-fix if missing).\n\n#### 3. The 5 Skill Killers (Skills Master Class)\n\n| # | Killer | Symptom | Fix |\n|---|---|---|---|\n| 1 | Description doesn't trigger | Reads like a summary or contains \"Use when…\" clauses. | Rewrite as a one-sentence imperative (verb-first, under 200 chars). Move trigger phrases to skill-finder's trigger table (Category 9). |\n| 2 | Over-defining process | Steps prescribed too rigidly when the agent should have degrees of freedom. | Loosen prescriptive language; distinguish \"must\" from \"should\" from \"may\". |\n| 3 | Stating the obvious | Content the agent already knows from general training. | Cut. (BUT: be careful — what's obvious to Claude may not be to Gemini. Provider neutrality wins ties.) |\n| 4 | Missing Gotchas | No `## Gotchas` section, empty section, or generic \"be careful\" content. | Generate draft gotchas — see Step 7 below. |\n| 5 | Monolithic blob | SKILL.md > 5000 words. | Push detail to `references\u002F\u003Ctopic>.md`. |\n\n#### 4. Anthropic troubleshooting playbook (PDF p25-27)\n\nCheck for the four diagnoses:\n\n- **Under-trigger** signals: skill doesn't load when it should; users\n  manually enabling it; support questions about when to use it. Fix:\n  add detail\u002Fkeywords to the description, especially for technical\n  terms.\n- **Over-trigger** signals: skill loads for irrelevant queries; users\n  disabling it; confusion about purpose. Fix: add negative triggers,\n  be more specific.\n- **Instructions not followed**: Overview is too verbose, critical\n  instructions buried, ambiguous language, or model laziness. Fix:\n  put critical instructions at the top; use `## Important` headers;\n  prefer code over prose; add explicit encouragement for thorough\n  steps.\n- **Large context**: SKILL.md too large; too many enabled skills\n  simultaneously. Fix: 5000-word ceiling; references\u002F extraction.\n\n#### 5. Directory structure\n\n- `SKILL.md` exists at the skill root (case-sensitive, exactly that\n  name).\n- No `README.md` inside the skill folder (any depth).\n- No `INDEX.md` inside the skill folder (parent `src\u002Fskills\u002FINDEX.md`\n  is fine).\n- `scripts\u002F` directory exists (may be empty with `.gitkeep`).\n- If a `templates\u002F` directory exists, that's a leftover from before\n  v0.6.0 — should be `assets\u002F`.\n- If `mcp\u002F` exists: `server.py` and `mcp-config.json` are present;\n  `SERVER.md` is present (NOT `README.md`).\n- If `mcp\u002F` exists: the server name in both `mcp-config.json` and the\n  `FastMCP(...)` call in `server.py` is prefixed with `processkit-`.\n- If `metadata.processkit.commands` is non-empty: a `commands\u002F`\n  directory exists with one file per entry in the list.\n- If `commands\u002F` exists but `metadata.processkit.commands` is absent\n  or empty: flag as should-fix (orphaned adapter files).\n- If harness projection directories exist, `.claude\u002Fcommands\u002F` and\n  `.agents\u002Fskills\u002F` contain exactly the same command names as the canonical\n  `commands\u002F` files.\n\n#### 6. Description quality\n\nBeyond the \u003C 1024 char length check:\n\n- Description is a **one-sentence imperative** — verb-first (e.g.,\n  \"Author X\", \"Review Y for Z\", \"Generate W\"). NOT narrative (\"This\n  skill helps you…\") and NOT passive (\"A tool for…\"). Violation =\n  should-fix.\n- Description is **under 200 characters**. Anything longer is usually\n  a sign of \"Use when…\" clauses being smuggled in — should-fix.\n- Description does **NOT** contain \"Use when the user says…\" or\n  similar trigger language. Trigger phrases belong in skill-finder's\n  trigger-phrase table (Category 9), not here. Violation = must-fix.\n- Description starts with the verb, not the noun. \"Recommend models\"\n  beats \"Model recommendation skill that recommends models.\" Violation\n  = nit.\n\n#### 7. Generate Gotchas (the killer feature)\n\nIf the Gotchas section is missing, empty, or generic, generate a draft.\n\nProcess:\n1. Read the skill's Overview section to understand what it does.\n2. Read the skill's `provides:` block to understand what it produces.\n3. Read the skill's `uses:` block to understand what it depends on.\n4. For each, ask the five Gotcha-generation questions from skill-builder\n   Step 5a:\n   - What would an agent get wrong on the first try?\n   - What does the agent need to KNOW that the user wouldn't think to\n     say?\n   - Where would the agent helpfully overshoot?\n   - Where would the agent helpfully undershoot?\n   - What conventions are unique to this skill that the agent has to\n     learn?\n5. Draft 5-10 specific, provider-neutral failure modes.\n6. Each draft Gotcha must have BOTH a specific failure AND a specific\n   countermeasure. Generic items (\"be careful\", \"double-check\") are\n   not acceptable drafts.\n7. Cross-reference Anti-patterns under Full reference if some draft\n   items overlap — Gotchas should be agent-only.\n\nFor domain skills (language\u002Fframework\u002Finfrastructure), useful seed\npatterns to consider:\n\n- Hallucinated APIs (the agent invents library functions that don't\n  exist).\n- Stale-syntax (the agent uses syntax from an older language version\n  than the project targets).\n- Ignoring project-specific conventions in favor of \"best practice\"\n  defaults.\n- Skipping verification steps when \"the obvious thing\" looks right.\n- Provider blind spots (agents on harnesses without filesystem access\n  hallucinating file contents).\n\n#### 8. Cross-reference quality\n\n- Skills mentioned in Overview\u002FFull reference exist in `src\u002Fskills\u002F`\n  (no broken references).\n- The `uses:` block is consistent with the body — every dependency\n  mentioned in the body should appear in `uses:`, and every dependency\n  in `uses:` should be referenced at least once in the body with its\n  purpose explained.\n\n#### 9. skill-finder registration\n\n- The skill has at least 1 entry in the trigger-phrase table in\n  `src\u002Fskills\u002Fskill-finder\u002FSKILL.md`. Missing = must-fix — a skill\n  not in skill-finder is invisible to agents who don't know its name.\n- The skill has a one-liner in the appropriate by-category section of\n  `skill-finder`. Missing = should-fix.\n- The skill's `description:` field does NOT contain \"Use when the user\n  says…\" or similar trigger language — those phrases belong here in\n  the trigger table, not in the description. Violation = must-fix\n  (cross-links with Category 6).\n\n#### 10. Command adapter hygiene\n\nRun this category only if `metadata.processkit.commands` is non-empty OR\n`commands\u002F` directory exists.\n\n- Every entry in `metadata.processkit.commands` has a corresponding\n  `commands\u002F\u003Cname>.md` file. Missing file = must-fix.\n- Every `commands\u002F\u003Cname>.md` file has a corresponding entry in\n  `metadata.processkit.commands`. Orphan file = should-fix.\n- Each processkit command name starts with the reserved `pk-` prefix.\n  Unprefixed processkit commands = must-fix.\n- The `argument-hint` in each adapter file matches the `args` field in\n  the corresponding `metadata.processkit.commands` entry. Mismatch =\n  should-fix.\n- `.claude\u002Fcommands\u002F\u003Cname>.md` has the same `argument-hint` as the\n  canonical adapter. Mismatch = must-fix.\n- `.agents\u002Fskills\u002F\u003Cname>\u002FSKILL.md` exists for each canonical command when\n  the project ships `.agents\u002Fskills\u002F`. Missing or extra command shims =\n  must-fix.\n- Each adapter file body is exactly one sentence invoking the skill\n  and workflow. Multi-line bodies with workflow steps are should-fix —\n  logic belongs in SKILL.md, not in the adapter.\n- `allowed-tools` in each adapter is present and scoped narrowly.\n  `Bash(*)` without justification = should-fix.\n- SKILL.md Overview mentions the commands (one line per command\n  referencing the adapter path). Absent mention = should-fix.\n\n#### 11. Security and permission audit\n\nRun this category for all skills (not only those with MCP servers).\n\n**MCP tools** (if `mcp\u002F` exists):\n\n- Every `@server.tool()` has explicit annotations. Missing annotations\n  on any tool = must-fix.\n- `readOnlyHint: true` on all query \u002F list \u002F get \u002F profile tools that\n  do not write state. Missing = should-fix.\n- `destructiveHint: true` on any tool that deletes, overwrites, or\n  resets persistent state. Missing = must-fix.\n- `openWorldHint: true` on any tool that makes HTTP or external API\n  calls. Missing = must-fix.\n- `idempotentHint: true` on tools where a duplicate call is harmless\n  (e.g. delete by ID, set-config to same value). Missing = nit.\n\n**SKILL.md permission surface:**\n\n- If the skill calls MCP tools marked `destructiveHint: true`, the\n  Gotchas section mentions that the agent must show the operation to\n  the user and confirm before calling. Missing = should-fix.\n- If the skill calls MCP tools marked `openWorldHint: true`, the\n  Gotchas section notes that external calls are best-effort and\n  failures should not be retried silently. Missing = should-fix.\n- If the skill ships scripts in `scripts\u002F` that write to the\n  filesystem outside the skill's own directory, or make network\n  calls, the Overview section includes a **Permissions** note listing\n  what the scripts touch. Missing = should-fix.\n\n#### 12. Behavioral completeness\n\nCheck that the skill encodes enough to prevent silent execution\nfailures — cases where the agent understands the workflow but then\nfails to execute a step.\n\n- **Skills that direct the agent to create entities** (WorkItems,\n  DecisionRecords, Discussions, etc.): the Gotchas section must\n  include a rule about calling the tool in the same turn as the\n  commitment. Missing = should-fix.\n- **Skills with multi-step workflows**: each step that produces a\n  persistent artifact (entity, file, commit) should have a\n  corresponding Gotcha about not deferring the write. Missing =\n  should-fix.\n- **All skills**: are Gotchas specific (failure described +\n  concrete countermeasure), or are they generic encouragements\n  (\"be careful\", \"double-check\") with no actionable fix? All-generic\n  Gotchas = should-fix.\n- **Process skills in particular**: at least one Gotcha must cover\n  the gap between verbal commitment (\"I'll do X\") and execution\n  (\"called the tool\"). Missing = should-fix.\n\n#### 13. Library expert specialization\n\nRun this category when the skill is a library expert, API expert, SDK\nguide, framework recipe book, or similar version-sensitive domain skill.\n\n- `metadata.library` exists with `name`, `version`, `homepage`,\n  `docs`, and `source_date`. Missing `metadata.library` on a library\n  expert = must-fix; missing individual fields = should-fix.\n- The Intro or Overview states the version range the skill covers.\n  Ambiguous \"latest\" claims are should-fix unless backed by a live-doc\n  workflow.\n- Recipe examples use one major version's API at a time. Mixed major\n  versions = must-fix because they generate broken copy-paste code.\n- Each recipe includes imports, setup, or prerequisites needed to run\n  it. Missing runnable context = should-fix.\n- The skill has a RAG Escalation or live-doc escalation section that\n  says when to leave baked-in recipes. Missing escalation policy =\n  should-fix.\n- Gotchas include stale API risk and project-version mismatch risk.\n  Missing both = should-fix.\n\n### The findings report\n\nAfter running through all 13 categories, output a structured report:\n\n```markdown\n# Review: \u003Cskill-name>\n\n## Must-fix (severity: high)\n\n- [Category N] \u003Cfinding>: \u003Cfix>\n\n## Should-fix (severity: medium)\n\n- [Category N] \u003Cfinding>: \u003Cfix>\n\n## Nit (severity: low)\n\n- [Category N] \u003Cfinding>: \u003Cfix>\n\n## Generated Gotchas draft\n\n(Only if Category 7 was triggered.)\n\n[5-10 draft gotchas in the standard format]\n\n## Trigger test cases\n\n(Only generated for skills with weak descriptions.)\n\nShould trigger:\n  - \"...\"\nShould NOT trigger:\n  - \"...\"\n```\n\n### Bulk operation: gotchas-pass over all skills\n\nTo populate Gotchas across the catalog in one batch:\n\n1. Iterate over every `src\u002Fskills\u002F*\u002FSKILL.md`.\n2. For each, run only Category 7 (skip the other categories for speed).\n3. Skip skills whose Gotchas section is already present and substantive\n   (≥ 3 specific items, not all generic).\n4. Output the draft Gotchas as a patch the user can apply.\n5. The user reviews the patch and either accepts it as-is, or polishes\n   each draft before committing.\n\nThis is the workflow that produced the initial Gotchas content for\nprocesskit's 100+ pre-existing skills (Thread 1 task #4).\n\nA similar bulk pass can be run for **Category 11 (security audit)**:\niterate every skill with an `mcp\u002F` directory and check that all tools\nhave annotations. Output missing annotations as a patch the user can\napply to each `server.py`.\n\nA bulk pass for **Category 12 (behavioral completeness)** is useful\nafter any behavioral norm is added to AGENTS.md — verify that the\ncorresponding skill's Gotchas already encode the same rule.\n\nThis skill also provides the `\u002Fpk-skill-audit` command for direct\ninvocation — see `commands\u002Fpk-skill-audit.md`.\n\n## Gotchas\n\nAgent-specific failure modes when running skill-reviewer:\n\n- **Treating \"Stating the obvious\" as license to gut content.** Killer\n  #3 says cut what the agent already knows — but provider neutrality\n  trumps. If something is obvious to Claude but not Gemini, KEEP IT.\n  When in doubt, leave content in.\n- **Generating Gotchas that just describe the skill's normal use.**\n  *\"Always remember to call the function\"* is not a gotcha; it's\n  documentation. Gotchas describe failure modes and countermeasures,\n  not happy paths.\n- **Generating identical Gotchas across multiple skills.** Each skill's\n  Gotchas should be specific to that skill. If the same item appears\n  for python-best-practices and rust-conventions, it's too generic —\n  push it into a shared meta-skill or remove it.\n- **Reporting nits as must-fix.** \"Must-fix\" is reserved for things\n  that genuinely break the skill (frontmatter invalid, Gotchas\n  missing entirely, name colliding with reserved words). Style and\n  tone preferences are nits or should-fixes at most. Inflating\n  severity trains users to ignore the report.\n- **Skipping Category 1 because the file \"looks right\".** Frontmatter\n  bugs are the most common silent failure. ALWAYS run Category 1\n  even when the skill body looks polished.\n- **Hallucinating Gotchas based on the skill's name alone.** Read the\n  Overview and Full reference before drafting Gotchas. A title-only\n  draft will produce generic content that the user has to throw\n  away.\n- **Forgetting to check `uses:` references actually exist.** A typo\n  in a skill name under `uses:` (e.g., `index-managment`) silently\n  breaks delegation. This is a must-fix that's easy to miss.\n- **Reviewing your own work without skepticism.** If skill-reviewer\n  is being used to review a skill that skill-builder just created,\n  the temptation is to rubber-stamp. Run all 13 categories anyway —\n  the whole point of the brownfield\u002Fgreenfield split is independent\n  validation.\n- **Skipping Category 10 because no commands\u002F folder is visible.**\n  Category 10 must also run when `metadata.processkit.commands` is\n  non-empty even if the `commands\u002F` directory is absent — the absence\n  itself is the must-fix finding.\n- **Skipping Category 11 for \"knowledge-only\" skills.** Even skills\n  without an MCP server can ship scripts or instruct the agent to call\n  external services. Run the SKILL.md permission surface checks for\n  every skill, not only those with `mcp\u002F`.\n- **Treating missing annotations as a nit.** Missing MCP tool\n  annotations are must-fix, not nits. A harness cannot surface\n  confirmation prompts for destructive tools if it doesn't know the\n  tool is destructive. The user may unknowingly approve state-destroying\n  operations.\n\n## Full reference\n\n### Severity definitions\n\n| Severity | Definition | Examples |\n|---|---|---|\n| **Must-fix** | The skill is broken or non-compliant in a way that prevents it from working. | Missing frontmatter, invalid YAML, missing Gotchas, name collision with reserved words, broken `uses:` reference, `mcp\u002FREADME.md` violation. |\n| **Should-fix** | The skill works but has known quality issues that will cause problems for users. | Generic Gotchas, description without trigger phrases, monolithic SKILL.md > 5000 words, buried critical instructions. |\n| **Nit** | Style or tone preference. The skill works fine; reasonable people would disagree on whether to fix. | Verbose Intro (4-5 sentences instead of 1-3), inconsistent capitalization in headings, redundant cross-references. |\n\n### Running skill-reviewer over a skill\n\nManual flow:\n1. Read the SKILL.md file in full.\n2. Walk through Categories 1-13.\n3. For each finding, classify severity.\n4. Generate the report in the format above.\n\n### Running the bulk gotchas-pass\n\nFor populating Gotchas across the catalog:\n\n1. Identify the target skills:\n   ```\n   query_entities(kind=\"Skill\")  # via index-management\n   ```\n2. For each, check whether the Gotchas section is present and substantive:\n   - Present: section heading `## Gotchas` exists.\n   - Substantive: 3+ items, each with both a specific failure AND a\n     specific countermeasure (not generic).\n3. For skills failing the check, generate a Gotchas draft using the\n   process in Category 7.\n4. Output as a series of patches (one per skill) that can be applied\n   in batch.\n\n### Anti-patterns when reviewing\n\n- **Drive-by review.** Skimming the SKILL.md and producing a one-line\n  report is worse than no review. Take the time.\n- **\"Nothing wrong\"** as a verdict. There's almost always something\n  worth raising as a should-fix or nit. If you found nothing,\n  re-read with the 5 Skill Killers in mind.\n- **Suggesting rewrites of working code\u002Fcontent.** skill-reviewer\n  surfaces problems and proposes specific fixes. It does NOT rewrite\n  the whole skill — that's the user's call.\n- **Reviewing without checking the format spec.** The canonical\n  source of truth is `src\u002Fskills\u002FFORMAT.md`. Re-read it if you're\n  unsure about a convention.\n\n### Cross-references\n\n- `skill-builder` — the greenfield counterpart; same checklist applied\n  during creation.\n- `index-management` — query existing skills; locate the skill being\n  reviewed; check that `uses:` references resolve.\n- `event-log` — log `skill.reviewed` events with the findings count.\n- `src\u002Fskills\u002FFORMAT.md` — the canonical format spec.\n","# Skill Reviewer\n\n## Intro\n\n`skill-reviewer` audits an existing processkit skill against the\ncanonical format, the 5 Skill Killers, the Agent Skills standard, and\nAnthropic's troubleshooting playbook. It produces a categorized findings\nreport (must-fix \u002F should-fix \u002F nit) and, crucially, **generates draft\nGotchas content for any skill whose Gotchas section is missing, empty,\nor generic** — the killer feature that lets one skill-reviewer pass\npopulate the Gotchas section of every skill in the catalog.\n\nskill-reviewer is the brownfield counterpart to `skill-builder`:\nskill-builder is for greenfield (creating new skills), skill-reviewer\nis for everything else (auditing, fixing, retrofitting).\n\n## Overview\n\n### When to use skill-reviewer\n\n- The user says *\"review the X skill\"*, *\"check skills\u002FY\"*, *\"is this\n  skill good\"*, *\"audit skills\u002F\"*, *\"lint the X skill\"*, or *\"generate\n  gotchas for X\"*.\n- A new skill has just been created and you want a second-pass quality\n  check before committing.\n- The Agent Skills standard or the format spec changed, and existing\n  skills need a sweep.\n- Bulk operation: **populate Gotchas across the whole catalog** by\n  running skill-reviewer over every skill in `src\u002Fskills\u002F`.\n\n### The review checklist (13 categories)\n\nFor each skill, check all 13 categories. Surface findings under one\nof the three severity buckets at the end.\n\n#### 1. Frontmatter compliance (Agent Skills standard)\n\n- Top-level `name` exists, kebab-case, matches the directory name.\n- Top-level `name` does NOT contain \"claude\" or \"anthropic\" (reserved\n  by Anthropic).\n- Top-level `description` exists and is \u003C 1024 characters.\n- Top-level `description` includes BOTH what AND when (with literal\n  trigger phrases the user would say).\n- NO XML angle brackets (`\u003C` \u002F `>`) anywhere in frontmatter.\n- `metadata.processkit.apiVersion` is present and equals\n  `processkit.projectious.work\u002Fv1`.\n- `metadata.processkit.id` is present and equals `SKILL-\u003Cname>`.\n- `metadata.processkit.version` is a valid semver string.\n- `metadata.processkit.created` is a valid ISO 8601 timestamp.\n- `metadata.processkit.category` is one of the registered values\n  (`process`, `language`, `framework`, `infrastructure`, `architecture`,\n  `design`, `data`, `ai`, `api`, `security`, `observability`,\n  `database`, `performance`, `meta`).\n- For `category: process`, `metadata.processkit.layer` is an integer\n  0-4. For other categories, `layer` is `null` or absent.\n- Every `uses:` entry is an OBJECT with a `skill` field AND a `purpose`\n  field. Bare strings or missing-purpose entries are must-fix.\n- Every skill name listed under `uses:` actually exists in\n  `src\u002Fskills\u002F`.\n\n#### 2. Section structure\n\n- The body has these four section headings IN THIS ORDER:\n  `## Intro`, `## Overview`, `## Gotchas`, `## Full reference`.\n- No headings starting with `## Level 1`, `## Level 2`, or `## Level 3`\n  (those were dropped in v0.6.0).\n- The Intro is 1-3 sentences. Longer is should-fix.\n- The Gotchas section is REQUIRED (must-fix if missing).\n\n#### 3. The 5 Skill Killers (Skills Master Class)\n\n| # | Killer | Symptom | Fix |\n|---|---|---|---|\n| 1 | Description doesn't trigger | Reads like a summary or contains \"Use when…\" clauses. | Rewrite as a one-sentence imperative (verb-first, under 200 chars). Move trigger phrases to skill-finder's trigger table (Category 9). |\n| 2 | Over-defining process | Steps prescribed too rigidly when the agent should have degrees of freedom. | Loosen prescriptive language; distinguish \"must\" from \"should\" from \"may\". |\n| 3 | Stating the obvious | Content the agent already knows from general training. | Cut. (BUT: be careful — what's obvious to Claude may not be to Gemini. Provider neutrality wins ties.) |\n| 4 | Missing Gotchas | No `## Gotchas` section, empty section, or generic \"be careful\" content. | Generate draft gotchas — see Step 7 below. |\n| 5 | Monolithic blob | SKILL.md > 5000 words. | Push detail to `references\u002F\u003Ctopic>.md`. |\n\n#### 4. Anthropic troubleshooting playbook (PDF p25-27)\n\nCheck for the four diagnoses:\n\n- **Under-trigger** signals: skill doesn't load when it should; users\n  manually enabling it; support questions about when to use it. Fix:\n  add detail\u002Fkeywords to the description, especially for technical\n  terms.\n- **Over-trigger** signals: skill loads for irrelevant queries; users\n  disabling it; confusion about purpose. Fix: add negative triggers,\n  be more specific.\n- **Instructions not followed**: Overview is too verbose, critical\n  instructions buried, ambiguous language, or model laziness. Fix:\n  put critical instructions at the top; use `## Important` headers;\n  prefer code over prose; add explicit encouragement for thorough\n  steps.\n- **Large context**: SKILL.md too large; too many enabled skills\n  simultaneously. Fix: 5000-word ceiling; references\u002F extraction.\n\n#### 5. Directory structure\n\n- `SKILL.md` exists at the skill root (case-sensitive, exactly that\n  name).\n- No `README.md` inside the skill folder (any depth).\n- No `INDEX.md` inside the skill folder (parent `src\u002Fskills\u002FINDEX.md`\n  is fine).\n- `scripts\u002F` directory exists (may be empty with `.gitkeep`).\n- If a `templates\u002F` directory exists, that's a leftover from before\n  v0.6.0 — should be `assets\u002F`.\n- If `mcp\u002F` exists: `server.py` and `mcp-config.json` are present;\n  `SERVER.md` is present (NOT `README.md`).\n- If `mcp\u002F` exists: the server name in both `mcp-config.json` and the\n  `FastMCP(...)` call in `server.py` is prefixed with `processkit-`.\n- If `metadata.processkit.commands` is non-empty: a `commands\u002F`\n  directory exists with one file per entry in the list.\n- If `commands\u002F` exists but `metadata.processkit.commands` is absent\n  or empty: flag as should-fix (orphaned adapter files).\n- If harness projection directories exist, `.claude\u002Fcommands\u002F` and\n  `.agents\u002Fskills\u002F` contain exactly the same command names as the canonical\n  `commands\u002F` files.\n\n#### 6. Description quality\n\nBeyond the \u003C 1024 char length check:\n\n- Description is a **one-sentence imperative** — verb-first (e.g.,\n  \"Author X\", \"Review Y for Z\", \"Generate W\"). NOT narrative (\"This\n  skill helps you…\") and NOT passive (\"A tool for…\"). Violation =\n  should-fix.\n- Description is **under 200 characters**. Anything longer is usually\n  a sign of \"Use when…\" clauses being smuggled in — should-fix.\n- Description does **NOT** contain \"Use when the user says…\" or\n  similar trigger language. Trigger phrases belong in skill-finder's\n  trigger-phrase table (Category 9), not here. Violation = must-fix.\n- Description starts with the verb, not the noun. \"Recommend models\"\n  beats \"Model recommendation skill that recommends models.\" Violation\n  = nit.\n\n#### 7. Generate Gotchas (the killer feature)\n\nIf the Gotchas section is missing, empty, or generic, generate a draft.\n\nProcess:\n1. Read the skill's Overview section to understand what it does.\n2. Read the skill's `provides:` block to understand what it produces.\n3. Read the skill's `uses:` block to understand what it depends on.\n4. For each, ask the five Gotcha-generation questions from skill-builder\n   Step 5a:\n   - What would an agent get wrong on the first try?\n   - What does the agent need to KNOW that the user wouldn't think to\n     say?\n   - Where would the agent helpfully overshoot?\n   - Where would the agent helpfully undershoot?\n   - What conventions are unique to this skill that the agent has to\n     learn?\n5. Draft 5-10 specific, provider-neutral failure modes.\n6. Each draft Gotcha must have BOTH a specific failure AND a specific\n   countermeasure. Generic items (\"be careful\", \"double-check\") are\n   not acceptable drafts.\n7. Cross-reference Anti-patterns under Full reference if some draft\n   items overlap — Gotchas should be agent-only.\n\nFor domain skills (language\u002Fframework\u002Finfrastructure), useful seed\npatterns to consider:\n\n- Hallucinated APIs (the agent invents library functions that don't\n  exist).\n- Stale-syntax (the agent uses syntax from an older language version\n  than the project targets).\n- Ignoring project-specific conventions in favor of \"best practice\"\n  defaults.\n- Skipping verification steps when \"the obvious thing\" looks right.\n- Provider blind spots (agents on harnesses without filesystem access\n  hallucinating file contents).\n\n#### 8. Cross-reference quality\n\n- Skills mentioned in Overview\u002FFull reference exist in `src\u002Fskills\u002F`\n  (no broken references).\n- The `uses:` block is consistent with the body — every dependency\n  mentioned in the body should appear in `uses:`, and every dependency\n  in `uses:` should be referenced at least once in the body with its\n  purpose explained.\n\n#### 9. skill-finder registration\n\n- The skill has at least 1 entry in the trigger-phrase table in\n  `src\u002Fskills\u002Fskill-finder\u002FSKILL.md`. Missing = must-fix — a skill\n  not in skill-finder is invisible to agents who don't know its name.\n- The skill has a one-liner in the appropriate by-category section of\n  `skill-finder`. Missing = should-fix.\n- The skill's `description:` field does NOT contain \"Use when the user\n  says…\" or similar trigger language — those phrases belong here in\n  the trigger table, not in the description. Violation = must-fix\n  (cross-links with Category 6).\n\n#### 10. Command adapter hygiene\n\nRun this category only if `metadata.processkit.commands` is non-empty OR\n`commands\u002F` directory exists.\n\n- Every entry in `metadata.processkit.commands` has a corresponding\n  `commands\u002F\u003Cname>.md` file. Missing file = must-fix.\n- Every `commands\u002F\u003Cname>.md` file has a corresponding entry in\n  `metadata.processkit.commands`. Orphan file = should-fix.\n- Each processkit command name starts with the reserved `pk-` prefix.\n  Unprefixed processkit commands = must-fix.\n- The `argument-hint` in each adapter file matches the `args` field in\n  the corresponding `metadata.processkit.commands` entry. Mismatch =\n  should-fix.\n- `.claude\u002Fcommands\u002F\u003Cname>.md` has the same `argument-hint` as the\n  canonical adapter. Mismatch = must-fix.\n- `.agents\u002Fskills\u002F\u003Cname>\u002FSKILL.md` exists for each canonical command when\n  the project ships `.agents\u002Fskills\u002F`. Missing or extra command shims =\n  must-fix.\n- Each adapter file body is exactly one sentence invoking the skill\n  and workflow. Multi-line bodies with workflow steps are should-fix —\n  logic belongs in SKILL.md, not in the adapter.\n- `allowed-tools` in each adapter is present and scoped narrowly.\n  `Bash(*)` without justification = should-fix.\n- SKILL.md Overview mentions the commands (one line per command\n  referencing the adapter path). Absent mention = should-fix.\n\n#### 11. Security and permission audit\n\nRun this category for all skills (not only those with MCP servers).\n\n**MCP tools** (if `mcp\u002F` exists):\n\n- Every `@server.tool()` has explicit annotations. Missing annotations\n  on any tool = must-fix.\n- `readOnlyHint: true` on all query \u002F list \u002F get \u002F profile tools that\n  do not write state. Missing = should-fix.\n- `destructiveHint: true` on any tool that deletes, overwrites, or\n  resets persistent state. Missing = must-fix.\n- `openWorldHint: true` on any tool that makes HTTP or external API\n  calls. Missing = must-fix.\n- `idempotentHint: true` on tools where a duplicate call is harmless\n  (e.g. delete by ID, set-config to same value). Missing = nit.\n\n**SKILL.md permission surface:**\n\n- If the skill calls MCP tools marked `destructiveHint: true`, the\n  Gotchas section mentions that the agent must show the operation to\n  the user and confirm before calling. Missing = should-fix.\n- If the skill calls MCP tools marked `openWorldHint: true`, the\n  Gotchas section notes that external calls are best-effort and\n  failures should not be retried silently. Missing = should-fix.\n- If the skill ships scripts in `scripts\u002F` that write to the\n  filesystem outside the skill's own directory, or make network\n  calls, the Overview section includes a **Permissions** note listing\n  what the scripts touch. Missing = should-fix.\n\n#### 12. Behavioral completeness\n\nCheck that the skill encodes enough to prevent silent execution\nfailures — cases where the agent understands the workflow but then\nfails to execute a step.\n\n- **Skills that direct the agent to create entities** (WorkItems,\n  DecisionRecords, Discussions, etc.): the Gotchas section must\n  include a rule about calling the tool in the same turn as the\n  commitment. Missing = should-fix.\n- **Skills with multi-step workflows**: each step that produces a\n  persistent artifact (entity, file, commit) should have a\n  corresponding Gotcha about not deferring the write. Missing =\n  should-fix.\n- **All skills**: are Gotchas specific (failure described +\n  concrete countermeasure), or are they generic encouragements\n  (\"be careful\", \"double-check\") with no actionable fix? All-generic\n  Gotchas = should-fix.\n- **Process skills in particular**: at least one Gotcha must cover\n  the gap between verbal commitment (\"I'll do X\") and execution\n  (\"called the tool\"). Missing = should-fix.\n\n#### 13. Library expert specialization\n\nRun this category when the skill is a library expert, API expert, SDK\nguide, framework recipe book, or similar version-sensitive domain skill.\n\n- `metadata.library` exists with `name`, `version`, `homepage`,\n  `docs`, and `source_date`. Missing `metadata.library` on a library\n  expert = must-fix; missing individual fields = should-fix.\n- The Intro or Overview states the version range the skill covers.\n  Ambiguous \"latest\" claims are should-fix unless backed by a live-doc\n  workflow.\n- Recipe examples use one major version's API at a time. Mixed major\n  versions = must-fix because they generate broken copy-paste code.\n- Each recipe includes imports, setup, or prerequisites needed to run\n  it. Missing runnable context = should-fix.\n- The skill has a RAG Escalation or live-doc escalation section that\n  says when to leave baked-in recipes. Missing escalation policy =\n  should-fix.\n- Gotchas include stale API risk and project-version mismatch risk.\n  Missing both = should-fix.\n\n### The findings report\n\nAfter running through all 13 categories, output a structured report:\n\n```markdown\n# Review: \u003Cskill-name>\n\n## Must-fix (severity: high)\n\n- [Category N] \u003Cfinding>: \u003Cfix>\n\n## Should-fix (severity: medium)\n\n- [Category N] \u003Cfinding>: \u003Cfix>\n\n## Nit (severity: low)\n\n- [Category N] \u003Cfinding>: \u003Cfix>\n\n## Generated Gotchas draft\n\n(Only if Category 7 was triggered.)\n\n[5-10 draft gotchas in the standard format]\n\n## Trigger test cases\n\n(Only generated for skills with weak descriptions.)\n\nShould trigger:\n  - \"...\"\nShould NOT trigger:\n  - \"...\"\n```\n\n### Bulk operation: gotchas-pass over all skills\n\nTo populate Gotchas across the catalog in one batch:\n\n1. Iterate over every `src\u002Fskills\u002F*\u002FSKILL.md`.\n2. For each, run only Category 7 (skip the other categories for speed).\n3. Skip skills whose Gotchas section is already present and substantive\n   (≥ 3 specific items, not all generic).\n4. Output the draft Gotchas as a patch the user can apply.\n5. The user reviews the patch and either accepts it as-is, or polishes\n   each draft before committing.\n\nThis is the workflow that produced the initial Gotchas content for\nprocesskit's 100+ pre-existing skills (Thread 1 task #4).\n\nA similar bulk pass can be run for **Category 11 (security audit)**:\niterate every skill with an `mcp\u002F` directory and check that all tools\nhave annotations. Output missing annotations as a patch the user can\napply to each `server.py`.\n\nA bulk pass for **Category 12 (behavioral completeness)** is useful\nafter any behavioral norm is added to AGENTS.md — verify that the\ncorresponding skill's Gotchas already encode the same rule.\n\nThis skill also provides the `\u002Fpk-skill-audit` command for direct\ninvocation — see `commands\u002Fpk-skill-audit.md`.\n\n## Gotchas\n\nAgent-specific failure modes when running skill-reviewer:\n\n- **Treating \"Stating the obvious\" as license to gut content.** Killer\n  #3 says cut what the agent already knows — but provider neutrality\n  trumps. If something is obvious to Claude but not Gemini, KEEP IT.\n  When in doubt, leave content in.\n- **Generating Gotchas that just describe the skill's normal use.**\n  *\"Always remember to call the function\"* is not a gotcha; it's\n  documentation. Gotchas describe failure modes and countermeasures,\n  not happy paths.\n- **Generating identical Gotchas across multiple skills.** Each skill's\n  Gotchas should be specific to that skill. If the same item appears\n  for python-best-practices and rust-conventions, it's too generic —\n  push it into a shared meta-skill or remove it.\n- **Reporting nits as must-fix.** \"Must-fix\" is reserved for things\n  that genuinely break the skill (frontmatter invalid, Gotchas\n  missing entirely, name colliding with reserved words). Style and\n  tone preferences are nits or should-fixes at most. Inflating\n  severity trains users to ignore the report.\n- **Skipping Category 1 because the file \"looks right\".** Frontmatter\n  bugs are the most common silent failure. ALWAYS run Category 1\n  even when the skill body looks polished.\n- **Hallucinating Gotchas based on the skill's name alone.** Read the\n  Overview and Full reference before drafting Gotchas. A title-only\n  draft will produce generic content that the user has to throw\n  away.\n- **Forgetting to check `uses:` references actually exist.** A typo\n  in a skill name under `uses:` (e.g., `index-managment`) silently\n  breaks delegation. This is a must-fix that's easy to miss.\n- **Reviewing your own work without skepticism.** If skill-reviewer\n  is being used to review a skill that skill-builder just created,\n  the temptation is to rubber-stamp. Run all 13 categories anyway —\n  the whole point of the brownfield\u002Fgreenfield split is independent\n  validation.\n- **Skipping Category 10 because no commands\u002F folder is visible.**\n  Category 10 must also run when `metadata.processkit.commands` is\n  non-empty even if the `commands\u002F` directory is absent — the absence\n  itself is the must-fix finding.\n- **Skipping Category 11 for \"knowledge-only\" skills.** Even skills\n  without an MCP server can ship scripts or instruct the agent to call\n  external services. Run the SKILL.md permission surface checks for\n  every skill, not only those with `mcp\u002F`.\n- **Treating missing annotations as a nit.** Missing MCP tool\n  annotations are must-fix, not nits. A harness cannot surface\n  confirmation prompts for destructive tools if it doesn't know the\n  tool is destructive. The user may unknowingly approve state-destroying\n  operations.\n\n## Full reference\n\n### Severity definitions\n\n| Severity | Definition | Examples |\n|---|---|---|\n| **Must-fix** | The skill is broken or non-compliant in a way that prevents it from working. | Missing frontmatter, invalid YAML, missing Gotchas, name collision with reserved words, broken `uses:` reference, `mcp\u002FREADME.md` violation. |\n| **Should-fix** | The skill works but has known quality issues that will cause problems for users. | Generic Gotchas, description without trigger phrases, monolithic SKILL.md > 5000 words, buried critical instructions. |\n| **Nit** | Style or tone preference. The skill works fine; reasonable people would disagree on whether to fix. | Verbose Intro (4-5 sentences instead of 1-3), inconsistent capitalization in headings, redundant cross-references. |\n\n### Running skill-reviewer over a skill\n\nManual flow:\n1. Read the SKILL.md file in full.\n2. Walk through Categories 1-13.\n3. For each finding, classify severity.\n4. Generate the report in the format above.\n\n### Running the bulk gotchas-pass\n\nFor populating Gotchas across the catalog:\n\n1. Identify the target skills:\n   ```\n   query_entities(kind=\"Skill\")  # via index-management\n   ```\n2. For each, check whether the Gotchas section is present and substantive:\n   - Present: section heading `## Gotchas` exists.\n   - Substantive: 3+ items, each with both a specific failure AND a\n     specific countermeasure (not generic).\n3. For skills failing the check, generate a Gotchas draft using the\n   process in Category 7.\n4. Output as a series of patches (one per skill) that can be applied\n   in batch.\n\n### Anti-patterns when reviewing\n\n- **Drive-by review.** Skimming the SKILL.md and producing a one-line\n  report is worse than no review. Take the time.\n- **\"Nothing wrong\"** as a verdict. There's almost always something\n  worth raising as a should-fix or nit. If you found nothing,\n  re-read with the 5 Skill Killers in mind.\n- **Suggesting rewrites of working code\u002Fcontent.** skill-reviewer\n  surfaces problems and proposes specific fixes. It does NOT rewrite\n  the whole skill — that's the user's call.\n- **Reviewing without checking the format spec.** The canonical\n  source of truth is `src\u002Fskills\u002FFORMAT.md`. Re-read it if you're\n  unsure about a convention.\n\n### Cross-references\n\n- `skill-builder` — the greenfield counterpart; same checklist applied\n  during creation.\n- `index-management` — query existing skills; locate the skill being\n  reviewed; check that `uses:` references resolve.\n- `event-log` — log `skill.reviewed` events with the findings count.\n- `src\u002Fskills\u002FFORMAT.md` — the canonical format spec.\n",{"name":5,"metadata":49,"description":79},{"processkit":50},{"id":51,"uses":52,"layer":62,"created":63,"version":64,"category":65,"commands":66,"provides":71,"apiVersion":78},"SKILL-skill-reviewer",[53,56,59],{"skill":54,"purpose":55},"skill-builder","Reference the canonical skill format and the 5 Skill Killers checklist that skill-builder enforces during creation. skill-reviewer applies the same rules to existing skills.",{"skill":57,"purpose":58},"index-management","Locate the skill being reviewed and surface related skills it should reference under uses:.",{"skill":60,"purpose":61},"event-log","Log skill.reviewed events so the project's audit trail captures review history.",3,1775606400000,"1.0.0","processkit",[67],{"args":68,"name":69,"description":70},"skill-name","pk-skill-audit","Run a full 13-category review of the named skill",{"assets":72,"mcp_tools":73,"processes":74,"primitives":77},[],[],[75,76],"skill-review","gotchas-generation",[],"processkit.projectious.work\u002Fv2","Audit an existing processkit skill against the 5 Skill Killers, the Agent Skills standard, and Anthropic's troubleshooting playbook, producing a categorized findings report and draft Gotchas.\n","main","7231f100e266e30d2c1699ecde5c00602471dc3c","src\u002Fcontext\u002Fskills\u002Fprocesskit\u002Fskill-reviewer","https:\u002F\u002Fgithub.com\u002Fprojectious-work\u002Fprocesskit","https:\u002F\u002Fprojectious-work.github.io\u002Fprocesskit\u002F","https:\u002F\u002Fgithub.com\u002Fprojectious-work","2026-11-11 23:24:18",{"source":8,"skillName":5,"commands":88,"prompt":98,"zipAvailable":99},[89,92,95],{"tool":90,"command":91},"npx","npx skills add projectious-work\u002Fprocesskit --skill skill-reviewer",{"tool":93,"command":94},"bunx","bunx skills add projectious-work\u002Fprocesskit --skill skill-reviewer",{"tool":96,"command":97},"pnpm","pnpm dlx skills add projectious-work\u002Fprocesskit --skill skill-reviewer","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fprojectious-work\u002Fprocesskit\u002Fblob\u002Fmain\u002Fsrc\u002Fcontext\u002Fskills\u002Fprocesskit\u002Fskill-reviewer\u002FSKILL.md",true,[101,107],{"path":102,"resourceType":103,"resourceTypeDesc":104,"sizeBytes":105,"sourceHtmlUrl":106},"commands\u002Fpk-skill-audit.md","REFERENCE","参考文档",141,"https:\u002F\u002Fgithub.com\u002Fprojectious-work\u002Fprocesskit\u002Fblob\u002Fmain\u002Fsrc\u002Fcontext\u002Fskills\u002Fprocesskit\u002Fskill-reviewer\u002Fcommands\u002Fpk-skill-audit.md",{"path":108,"resourceType":109,"resourceTypeDesc":110,"sizeBytes":14,"sourceHtmlUrl":111},"scripts\u002F.gitkeep","SCRIPT","脚本","https:\u002F\u002Fgithub.com\u002Fprojectious-work\u002Fprocesskit\u002Fblob\u002Fmain\u002Fsrc\u002Fcontext\u002Fskills\u002Fprocesskit\u002Fskill-reviewer\u002Fscripts\u002F.gitkeep",[113,120,127,133,156,175],{"skillNo":114,"name":115,"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":116,"categories":117,"keywords":118,"publishedAt":119},"01M7665G2ED3624R3X93304265","standard-ui-design","https:\u002F\u002Fgithub.com\u002Fprojectious-work\u002Fprocesskit\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fstandard-ui-design\u002FSKILL.md",[],[],"2026-11-12 13:55:43",{"skillNo":121,"name":122,"nameZh":6,"description":123,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":124,"categories":125,"keywords":126,"publishedAt":119},"01M76651QTC6W0HJ4AFW2MMR3V","standard-agent-use","Company standard for working with AI agents. Use when deciding what to delegate to agents, capturing open work, keeping agent memory or knowledge notes, improving an agent workflow, or introducing agent work in a team or for a client.","https:\u002F\u002Fgithub.com\u002Fprojectious-work\u002Fprocesskit\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fstandard-agent-use\u002FSKILL.md",[],[],{"skillNo":128,"name":129,"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":130,"categories":131,"keywords":132,"publishedAt":119},"01M76650NN4T8P241021NBR64S","standard-agent-building","https:\u002F\u002Fgithub.com\u002Fprojectious-work\u002Fprocesskit\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fstandard-agent-building\u002FSKILL.md",[],[],{"skillNo":134,"name":135,"nameZh":6,"description":136,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":137,"categories":138,"keywords":140,"publishedAt":44},"01M6X5MTC89B4ZGXW3JVSQCPF8","skill-management","Manage Skill package manifests as v1 ontology entities.","https:\u002F\u002Fgithub.com\u002Fprojectious-work\u002Fprocesskit\u002Fblob\u002Fmain\u002Fsrc\u002Fcontext\u002Fskills\u002Fprocesskit\u002Fskill-management\u002FSKILL.md",[139],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},[141,142,143,144,145,146,38,147,148,149,150,151,135,152,153,154,155],"management","skill management","ontology","v1","create","as","transition","get","manage","update","alpha","list","package","skill package","manifests",{"skillNo":157,"name":158,"nameZh":6,"description":159,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":160,"categories":161,"keywords":163,"publishedAt":44},"01M6X5MCV74NHW5MEDNHZRF2C8","schema-management","Manage generated processkit schema contracts through deterministic MCP rebuild and inspection workflows.","https:\u002F\u002Fgithub.com\u002Fprojectious-work\u002Fprocesskit\u002Fblob\u002Fmain\u002Fsrc\u002Fcontext\u002Fskills\u002Fprocesskit\u002Fschema-management\u002FSKILL.md",[162],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},[164,165,141,166,167,168,169,35,170,171,172,173,38,148,149,151,174],"schema","schema management","contract","rebuild","contracts","workflows","generated","MCP","validation","deterministic","mode",{"skillNo":176,"name":177,"nameZh":6,"description":178,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":179,"categories":180,"keywords":181,"publishedAt":44},"01M6X5M5G6YJ96AMZGSCZJ1D13","proposition-management","Record claims, risks, beliefs, world facts, WSJF estimates, and assumptions through the Proposition interface.","https:\u002F\u002Fgithub.com\u002Fprojectious-work\u002Fprocesskit\u002Fblob\u002Fmain\u002Fsrc\u002Fcontext\u002Fskills\u002Fprocesskit\u002Fproposition-management\u002FSKILL.md",[],[182,141,183,184,145,185,186,187,38,148,32,30,150,188,151,189,190,191,192,37],"proposition management","proposition","interface","claims","record","world","WSJF","index","updates","query","facts"]