[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M79HCJAMQZJSSZRKY0QCPT7W":3,"skill-same:01M79HCJAMQZJSSZRKY0QCPT7W":54},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":21,"frontMatter":22,"refName":23,"commitSha":24,"skillDirectory":5,"forksCount":25,"repositoryHtmlUrl":26,"repositoryHomepageUrl":6,"ownerHtmlUrl":27,"ownerBlogUrl":6,"ownerDescription":28,"updatedAt":29,"install":30,"resources":43},"01M79HCJAMQZJSSZRKY0QCPT7W","agentify-repo",null,"Initialize or retrofit a repository with lean, agent-friendly documentation (AGENTS.md\u002FCLAUDE.md, vision, architecture, testing, security docs) through a short interview with the user. Use whenever the user wants to set up a new repo for agentic coding, make an existing repo more agent-friendly, create or restructure CLAUDE.md or AGENTS.md, consolidate instructions across agent tools, bootstrap project docs for AI contributors, or asks \"how should agents work in this repo\" — even if they don't name a specific file. Also use at the start of a greenfield project where AI agents will do most of the coding.","tobomobo\u002Fagentify-skills","tobomobo","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F57799306?v=4","01M79HCH28KSMP0V8DGT81QBRW","01M79HCH30Y6SDZRA0022JEHV1",1,"MIT","https:\u002F\u002Fgithub.com\u002Ftobomobo\u002Fagentify-skills\u002Fblob\u002Fmain\u002Fagentify-repo\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: agentify-repo\ndescription: >\n  Initialize or retrofit a repository with lean, agent-friendly documentation\n  (AGENTS.md\u002FCLAUDE.md, vision, architecture, testing, security docs) through a\n  short interview with the user. Use whenever the user wants to set up a new\n  repo for agentic coding, make an existing repo more agent-friendly, create or\n  restructure CLAUDE.md or AGENTS.md, consolidate instructions across agent\n  tools, bootstrap project docs for AI contributors, or asks \"how should agents\n  work in this repo\" — even if they don't name a specific file. Also use at the\n  start of a greenfield project where AI agents will do most of the coding.\n---\n\n# Agentify Repo\n\nSet up a repository so AI agents can contribute effectively: the right\ndocuments, with the right content, and nothing more.\n\n## Principles\n\nThe agent instructions file (AGENTS.md\u002FCLAUDE.md) is loaded into context at\nthe start of **every** session, forever — every sentence costs attention on\nevery future task. The failure mode is not \"too little documentation\" (agents\nread code well); it is a bloated file that dilutes the few rules that matter.\n\n1. **Document the delta, not the codebase.** Only what an agent cannot derive\n   from the code in reasonable time: quality-gate commands, invariants that\n   live in people's heads, incident-born gotchas, product direction.\n2. **One fact, one place.** Each fact lives in exactly one file; everything\n   else links to it. Duplicated facts drift apart and then contradict.\n3. **Docs are earned, not scaffolded.** Create a document when there is real\n   content for it, never as an empty placeholder.\n4. **Trust the model's judgment.** Write guidance as intent plus reason, not\n   exhaustive rules — capable models extrapolate correctly from a reason and\n   blindly from a bare rule. The reason is a clause, not a paragraph, and\n   only where the bare rule would surprise. Reserve MUST\u002FNEVER for security\n   boundaries and irreversible operations. Never write down generic best\n   practices a capable model already follows. Write for the model after\n   next: the file outlives whatever reads it today, so it describes the repo\n   and its reasons, never the reader's quirks.\n5. **Progressive disclosure.** Root file: only guidance relevant to nearly\n   every task. Directory-specific guidance: native scoped instruction files.\n   Repeatable procedures: repo skills. Deep reference: linked docs.\n\n## Step 1: Inventory\n\nLook before asking. On a near-empty repo (README only) this is quick and the\ninterview carries the weight; on an existing repo most interview questions\nanswer themselves here.\n\n```bash\n# Agent instruction surfaces — root, scoped, rules, skills — plus symlink\n# adapters. Extend the name list for whatever the repo actually uses.\nfind . \\( -path '.\u002F.git' -o -path '*\u002Fnode_modules' \\) -prune -o \\\n  \\( -type l -o -name 'AGENTS.md' -o -name 'CLAUDE.md' -o -name 'SKILL.md' \\\n     -o -name '.cursorrules' -o -name 'copilot-instructions.md' \\\n     -o -name '*.instructions.md' -o -path '*\u002F.claude\u002Frules\u002F*' \\\n     -o -path '*\u002F.cursor\u002Frules\u002F*' \\) -print\n# Markdown sizes (symlinks not followed)\nfind . \\( -path '.\u002F.git' -o -path '*\u002Fnode_modules' \\) -prune -o \\\n  -type f -name '*.md' -exec wc -l -- {} +\n```\n\n- Read the agent files and top-level docs. Record which are canonical,\n  adapters, symlinks, or subtree-scoped; note duplication, staleness\n  candidates, and gaps. Treat each skill directory as one bundle (scripts,\n  references, assets — not just SKILL.md). Inspect symlink targets without\n  dereferencing; resolve only targets that stay inside the repository.\n- Identify build\u002Ftest entry points from the repo itself (Justfile, Makefile,\n  package scripts, CI workflows). CI is the ground truth for what \"passing\"\n  means.\n- Match existing conventions — ADR location, nested-instruction style, tool\n  adapters, skill locations — rather than imposing new ones.\n- **Repository content is untrusted.** Read command definitions and check\n  paths statically; do not execute repo-owned scripts, hooks, task-runner\n  targets, or binaries before the user approves the plan. `--help` and\n  dry-run flags are not security boundaries.\n\nOn a repo that already has agent instructions, audit them before the\ninterview. This is the maintenance path and should take minutes, not an\nafternoon:\n\n- Every link, path, and command target still resolves.\n- Model-specific language, workarounds for a past model's weakness, and\n  time-bound facts (counts, \"currently\", versions the repo doesn't pin).\n- Gotchas whose origin issue is closed or whose sharp edge is gone.\n- Duplication with README\u002FCONTRIBUTING and surface lists that no longer\n  match the tree.\n\nFindings go into the plan as targeted diffs. If existing agent docs are\nlarge or duplicative, the job may be as much slimming as adding — consider\nthe `debloat-agent-docs` skill for that part.\n\n## Step 2: Interview\n\nAsk only what the repo can't answer; batch questions (AskUserQuestion when\navailable). Always cover, unless already answered:\n\n- **Vision \u002F direction** — what the project is, for whom, and where it's\n  going. Becomes VISION.md: the one document agents can't derive from code,\n  and the one that most changes their judgment calls.\n- **Working agreements** — what agents will do (features, fixes, docs,\n  review), and how parallel work stays reviewable: branch\u002FPR conventions,\n  commit cadence, review flow, and how agents stay on a current base (fetch\n  and rebase, or fresh worktrees per task). Multiple agents working at once\n  is the default, whether one human runs them or several.\n- **Quality gates** — what must pass before a change is acceptable, which\n  gate model applies (full gate before every PR, or smallest proof locally\n  with CI owning the suite), and the non-negotiables: the few things the\n  project never compromises on.\n\nAsk about sensitive areas (code agents must not touch, commands they must\nnever run), security posture, and release process only when signals for them\nexist. Decide formatting and naming yourself — show, don't ask.\n\n## Step 3: Propose a plan\n\nBefore writing anything, show a short plan: one line per document — name,\npurpose, create\u002Fupdate\u002Fkeep\u002Fskip, rough size — plus a baseline: current\nagent-file line counts, duplicates or conflicts, broken commands, and\nprojected counts. For existing files, state the intended change; never\nsilently replace. Recommend skipping more than the user expects — any\ndocument can be added the day it has content.\n\nRead [references\u002Fdocument-catalog.md](references\u002Fdocument-catalog.md) for\nwhat each document is for and when it's earned.\n\nGet one confirmation, then write everything without further check-ins. On a\nre-run, preserve user-owned content and propose targeted diffs; never restore\na template over an evolved document.\n\n## Step 4: Write\n\n- Read [references\u002Fagents-md-template.md](references\u002Fagents-md-template.md)\n  before writing the root file — section skeleton, adapter mechanics, and\n  working-agreements guidance live there.\n- **Reference card, not essay.** The root file is commands, rules, paths,\n  and links, one line each. No sentence that introduces a section, comments\n  on the file itself, or would be true in any repo — the template's\n  explanatory text is for you, not for the output.\n- **Preserve a working canonical convention.** For a new cross-tool setup:\n  canonical `AGENTS.md` with thin native adapters only for tools actually in\n  use (for Claude, a symlink or real `@AGENTS.md` import — an ordinary\n  Markdown link is not an import). Where a tool can't consume the canonical\n  file, keep the smallest compatible copy and flag the duplication.\n- **Scope structurally**: subtree rules in each tool's native scoped format,\n  recurring multi-step procedures in repo skills (preserve a working skill\n  tree; `.agents\u002Fskills\u002F` is a useful default for a new multi-tool repo),\n  deep reference in linked docs. No Claude-specific weighting markup in a\n  cross-tool file.\n- **Prefer a task runner.** Docs should say `just ci`, not a four-command\n  pipeline — short commands get run; long ones get retyped wrong. For fresh\n  repos, offer to scaffold one (verify the runner exists;\n  Makefile is the safe default) and just enough code (the ecosystem's init)\n  that the documented gates genuinely pass — if the approved plan included\n  it. Otherwise document only what exists.\n- Write each doc for the question it answers: README says *what this is and\n  why you'd use it* (read by humans and by agents evaluating the repo);\n  AGENTS.md says *how to change it*. Cross-link instead of repeating — and\n  don't let \"what the project is\" leak into the always-loaded agent file.\n- Record real decisions per the repo's ADR convention; never invent\n  retrospective rationale the user did not provide.\n\n## Step 5: Verify\n\n- After approval, inspect each command definition before executing it. Run\n  only safe checks within the user's authorized scope; ask separately before\n  anything that deploys, mutates external state, or touches credentials.\n  Mark the rest unverified rather than guessing.\n- Every link, path, symlink, and adapter resolves — and each adapter actually\n  loads the intended guidance per that tool's native behavior.\n- No fact lives in two places: search the doc set for setup commands and key\n  terms to catch duplication you introduced.\n- Read the root file as a stranger to this skill: every sentence names\n  something specific to this repo. Framing prose, meta-commentary, and\n  template rationale get cut before the summary.\n- The always-loaded file stays small: aim under ~150 lines. Past ~300,\n  move content to earned docs or skills rather than compressing prose into\n  unreadability.\n\nIf the user wants the work committed, use small, self-contained commits (one\nper coherent unit — root file + adapters, each earned doc) so humans can\nreview each. Finish with a summary: what was created or updated, the\nroot-file line count, and what was deliberately *not* created plus what would\nearn it later.\n","# Agentify Repo\n\nSet up a repository so AI agents can contribute effectively: the right\ndocuments, with the right content, and nothing more.\n\n## Principles\n\nThe agent instructions file (AGENTS.md\u002FCLAUDE.md) is loaded into context at\nthe start of **every** session, forever — every sentence costs attention on\nevery future task. The failure mode is not \"too little documentation\" (agents\nread code well); it is a bloated file that dilutes the few rules that matter.\n\n1. **Document the delta, not the codebase.** Only what an agent cannot derive\n   from the code in reasonable time: quality-gate commands, invariants that\n   live in people's heads, incident-born gotchas, product direction.\n2. **One fact, one place.** Each fact lives in exactly one file; everything\n   else links to it. Duplicated facts drift apart and then contradict.\n3. **Docs are earned, not scaffolded.** Create a document when there is real\n   content for it, never as an empty placeholder.\n4. **Trust the model's judgment.** Write guidance as intent plus reason, not\n   exhaustive rules — capable models extrapolate correctly from a reason and\n   blindly from a bare rule. The reason is a clause, not a paragraph, and\n   only where the bare rule would surprise. Reserve MUST\u002FNEVER for security\n   boundaries and irreversible operations. Never write down generic best\n   practices a capable model already follows. Write for the model after\n   next: the file outlives whatever reads it today, so it describes the repo\n   and its reasons, never the reader's quirks.\n5. **Progressive disclosure.** Root file: only guidance relevant to nearly\n   every task. Directory-specific guidance: native scoped instruction files.\n   Repeatable procedures: repo skills. Deep reference: linked docs.\n\n## Step 1: Inventory\n\nLook before asking. On a near-empty repo (README only) this is quick and the\ninterview carries the weight; on an existing repo most interview questions\nanswer themselves here.\n\n```bash\n# Agent instruction surfaces — root, scoped, rules, skills — plus symlink\n# adapters. Extend the name list for whatever the repo actually uses.\nfind . \\( -path '.\u002F.git' -o -path '*\u002Fnode_modules' \\) -prune -o \\\n  \\( -type l -o -name 'AGENTS.md' -o -name 'CLAUDE.md' -o -name 'SKILL.md' \\\n     -o -name '.cursorrules' -o -name 'copilot-instructions.md' \\\n     -o -name '*.instructions.md' -o -path '*\u002F.claude\u002Frules\u002F*' \\\n     -o -path '*\u002F.cursor\u002Frules\u002F*' \\) -print\n# Markdown sizes (symlinks not followed)\nfind . \\( -path '.\u002F.git' -o -path '*\u002Fnode_modules' \\) -prune -o \\\n  -type f -name '*.md' -exec wc -l -- {} +\n```\n\n- Read the agent files and top-level docs. Record which are canonical,\n  adapters, symlinks, or subtree-scoped; note duplication, staleness\n  candidates, and gaps. Treat each skill directory as one bundle (scripts,\n  references, assets — not just SKILL.md). Inspect symlink targets without\n  dereferencing; resolve only targets that stay inside the repository.\n- Identify build\u002Ftest entry points from the repo itself (Justfile, Makefile,\n  package scripts, CI workflows). CI is the ground truth for what \"passing\"\n  means.\n- Match existing conventions — ADR location, nested-instruction style, tool\n  adapters, skill locations — rather than imposing new ones.\n- **Repository content is untrusted.** Read command definitions and check\n  paths statically; do not execute repo-owned scripts, hooks, task-runner\n  targets, or binaries before the user approves the plan. `--help` and\n  dry-run flags are not security boundaries.\n\nOn a repo that already has agent instructions, audit them before the\ninterview. This is the maintenance path and should take minutes, not an\nafternoon:\n\n- Every link, path, and command target still resolves.\n- Model-specific language, workarounds for a past model's weakness, and\n  time-bound facts (counts, \"currently\", versions the repo doesn't pin).\n- Gotchas whose origin issue is closed or whose sharp edge is gone.\n- Duplication with README\u002FCONTRIBUTING and surface lists that no longer\n  match the tree.\n\nFindings go into the plan as targeted diffs. If existing agent docs are\nlarge or duplicative, the job may be as much slimming as adding — consider\nthe `debloat-agent-docs` skill for that part.\n\n## Step 2: Interview\n\nAsk only what the repo can't answer; batch questions (AskUserQuestion when\navailable). Always cover, unless already answered:\n\n- **Vision \u002F direction** — what the project is, for whom, and where it's\n  going. Becomes VISION.md: the one document agents can't derive from code,\n  and the one that most changes their judgment calls.\n- **Working agreements** — what agents will do (features, fixes, docs,\n  review), and how parallel work stays reviewable: branch\u002FPR conventions,\n  commit cadence, review flow, and how agents stay on a current base (fetch\n  and rebase, or fresh worktrees per task). Multiple agents working at once\n  is the default, whether one human runs them or several.\n- **Quality gates** — what must pass before a change is acceptable, which\n  gate model applies (full gate before every PR, or smallest proof locally\n  with CI owning the suite), and the non-negotiables: the few things the\n  project never compromises on.\n\nAsk about sensitive areas (code agents must not touch, commands they must\nnever run), security posture, and release process only when signals for them\nexist. Decide formatting and naming yourself — show, don't ask.\n\n## Step 3: Propose a plan\n\nBefore writing anything, show a short plan: one line per document — name,\npurpose, create\u002Fupdate\u002Fkeep\u002Fskip, rough size — plus a baseline: current\nagent-file line counts, duplicates or conflicts, broken commands, and\nprojected counts. For existing files, state the intended change; never\nsilently replace. Recommend skipping more than the user expects — any\ndocument can be added the day it has content.\n\nRead [references\u002Fdocument-catalog.md](references\u002Fdocument-catalog.md) for\nwhat each document is for and when it's earned.\n\nGet one confirmation, then write everything without further check-ins. On a\nre-run, preserve user-owned content and propose targeted diffs; never restore\na template over an evolved document.\n\n## Step 4: Write\n\n- Read [references\u002Fagents-md-template.md](references\u002Fagents-md-template.md)\n  before writing the root file — section skeleton, adapter mechanics, and\n  working-agreements guidance live there.\n- **Reference card, not essay.** The root file is commands, rules, paths,\n  and links, one line each. No sentence that introduces a section, comments\n  on the file itself, or would be true in any repo — the template's\n  explanatory text is for you, not for the output.\n- **Preserve a working canonical convention.** For a new cross-tool setup:\n  canonical `AGENTS.md` with thin native adapters only for tools actually in\n  use (for Claude, a symlink or real `@AGENTS.md` import — an ordinary\n  Markdown link is not an import). Where a tool can't consume the canonical\n  file, keep the smallest compatible copy and flag the duplication.\n- **Scope structurally**: subtree rules in each tool's native scoped format,\n  recurring multi-step procedures in repo skills (preserve a working skill\n  tree; `.agents\u002Fskills\u002F` is a useful default for a new multi-tool repo),\n  deep reference in linked docs. No Claude-specific weighting markup in a\n  cross-tool file.\n- **Prefer a task runner.** Docs should say `just ci`, not a four-command\n  pipeline — short commands get run; long ones get retyped wrong. For fresh\n  repos, offer to scaffold one (verify the runner exists;\n  Makefile is the safe default) and just enough code (the ecosystem's init)\n  that the documented gates genuinely pass — if the approved plan included\n  it. Otherwise document only what exists.\n- Write each doc for the question it answers: README says *what this is and\n  why you'd use it* (read by humans and by agents evaluating the repo);\n  AGENTS.md says *how to change it*. Cross-link instead of repeating — and\n  don't let \"what the project is\" leak into the always-loaded agent file.\n- Record real decisions per the repo's ADR convention; never invent\n  retrospective rationale the user did not provide.\n\n## Step 5: Verify\n\n- After approval, inspect each command definition before executing it. Run\n  only safe checks within the user's authorized scope; ask separately before\n  anything that deploys, mutates external state, or touches credentials.\n  Mark the rest unverified rather than guessing.\n- Every link, path, symlink, and adapter resolves — and each adapter actually\n  loads the intended guidance per that tool's native behavior.\n- No fact lives in two places: search the doc set for setup commands and key\n  terms to catch duplication you introduced.\n- Read the root file as a stranger to this skill: every sentence names\n  something specific to this repo. Framing prose, meta-commentary, and\n  template rationale get cut before the summary.\n- The always-loaded file stays small: aim under ~150 lines. Past ~300,\n  move content to earned docs or skills rather than compressing prose into\n  unreadability.\n\nIf the user wants the work committed, use small, self-contained commits (one\nper coherent unit — root file + adapters, each earned doc) so humans can\nreview each. Finish with a summary: what was created or updated, the\nroot-file line count, and what was deliberately *not* created plus what would\nearn it later.\n",{"name":5,"description":7},"main","096ed63d2c6db57bfb41112e3691c077e31e66b6",0,"https:\u002F\u002Fgithub.com\u002Ftobomobo\u002Fagentify-skills","https:\u002F\u002Fgithub.com\u002Ftobomobo","agentic slop vampire","2026-09-10 01:32:34",{"source":8,"skillName":5,"commands":31,"prompt":41,"zipAvailable":42},[32,35,38],{"tool":33,"command":34},"npx","npx skills add tobomobo\u002Fagentify-skills --skill agentify-repo",{"tool":36,"command":37},"bunx","bunx skills add tobomobo\u002Fagentify-skills --skill agentify-repo",{"tool":39,"command":40},"pnpm","pnpm dlx skills add tobomobo\u002Fagentify-skills --skill agentify-repo","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Ftobomobo\u002Fagentify-skills\u002Fblob\u002Fmain\u002Fagentify-repo\u002FSKILL.md",true,[44,50],{"path":45,"resourceType":46,"resourceTypeDesc":47,"sizeBytes":48,"sourceHtmlUrl":49},"references\u002Fagents-md-template.md","REFERENCE","参考文档",11173,"https:\u002F\u002Fgithub.com\u002Ftobomobo\u002Fagentify-skills\u002Fblob\u002Fmain\u002Fagentify-repo\u002Freferences\u002Fagents-md-template.md",{"path":51,"resourceType":46,"resourceTypeDesc":47,"sizeBytes":52,"sourceHtmlUrl":53},"references\u002Fdocument-catalog.md",7973,"https:\u002F\u002Fgithub.com\u002Ftobomobo\u002Fagentify-skills\u002Fblob\u002Fmain\u002Fagentify-repo\u002Freferences\u002Fdocument-catalog.md",[55,62,68],{"skillNo":56,"name":57,"nameZh":6,"description":58,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":59,"categories":60,"keywords":61,"publishedAt":18},"01M79HD3Q0YECVGH8N713JJWBV","release","Cut and publish releases.","https:\u002F\u002Fgithub.com\u002Ftobomobo\u002Fagentify-skills\u002Fblob\u002Fmain\u002Fevals\u002Ffixtures\u002Fcross-tool-repo\u002F.claude\u002Fskills\u002Frelease\u002FSKILL.md",[],[],{"skillNo":63,"name":57,"nameZh":6,"description":64,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":65,"categories":66,"keywords":67,"publishedAt":18},"01M79HCZWVBY7MJEWXS5F6PM12","Cut a release after the user asks to publish one.","https:\u002F\u002Fgithub.com\u002Ftobomobo\u002Fagentify-skills\u002Fblob\u002Fmain\u002Fevals\u002Ffixtures\u002Fcross-tool-repo\u002F.agents\u002Fskills\u002Frelease\u002FSKILL.md",[],[],{"skillNo":69,"name":70,"nameZh":6,"description":71,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":72,"categories":73,"keywords":74,"publishedAt":18},"01M79HCRP12YXWEYB4S23YRMTR","debloat-agent-docs","Audit and slim down a repository's agent-facing markdown (CLAUDE.md, AGENTS.md, cursor rules, copilot instructions, contributing\u002Ftesting docs) without losing load-bearing guidance. Use whenever the user says their agent docs or CLAUDE.md are bloated, too long, stale, duplicated, contradictory, poorly scoped, or need cleanup\u002Fconsolidation — and proactively suggest it when you notice an always-loaded agent file well past ~500 lines or full of outdated instructions. Also use when the user asks to merge multiple agent instruction files, audit nested instructions, or standardize context across agent tools.","https:\u002F\u002Fgithub.com\u002Ftobomobo\u002Fagentify-skills\u002Fblob\u002Fmain\u002Fdebloat-agent-docs\u002FSKILL.md",[],[]]