[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7EJ8YPQHCGA7B7CBV3PX0QT":3,"skill-same:01M7EJ8YPQHCGA7B7CBV3PX0QT":62},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":14,"repositoryHtmlUrl":27,"repositoryHomepageUrl":28,"ownerHtmlUrl":29,"ownerBlogUrl":30,"ownerDescription":6,"updatedAt":31,"install":32,"resources":45},"01M7EJ8YPQHCGA7B7CBV3PX0QT","agent-audit",null,"Audits an AI agent definition (a Claude Code subagent, a plugin's agents, or another framework's agent prompt and config) against the seven layers a production agent needs: model, instructions, tools, memory, loop, guardrails and human approval. A script lints Claude Code agent files for misspelled or ignored fields, no loop cap, no tool boundary and no approval rule; the rest is judged from the file with quoted evidence. Returns one finding per gap with the smallest fix, applies the fixes on request (editing the generator's sources when the agent is generated), and writes delegation test cases. Use when someone asks to review, audit, harden or improve their agents or subagents, asks why an agent misbehaves in production, loops, acts without asking or never gets picked, or wants an agent made production-ready, even if they never say \"audit\". Not for SKILL.md files (use tune-skill), designing a new agent team (use ai-team-blueprint), building a new skill from an idea (use skill-builder), reviewing ordinary code (use consistent-code-review), or red-teaming a live app (use pre-ship-security-red-team).","arm-pz\u002Fcaffeinity-skills","arm-pz","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F29532099?v=4","01M7EJ726AAB9GD0C5C5G9NJEG","Abdul Rahman","01M7EJ726X4ZD3R4ESGX90K47M",0,"MIT","https:\u002F\u002Fgithub.com\u002Farm-pz\u002Fcaffeinity-skills\u002Fblob\u002Fmain\u002Fplugins\u002Fagent-audit\u002Fskills\u002Fagent-audit\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: agent-audit\ndescription: >-\n  Audits an AI agent definition (a Claude Code subagent, a plugin's agents, or another\n  framework's agent prompt and config) against the seven layers a production agent needs:\n  model, instructions, tools, memory, loop, guardrails and human approval. A script lints\n  Claude Code agent files for misspelled or ignored fields, no loop cap, no tool boundary and\n  no approval rule; the rest is judged from the file with quoted evidence. Returns one finding\n  per gap with the smallest fix, applies the fixes on request (editing the generator's sources\n  when the agent is generated), and writes delegation test cases. Use when someone asks to\n  review, audit, harden or improve their agents or subagents, asks why an agent misbehaves in\n  production, loops, acts without asking or never gets picked, or wants an agent made\n  production-ready, even if they never say \"audit\". Not for SKILL.md files (use tune-skill),\n  designing a new agent team (use ai-team-blueprint), building a new skill from an idea (use\n  skill-builder), reviewing ordinary code (use consistent-code-review), or red-teaming a live\n  app (use pre-ship-security-red-team).\n---\n\n# agent-audit\n\nAn agent is a model using tools in a loop. When one fails in production, the model is rarely\nthe cause; usually a layer around it is missing: nothing caps the loop, every tool is granted,\nnothing is remembered between runs, and \"be careful\" stands in for an approval step. This skill\nchecks an agent for each layer and writes down what is missing, with evidence from the file.\n\n## Intake\n\nFind the agent files before asking anything:\n\n- Claude Code: `.claude\u002Fagents\u002F*.md`, `~\u002F.claude\u002Fagents\u002F*.md`, and `agents\u002F*.md` inside any\n  folder that has `.claude-plugin\u002Fplugin.json` (a plugin agent).\n- Other frameworks: the system prompt plus the code that sets tools, turn limits and approvals\n  (`Agent(...)` in the OpenAI Agents SDK, a LangGraph graph, an n8n workflow).\n\nThen ask in one message, only for what the files cannot tell you: what the agent is for, who\nruns it, and which of its actions reach outside (send, publish, delete, spend). If the user\ngives no answer, say what you assumed.\n\nIf a file says it is generated (\"Generated by ...\", \"do not edit\"), find the generator and its\nsources now. Every fix goes there, never into the generated file.\n\n## Workflow\n\n1. **Lint.** For Claude Code agent files run:\n\n       python scripts\u002Fcheck_agent.py \u003Cagent files or folders>\n\n   It fails on what Claude Code would misread (an unknown or misspelled field is silently\n   ignored, an invalid model, color, memory or maxTurns value) and warns on missing layers: no\n   tool boundary, no `maxTurns`, no stop condition, no rule that read content is data, no\n   approval rule, and fields a plugin agent ignores (`hooks`, `permissionMode`, `mcpServers`,\n   `initialPrompt`). Fix every FAIL first. Other frameworks skip this step; apply the same\n   questions by hand from `references\u002Flayers.md`.\n2. **Judge the seven layers.** Work through `references\u002Flayers.md` for each agent. For every\n   layer write PASS, GAP or RISK with a quote from the file or \"absent\". Judgment calls the\n   script cannot make live here: is the description specific enough to win routing against its\n   siblings, which tools are actually needed, and which actions are high-risk for this agent.\n3. **Check the family.** With more than one agent, compare descriptions pairwise: two agents\n   that claim the same request, and every `(use X)` handoff that names an agent or skill that\n   does not exist. Also check each agent against installed skills that do the same job.\n4. **Report.** One table per agent: layer, verdict, evidence, smallest fix. Order the findings\n   by risk: approval and guardrails (L7, L6) first, then loop (L5), tools (L3), memory (L4),\n   instructions (L2), model (L1). Close with what is shared across agents, since a fix in a\n   shared template fixes them all.\n5. **Fix on request.** Make the smallest edit that closes each gap, rerun the script, and show\n   before and after counts. Rebuild a generated agent from its sources and run the repo's own\n   staleness check.\n6. **Write delegation tests.** For each agent, one realistic request that must route to it and\n   one near miss from its \"Not for\" list that must not. If the repo has an eval format, write\n   them in it; otherwise use the table in `references\u002Flayers.md`. Say that they were written,\n   not run, until they are run.\n\n## Rules\n\n- **Evidence or nothing.** Every finding quotes the file or says the line is absent. No finding\n  comes from the agent's name or from what such agents usually do.\n- **An instruction is a wish; a permission is a guardrail.** Where a prompt rule could be a\n  `tools` or `disallowedTools` entry, a hook, or a `permissions.deny` rule, say so. A plugin\n  agent cannot carry hooks or a permission mode; name the project-agent or `hooks\u002Fhooks.json`\n  route instead.\n- **Do not cut tools you have not traced.** Before narrowing `tools`, list what each step\n  actually calls (scripts it runs, files it writes, MCP servers it uses). A narrower list that\n  breaks a step is worse than a warning.\n- **Do not change the model without evals.** Recommend a smaller model only for a step with a\n  passing eval set on that model.\n- **Never run the agent against production** to test it, and never send, publish, delete or\n  spend while auditing.\n\n## Output\n\n```\n\u003Cagent>: \u003Cn> RISK, \u003Cn> GAP, \u003Cn> PASS\n| Layer | Verdict | Evidence | Fix |\n...\nShared across agents: \u003Cone line per template-level fix>\nTests written: \u003Cn> (not run) | run: \u003Cpass>\u002F\u003Ctotal>\n```\n","# agent-audit\n\nAn agent is a model using tools in a loop. When one fails in production, the model is rarely\nthe cause; usually a layer around it is missing: nothing caps the loop, every tool is granted,\nnothing is remembered between runs, and \"be careful\" stands in for an approval step. This skill\nchecks an agent for each layer and writes down what is missing, with evidence from the file.\n\n## Intake\n\nFind the agent files before asking anything:\n\n- Claude Code: `.claude\u002Fagents\u002F*.md`, `~\u002F.claude\u002Fagents\u002F*.md`, and `agents\u002F*.md` inside any\n  folder that has `.claude-plugin\u002Fplugin.json` (a plugin agent).\n- Other frameworks: the system prompt plus the code that sets tools, turn limits and approvals\n  (`Agent(...)` in the OpenAI Agents SDK, a LangGraph graph, an n8n workflow).\n\nThen ask in one message, only for what the files cannot tell you: what the agent is for, who\nruns it, and which of its actions reach outside (send, publish, delete, spend). If the user\ngives no answer, say what you assumed.\n\nIf a file says it is generated (\"Generated by ...\", \"do not edit\"), find the generator and its\nsources now. Every fix goes there, never into the generated file.\n\n## Workflow\n\n1. **Lint.** For Claude Code agent files run:\n\n       python scripts\u002Fcheck_agent.py \u003Cagent files or folders>\n\n   It fails on what Claude Code would misread (an unknown or misspelled field is silently\n   ignored, an invalid model, color, memory or maxTurns value) and warns on missing layers: no\n   tool boundary, no `maxTurns`, no stop condition, no rule that read content is data, no\n   approval rule, and fields a plugin agent ignores (`hooks`, `permissionMode`, `mcpServers`,\n   `initialPrompt`). Fix every FAIL first. Other frameworks skip this step; apply the same\n   questions by hand from `references\u002Flayers.md`.\n2. **Judge the seven layers.** Work through `references\u002Flayers.md` for each agent. For every\n   layer write PASS, GAP or RISK with a quote from the file or \"absent\". Judgment calls the\n   script cannot make live here: is the description specific enough to win routing against its\n   siblings, which tools are actually needed, and which actions are high-risk for this agent.\n3. **Check the family.** With more than one agent, compare descriptions pairwise: two agents\n   that claim the same request, and every `(use X)` handoff that names an agent or skill that\n   does not exist. Also check each agent against installed skills that do the same job.\n4. **Report.** One table per agent: layer, verdict, evidence, smallest fix. Order the findings\n   by risk: approval and guardrails (L7, L6) first, then loop (L5), tools (L3), memory (L4),\n   instructions (L2), model (L1). Close with what is shared across agents, since a fix in a\n   shared template fixes them all.\n5. **Fix on request.** Make the smallest edit that closes each gap, rerun the script, and show\n   before and after counts. Rebuild a generated agent from its sources and run the repo's own\n   staleness check.\n6. **Write delegation tests.** For each agent, one realistic request that must route to it and\n   one near miss from its \"Not for\" list that must not. If the repo has an eval format, write\n   them in it; otherwise use the table in `references\u002Flayers.md`. Say that they were written,\n   not run, until they are run.\n\n## Rules\n\n- **Evidence or nothing.** Every finding quotes the file or says the line is absent. No finding\n  comes from the agent's name or from what such agents usually do.\n- **An instruction is a wish; a permission is a guardrail.** Where a prompt rule could be a\n  `tools` or `disallowedTools` entry, a hook, or a `permissions.deny` rule, say so. A plugin\n  agent cannot carry hooks or a permission mode; name the project-agent or `hooks\u002Fhooks.json`\n  route instead.\n- **Do not cut tools you have not traced.** Before narrowing `tools`, list what each step\n  actually calls (scripts it runs, files it writes, MCP servers it uses). A narrower list that\n  breaks a step is worse than a warning.\n- **Do not change the model without evals.** Recommend a smaller model only for a step with a\n  passing eval set on that model.\n- **Never run the agent against production** to test it, and never send, publish, delete or\n  spend while auditing.\n\n## Output\n\n```\n\u003Cagent>: \u003Cn> RISK, \u003Cn> GAP, \u003Cn> PASS\n| Layer | Verdict | Evidence | Fix |\n...\nShared across agents: \u003Cone line per template-level fix>\nTests written: \u003Cn> (not run) | run: \u003Cpass>\u002F\u003Ctotal>\n```\n",{"name":5,"description":7},"main","523bc2bf98727c9fd981424f86ecba66c6d6ecd6","plugins\u002Fagent-audit\u002Fskills\u002Fagent-audit","https:\u002F\u002Fgithub.com\u002Farm-pz\u002Fcaffeinity-skills","https:\u002F\u002Fskills.caffeinity.com\u002F","https:\u002F\u002Fgithub.com\u002Farm-pz","skills.caffeinity.com","2026-11-14 01:22:10",{"source":8,"skillName":5,"commands":33,"prompt":43,"zipAvailable":44},[34,37,40],{"tool":35,"command":36},"npx","npx skills add arm-pz\u002Fcaffeinity-skills --skill agent-audit",{"tool":38,"command":39},"bunx","bunx skills add arm-pz\u002Fcaffeinity-skills --skill agent-audit",{"tool":41,"command":42},"pnpm","pnpm dlx skills add arm-pz\u002Fcaffeinity-skills --skill agent-audit","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Farm-pz\u002Fcaffeinity-skills\u002Fblob\u002Fmain\u002Fplugins\u002Fagent-audit\u002Fskills\u002Fagent-audit\u002FSKILL.md",true,[46,52,56],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"references\u002Flayers.md","REFERENCE","参考文档",4799,"https:\u002F\u002Fgithub.com\u002Farm-pz\u002Fcaffeinity-skills\u002Fblob\u002Fmain\u002Fplugins\u002Fagent-audit\u002Fskills\u002Fagent-audit\u002Freferences\u002Flayers.md",{"path":53,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":54,"sourceHtmlUrl":55},"references\u002Ftrigger-cases.md",1789,"https:\u002F\u002Fgithub.com\u002Farm-pz\u002Fcaffeinity-skills\u002Fblob\u002Fmain\u002Fplugins\u002Fagent-audit\u002Fskills\u002Fagent-audit\u002Freferences\u002Ftrigger-cases.md",{"path":57,"resourceType":58,"resourceTypeDesc":59,"sizeBytes":60,"sourceHtmlUrl":61},"scripts\u002Fcheck_agent.py","SCRIPT","脚本",12888,"https:\u002F\u002Fgithub.com\u002Farm-pz\u002Fcaffeinity-skills\u002Fblob\u002Fmain\u002Fplugins\u002Fagent-audit\u002Fskills\u002Fagent-audit\u002Fscripts\u002Fcheck_agent.py",[63,70,77,84,91,98],{"skillNo":64,"name":65,"nameZh":6,"description":66,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":67,"categories":68,"keywords":69,"publishedAt":19},"01M7EJR8K8FDPBY02RFPRRW19M","tune-and-ship","Takes one published skill at a time from as-is to tuned, fact-checked, released and live, then moves to the next one in the catalog. Each run tunes the skill with tune-skill, checks every version number, flag, price and command it states against the primary source, re-reads its siblings' real SKILL.md files, polishes the listing with findable-polish, writes the version and release notes, opens one PR, and after the merge confirms the site page and marketplace copy are live. Use whenever the user wants to fine-tune their other skills, work through the catalog, give a published skill the full treatment, fact-check a skill's claims, or ship a tune instead of drafting it, even if they never say \"ship\". Not for a tune with no release (use tune-skill), drafts with nothing pushed (use tune-and-polish), a new skill (use skill-launch), regressions from the nightly health check (use caffeinity-routing's Descale brew), or a whole collection (use skill-estate).","https:\u002F\u002Fgithub.com\u002Farm-pz\u002Fcaffeinity-skills\u002Fblob\u002Fmain\u002Fplugins\u002Ftune-and-ship\u002Fskills\u002Ftune-and-ship\u002FSKILL.md",[],[],{"skillNo":71,"name":72,"nameZh":6,"description":73,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":19},"01M7EJR2D3H3EA8PT9PVR8DVTS","skill-estate","Takes a whole collection of agent skill repos from scattered to published, consistent and self-maintaining, in gated phases: inventory every repo, reconcile site, marketplace and README listings, test and tune the family, close workflow gaps, wire publishing so new skill repos go live on their own, bundle a plugin with agents, add trigger tests and directory listings, and leave a maintenance loop and a map. Each phase hands off to the skill that owns it and ends in PRs. Use when someone's skills site and GitHub disagree, wants all their skill repos audited or cleaned up, wants new skills to publish themselves, or wants that cleanup repeated on another org, even if they never say estate. Not for one skill (use tune-skill), one listing (use findable), the site pipeline alone (use publish-pipeline), or weekly upkeep (use caffeinity-routing's Descale brew).","https:\u002F\u002Fgithub.com\u002Farm-pz\u002Fcaffeinity-skills\u002Fblob\u002Fmain\u002Fplugins\u002Fskill-estate\u002Fskills\u002Fskill-estate\u002FSKILL.md",[],[],{"skillNo":78,"name":79,"nameZh":6,"description":80,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":81,"categories":82,"keywords":83,"publishedAt":19},"01M7EJQVD17QR594SGFS5Z1T34","paper-trail","Drafts the policy pages a small product needs before it takes money or data (privacy policy, terms of service, refund policy, cookie notice) from a fact sheet about what the product actually collects, stores, shares and charges, then checks the drafts for placeholders, missing sections and facts nobody supplied. Use when someone needs a privacy policy, terms, refund or cookie page for a site, app, SaaS, newsletter or digital product, asks what legal pages they need before launch, or wants existing policy pages checked against what the product really does, even if they never say \"legal\". Not legal advice and not for contracts, NDAs or employment documents (use a lawyer), and not for the launch plan itself (use solo-income-blueprint) or the site build (use site-prompt).","https:\u002F\u002Fgithub.com\u002Farm-pz\u002Fcaffeinity-skills\u002Fblob\u002Fmain\u002Fplugins\u002Fpaper-trail\u002Fskills\u002Fpaper-trail\u002FSKILL.md",[],[],{"skillNo":85,"name":86,"nameZh":6,"description":87,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":88,"categories":89,"keywords":90,"publishedAt":19},"01M7EJQM55D4W6D9SJB8WDVP27","course-builder","Designs a course for a group of learners (a class, cohort, workshop series, bootcamp or internal training): measurable outcomes, a module sequence with prerequisites, activities and assessments aligned to each outcome, a time budget that fits the hours available, and a rubric, then checks the plan with a script so no outcome goes untaught or unassessed. Use when someone is teaching others and asks for a curriculum, syllabus, course outline, module plan, cohort or workshop structure, or wants an existing course checked for gaps and overload, even if they never say \"curriculum\". Not for one person's own learning plan (use study-system), for being taught the topic (use topic-primer), or for marketing the course (use sales-copy).","https:\u002F\u002Fgithub.com\u002Farm-pz\u002Fcaffeinity-skills\u002Fblob\u002Fmain\u002Fplugins\u002Fcourse-builder\u002Fskills\u002Fcourse-builder\u002FSKILL.md",[],[],{"skillNo":92,"name":93,"nameZh":6,"description":94,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":95,"categories":96,"keywords":97,"publishedAt":19},"01M7EJQDFT8K0PW1HF7RPS6M03","wordsmith","Write, draft, rewrite, edit, proofread, polish and improve text while preserving the author's intended meaning, facts and voice. Use when the user wants clearer, shorter or better-toned writing; grammar, structure or flow fixes; emails, website or marketing copy, headlines, outlines or summaries; a different reading level; an accessibility, inclusivity or accuracy review; or content prepared for publication, even if they never say \"edit\". Also handles \u002Fwordsmith commands such as draft, critique, audit, tighten, tone and final. Not for resumes, CVs or LinkedIn profiles (use recruiter-lens if installed), not for code or code comments, and not for building .docx or .pdf files (write the text here, then use the file skill).","https:\u002F\u002Fgithub.com\u002Farm-pz\u002Fcaffeinity-skills\u002Fblob\u002Fmain\u002Fplugins\u002Fcaffeinity\u002Fskills\u002Fwordsmith\u002FSKILL.md",[],[],{"skillNo":99,"name":100,"nameZh":6,"description":101,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":102,"categories":103,"keywords":104,"publishedAt":19},"01M7EJQ3TFWMSW96T0E49W310A","web-preflight","Use when shipping, deploying, publishing, launching, reviewing, or pre-flighting a website or API and you want a lightweight check for accidental exposure of secrets, source or configuration files, admin and debug surfaces, unsafe redirects, missing security controls, and misleading 200 responses. Trigger phrases include \"before we ship\", \"security check\", \"preflight\", \"audit the site\", \"check the production URL\", \"is my .env exposed\", and \"why does \u002F.env return 200\", including when access-log or analytics evidence is supplied. Checks only sites the user owns or is authorized to assess, with bounded GET and HEAD requests and a fixed compact path set; distinguishes genuine resources from SPA catch-all and blanket WAF responses. Not for penetration testing, exploitation, credential testing, broad port scanning, or application-code review; use a dedicated authorized security-testing or code-review workflow instead.","https:\u002F\u002Fgithub.com\u002Farm-pz\u002Fcaffeinity-skills\u002Fblob\u002Fmain\u002Fplugins\u002Fcaffeinity\u002Fskills\u002Fweb-preflight\u002FSKILL.md",[],[]]