agentleFS
emiljuchnikowskiemiljuchnikowski/ai-os-template/.claude/skills/audit/SKILL.md

audit

Skill0 starsChanged 3 months agoMITView on GitHub

Weekly 4C gap report per domain + OS truth checks (read-only, scored). Use weekly (e.g. Friday) or when the owner asks "how is my AI OS doing".

193 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use delight, teach-impeccable and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • HK-hubSkill

    delight

    Add moments of joy, personality, and unexpected touches that make interfaces memorable and enjoyable to use. Elevates functional to delightful. Use when the user asks to add polish, personality, animations, micro-interactions, delight, or make an interface feel fun or memorable.

    Also in 31 projects that use this

  • HK-hubSkill

    teach-impeccable

    One-time setup that gathers design context for your project and saves it to your AI config file. Run once to establish persistent design guidelines.

    Also in 18 projects that use this

  • zwy123zwySkill

    normalize

    将用户随意输入的提示词规范化为结构清晰、便于大模型执行的标准化提示词。适用于需要整理、统一提示词格式时调用。

    Also in 18 projects that use this

  • ww-w-aiSkill

    pm-discovery

    PM Agent Team - Automated product discovery, strategy, and PRD generation. Runs PM analysis workflow to produce comprehensive PRD before PDCA Plan phase.

    Also in 3 projects that use this

  • ww-w-aiSkill

    qa-phase

    Run QA tests at various levels (L1-L5) for features under PDCA management. Integrates with MCP tools to execute structured test plans.

    Also in 3 projects that use this

  • +7 more from HK-hub

The file

Files
SKILL.md

Audit — 4C gaps + truth checks

Read-only: change NO files except writing the report. Delegate analysis to the os-auditor agent (it has no write tools by design). Motto: "does my OS still tell the truth?" — indexes and steering files are claims about reality; the audit checks every claim against the actual state.

Part 1 — 4C scoring

For each domain (per the list in CLAUDE.md), score each layer 0–5:

  1. Context — would a fresh session answer the domain's control questions without searching?
  2. Connections — is the domain's data reachable live (calendar, tasks, tools) without pasting?
  3. Capabilities — is there a skill that turns a short command into an artifact for this domain?
  4. Cadence — does anything happen without being asked? (0 is fine for the first weeks — cadence is deliberately last)

Part 2 — truth checks

  1. Routing integrity — every path, skill, agent and connection named in CLAUDE.md, connections.md and skills exists on disk; and the reverse: files in context/ and .claude/ that nothing routes to (orphans — unfindable without searching).
  2. Index truth — declared vs actual: the skill list in CLAUDE.md vs .claude/skills/, agents vs .claude/agents/, domains vs context/ folders, connections.md vs connections that actually work. If you build reference skills (with a chapters/ dir), also check their wiring: every declared consumer skill must actually reference them — an unreferenced reference skill is invisible at runtime.
  3. Freshness — classify each source and key context file: 🟢 fresh (updated on rhythm) / 🟡 drifting (rhythm slipping) / 🔴 frozen (was supposed to live, stands still) / ⚫ retired (should move to archives/) / 🔵 on-demand (no change is fine).
  4. Four context-failure modes — hunt for concrete instances: poisoning (a false/stale fact stated as true), bloat (a file so big the signal drowns — segmentation candidate), confusion (a missing fact an agent would guess), clash (contradicting facts between files — name the newer source, mark the older for archiving).
  5. Expertise vs situational — expertise context (who, goals, rules) belongs in context/ files; situational context (states, snapshots, live data) should be fetched just-in-time via connections. Flag situational data rotting in context/ and expertise you have to fetch from external systems.

Report → archives/audits/YYYY-MM-DD-audit.md

  • Table: domain × 4C with scores + total and delta vs the previous audit (start by reading the previous report in archives/audits/).
  • Truth-check table: routing / index / freshness / failure modes / placement — 🟢🟡🔴 + one-line reason each.
  • "What would get a wrong answer today?" — 2–3 concrete questions the owner could ask right now and get a wrong answer, with the cause (the shortest proof a gap is real).
  • Fix list awaiting approval (bike method — the audit changes nothing), grouped: (a) quick file fixes, (b) routing/index, (c) data catch-up, (d) cadence candidates — section (d) feeds /level-up.
  • Rhythm check: for each OS rhythm (brief, audit, level-up, ...) apply "never miss twice" — one miss is noise, two in a row means the rhythm is collapsing and needs action, not a note. Check it on data you already have (commits, report dates); don't build a new tracker.
  • Top 3 gaps, each with one concrete corrective action (one of them = the /level-up candidate).
  • Commit: docs: weekly 4C audit.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.