Aios audit

What this skill does

Runs the Four Cs Audit on the current Claude Code project. Reads (never writes) the project's operating manual, memory, skills, agents, MCPs, decisions, and references. Scores each of the Four Cs out of 25. Surfaces strengths and the top 3 leverage-weighted gaps with concrete next-step commands.

Scope is structural — "is the AIOS built right?" It is NOT a capability planner. Capability gaps ("you could build a daily brief if you connected calendar") belong to /level-up. The audit answers: are the files, folders, registries, and connections in good shape?

First run is the baseline. Re-run weekly to watch the score climb. That's the compounding hook.

Today's context

• Date: !date +%Y-%m-%d
• Project root: the current working directory

The Four Cs (scored 25 each = 100 total)

Layer	Test
Context	Knows the user — identity, team, voice, decisions, references
Connections	Reaches the user's stuff — MCPs, integrations, data sources
Capabilities	Knows how to do work — skills + agents
Cadence	Runs without being asked — schedules, hooks, recurring rituals

Execution

Step 0: Run the deterministic scorer (RUN it, don't read it)

python3 .claude/skills/aios-audit/scripts/four-cs-score.py

The bundled script does all the mechanical detection and every piece of point arithmetic — counts, rounding, floors, caps, and leverage multipliers — so weekly scores are comparable, not re-derived by hand. The first pass emits the mechanical counts plus a needs_judgment list of flags only you can decide. Steps 1-2 below are your guide for judging those flags; then re-run:

python3 .claude/skills/aios-audit/scripts/four-cs-score.py \
  --tier1-domains N --identity-captured 0|1 \
  --connected-tools "tool1,tool2" --stale-connections N \
  --connections-doc 0-3 --write-path 0-2

Use the final JSON's scores, total, stage, and gap leverage verbatim — never recompute them. Your judgment goes into the flags, the strengths, and the concrete next-step lines.

Step 1: Discover the project shape (judgment guide)

The audit looks for patterns and intent, not exact paths. File names vary. The script already counted the mechanical parts; use Glob and Read only to judge the flagged criteria against these signals:

Operating manual: CLAUDE.md (root), CLAUDE.local.md (gitignored). Memory: MEMORY.md (root), ~/.claude/projects/<id>/memory/MEMORY.md, or memory/ folder. Skills: .claude/skills/*/SKILL.md — count + frontmatter. Agents: .claude/agents/*.md — count + frontmatter. Match by frontmatter: only count files with name/description frontmatter; skip README.md (a folder README is not an agent). Connection mechanisms (any of these = "reachable"):

• MCPs: .mcp.json, .claude/settings.json (mcpServers key), .claude/settings.local.json
• API scripts: scripts/*.py|.js|.ts documented in CLAUDE.md
• Export pipelines: data/, imports/, exports/ with refresh script + last-run timestamp
• API keys + reference guide: .env entries + corresponding references/{tool}-api.md

Connections registry: connections.md (anywhere). Reference guides: references/{tool}-api.md, references/*-reference.md, or equivalent. Decisions: decisions/log.md, decisions.md, or any append-only decisions file. References / SOPs: references/, docs/, sops/ folders. Templates: templates/, .claude/templates/. Hooks / scheduled jobs: .claude/settings.json hooks key, or skill names matching morning-*, weekly-*, daily-*, monthly-*, standup.

Don't penalize for non-canonical names if equivalent intent is captured elsewhere.

Step 2: Score each C (25 points each)

The tables below are the rubric the script implements — they document what is scored and how to judge the flagged criteria. Don't hand-compute any of the math; the script owns it.

Context (25 pts)

Criterion	Points	How to detect
Operating manual exists and is substantive (>200 words)	5	Read CLAUDE.md, count words
Identity / role / voice captured	5	CLAUDE.md mentions who the user is + role/mission, OR .claude/rules/*.md exists
Persistent memory exists with multiple entries	5	MEMORY.md exists with >3 entries, OR memory/ has >3 files
Reference docs exist	5	references/, docs/, or sops/ has ≥1 file
Decisions captured	5	decisions/log.md or equivalent has ≥1 entry

Connections (25 pts) — domain-aware, mechanism-agnostic

A "reachable" connection counts via ANY mechanism: MCP, script, export pipeline, or .env key + references/{tool}-api.md. The kit is API-first; the audit doesn't prefer MCPs.

The 7 Tier-1 Universal Data Domains:

#	Domain	Examples
1	Outcomes / Financials	Stripe, QuickBooks, a metrics dashboard, Looker
2	Customer / people interactions	HubSpot, Salesforce, Gmail-as-CRM, community DMs
3	Calendar	Google Cal, Outlook, Calendly
4	Communication	Gmail, Outlook, Slack, Teams, Discord
5	Project / task tracking	ClickUp, Asana, Linear, Notion DB, Jira
6	Meeting intelligence	Granola, Otter, Fireflies, Gong, Zoom
7	Knowledge / files	Notion, Drive, Dropbox, Confluence, SharePoint

Tier-2 (bonus): AI service API keys (OpenRouter, Anthropic, OpenAI), decisions/history, content/publishing.

Criterion	Points	How to detect
Tier-1 domain coverage	10	1.4 pts per tier-1 domain reachable. Round to nearest 0.5. Cap 10.
Reference guide presence	5	0 if no connections are reachable. Otherwise start at 5, -1 per connected tool with no references/{tool}-api.md. Floor 0.
Auth / pipeline freshness	5	0 if no connections are reachable. Otherwise start at 5, -1 per connection in needs-auth/expired state, or script with no run within 30 days. Floor 0.
Documentation in connections.md	3	0 if missing; 1 sparse; 2 most; 3 covers all reachable.
Intentional + safely-scoped write path	2	+1 if at least one connection has an INTENTIONAL, scoped write path a cadence actually needs (send email, post update); +1 if keys are scoped to least privilege (read-only, or write/admin scopes no broader than a cadence needs). Read-only is fine when nothing needs to write — don't blanket-penalize it, and don't reward over-broad write keys.

Capabilities (25 pts)

Criterion	Points	How to detect
3+ skills installed	10	Count .claude/skills/*/SKILL.md
1+ user-built skill	10	A skill in .claude/skills/ beyond the box-shipped set (agent-builder, agents-team-builder, aios-audit, grill-me, hooks-builder, level-up, multi-brain, onboard, plugin-builder, routines-builder, session-handoff, skill-builder, workflow-builder). A fresh clone scores 0 here — this point is for skills you add. (Update this set — and the script's BOX_SKILLS — if the kit ships more.)
1+ user-built agent	5	An agent in .claude/agents/*.md beyond the box-shipped scribe + warden. A fresh clone scores 0.

Cadence (25 pts)

Criterion	Points	How to detect
1+ recurring/scheduled trigger	10	.claude/settings.json hooks, OR skill name matches morning-* / daily-* / weekly-* / monthly-* / standup
Recent activity / usage signal	10	Files in .claude/skills/ modified within 30 days, OR decisions/log.md has entry within 30 days
Templates folder populated	5	templates/ or .claude/templates/ has ≥1 file

Step 3: Identify top 3 gaps by leverage

The script already emits gaps_by_leverage using this rule — leverage = (points lost) × (impact multiplier) — take its top 3 as ranked. One exception you may adjust: the write-path 2x fires only "where a cadence needs one"; if no cadence needs a write, demote that gap to 1x and re-rank. The multiplier table, for reference:

Impact multipliers:

• 0 tier-1 domains reachable: 4x (AIOS is blind to the work)
• Operating manual missing or thin: 3x (foundation)
• ≤2 tier-1 domains reachable: 3x (Connections is the gateway to live data)
• 0 skills: 2x (no Capabilities = no AIOS)
• No recurring trigger: 2x (no Cadence = no autonomy)
• No intentional write path where a cadence needs one: 2x (viewer, not an OS)
• 0 reference guides for connected tools: 1.5x (every future skill re-researches the same APIs)
• No decisions log: 1.5x
• All others: 1x

For each of the top 3, write a one-line concrete next step:

• Need a new skill? Recommend skill-creator (Anthropic) or skill-builder (local), or "write SKILL.md at .claude/skills/<name>/SKILL.md with YAML frontmatter."
• Need to log a decision? "Append to decisions/log.md."
• Need to reach a tier-1 domain? Prefer API+script (write scripts/{tool}_api.py + save references/{tool}-api.md). Recommend claude mcp add only if no API path exists.
• Connected tool missing a reference guide? "Research the API once, save endpoints + auth + common queries to references/{tool}-api.md."
• Need a recurring trigger? "Add a hook to .claude/settings.json, or write a skill named daily-* you run each morning." For how to engineer a durable cadence/loop (triggers, brakes, verification), see references/agent-loops.md.

Step 4: Output the report

Print directly in chat (Markdown). Format:

# AIOS Audit — {date}
**Score: {total}/100** ({stage})

Stage thresholds:
- 0-39 → Stage 0: Foundation
- 40-69 → Stage 1: Built
- 70-89 → Stage 2: Compounding
- 90-100 → Stage 3: Autonomous

## Scoreboard

Context        {bar}  {n}/25  {label}
Connections    {bar}  {n}/25  {label}
Capabilities   {bar}  {n}/25  {label}
Cadence        {bar}  {n}/25  {label}

(bar = ## per 5pts, rounding each C score to the nearest whole point before rendering; label = "Strong" ≥20, "Solid" 15-19.99, "Thin" 8-14.99, "Missing" <8 — bands are continuous, so fractional Connections scores always land in exactly one band)

## Strengths
- {1-3 short bullets from highest-scoring criteria}

## Top 3 Gaps (ranked by leverage)
1. **{gap name}** (-{points} × {multiplier})
   → {concrete next-step}
2. **{gap name}** (-{points} × {multiplier})
   → {concrete next-step}
3. **{gap name}** (-{points} × {multiplier})
   → {concrete next-step}

## Suggested next: {single most leveraged action}

---
Structural gaps only. To explore CAPABILITY gaps (what your AIOS could DO that it can't yet), run /level-up after this audit.

Step 5: Offer to save the report

After printing, ask: "Save this audit to audits/audit-{date}.md so you can track score over time?" If yes, write it (creating audits/ folder if needed). This is the only writable side effect.

Notes

• Read-only by default. Never modify CLAUDE.md, memory, skills, or any project files. Only optional write is the audit report.
• Be flexible about file names. Don't penalize for using non-canonical names if intent is captured.
• Be honest, not generous. A 95/100 is a flex. Most setups land 40-70.
• Don't suggest skills that don't exist. Point at what's actually available.
• Speed matters. Report in under 60 seconds wall-clock. The script does the counting; read only the targeted files the judged flags need.
• Cadence detection is fuzzy. Infer from skill names if hooks/cron data isn't cleanly available.