agentleFS
cacardinalcacardinal/laila/.claude/skills/os-audit/SKILL.md

os-audit

Skill1 starsChanged 4 months agoMITView on GitHub

Scored Four-Cs health audit of Laila. Trigger on "os audit", "audit the OS", "system health score", "four Cs", "how healthy is the OS". NOT the command channel's /audit (that's the autonomy audit log). Read-only except one report file.

2 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 4 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    NOT for: the command channel's /audit (autonomy action log), stale-content hygiene passes, or pending-item review

    line 15
  2. Very specific:

    Never print secrets — env-file checks are presence-only (grep -c).

    line 109
  3. Distinctive:

    Don't recommend tools/skills that don't exist in this repo.

    line 112
  4. Distinctive:

    Before declaring this skill done (and again after edits):

    line 116
  5. Very specific:

    Idempotency test: run twice same day — second run overwrites the same file cleanly (no -2.md suffix, no duplicate append) and computes Δ …

    line 118
  6. Distinctive:

    After each run: if Alex corrected a score or a detection misfired (e.g., penalized a bulk-migration status.md timestamp despite git …

    line 123
  7. Distinctive:

    If the same correction happens twice, the fix is MANDATORY before the next run.

    line 124

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

OS Audit

Score Laila 0-100 across the Four Cs (Context, Connections, Capabilities, Cadence — 25 each), rank the top 5 gaps by leverage, write one report artifact. Structural health only — "is the OS built and running right?", not "what could it do next."

When to use

  • "os audit", "audit the OS", "system health score", "four Cs check"
  • Monthly security/health audit companion; after big infra changes
  • NOT for: the command channel's /audit (autonomy action log), stale-content hygiene passes, or pending-item review

Steps

  1. Gather inputs (read-only, batch the reads). config/integrations.json (connections[]), config/domain-triggers.json (active domains), root AGENTS.md, MEMORY.md (line count only), docs/background-monitoring.md. Count skills via frontmatter-only scan of skills/*/SKILL.md — do NOT read skill bodies except where step 4 requires a grep. Count agents/*.md.
  2. Score each C per rubric below. Be honest, not generous — 90+ is a flex; a real setup lands 60-85. Don't penalize non-canonical file names if the intent is captured elsewhere.
  3. Rank gaps: leverage = points lost × impact multiplier. Output top 5.
  4. Write the artifact (only write): state/audits/YYYY-MM-DD-os-audit.md (mkdir -p state/audits first). Same-day rerun overwrites. If a previous audit file exists, append one line: Δ vs YYYY-MM-DD: {+/-n} total ({per-C deltas}).
  5. Print the report in chat. Target under 60 seconds wall-clock.

Rubric

Context (25)

CriterionPtsDetection
Root AGENTS.md is a router, not an archive3Pointers to domain files/configs, not inline duplication of domain content. Penalize -1 per section that duplicates what a domain AGENTS.md or config file owns; penalize if >~400 lines
Registry completeness2Three diffs, -1 each mismatch (floor 0): domains on disk vs config/domain-triggers.json entries; agents in agents/ vs AGENTS.md router tables; loops in state/loops-registry.json vs the scheduler's actual job definitions
Active-domain coverage5Every state: active domain in domain-triggers.json has domains/<name>/AGENTS.md + tracking/status.md. -1 per missing file, floor 0
Status freshness5-1 per active domain whose status.md "Last updated" >30 days old AND no git commits touching that domain dir in 30 days (git log --since="30 days ago" -- domains/<name>/). Status files may share a bulk-migration timestamp — git activity is the tiebreaker before penalizing. Floor 0
MEMORY.md is a hot cache5≤150 lines = 5; 151-250 = 3; >250 = 1. It's a cache, not an archive
Three-layer knowledge system live5knowledge/entities/ populated, state/daily-notes/ has note within 3 days, knowledge/tacit/ exists, nightly consolidation artifact within 7 days

Connections (25) — mechanism-agnostic, read config/integrations.json connections[]

Seven tier-1 domains: Comms, Calendar, Tasks, CRM, Finance, Meetings, Knowledge. "Reachable" = any entry in connections[] for that domain via ANY mechanism (MCP, script, GraphQL, osascript, scrape).

CriterionPtsDetection
Tier-1 coverage18~2.5 pts per reachable tier-1 domain (18/7 each, round total to nearest 0.5, cap 18)
Freshness(deduction)-0.5 from coverage per connection with last_verified >30 days old. Floor 0
Comms routes PROVEN2If a comms-route health loop exists, read its last-run artifact: overall: ok AND last_run <48h = 2; degraded or stale = 1; critical/missing = 0. A declared route is only worth points if a health loop proved it
Reference docs3Start at 3; -1 per connected tool with reference_doc: null. Floor 0
Write capability2≥1 connection has can_write: true. All read-only = viewer, not an OS = 0

Capabilities (25)

CriterionPtsDetection
Skills cover the top recurring workflows10Frontmatter-only count, then check coverage of: daily brief, comms check, session wrap, reminders/tasks, CRM, and the household's own recurring routines. ~1.5 pts per covered workflow, cap 10
Subagents defined4agents/*.md ≥2 = 4; 1 = 2; 0 = 0
Delegation actually wired4grep -rl "subagent_type|Task tool|dispatch" skills/ — ≥3 skills delegate = 4; 1-2 = 2; 0 = 0 (agents that exist but are never dispatched don't count)
Review/critic gates exist3Any adversarial/review pass wired into a skill (e.g. a blind red-team grader, a prose review gate) = 3
Template compliance of new skills4Sample the 5 most recently modified SKILL.md files: -1 per file missing Verification or Self-improvement sections (per templates/skill-template.md). Floor 0. Pre-template legacy skills: note, half-penalty

Cadence (25) — scheduled-job count is necessary, NOT sufficient. Score health.

CriterionPtsDetection
Jobs loaded vs documented6Scheduler's loaded jobs (e.g. `launchctl listgrep -i <your-prefix>orcrontab -l`) vs the Background Monitoring table. -1 per documented job not loaded. Floor 0
Jobs actually producing8Recency of each job's state artifact (state/*-last.json files): each within 2× its schedule = healthy. Check your CLI's usage/failure log (e.g. state/agent-usage.jsonl, if present) for repeated failures. -2 per scheduled job in known-broken state. Floor 0
Delivery pipeline5Daily brief has a delivered/sent marker for today or yesterday = 5; brief generated but not sent = 2; neither = 0. If an OAuth token expiry cycle is blocking sends, this is the 4x gap
Dead-man's-switch wired3Presence-only grep of the untracked env file for ping URLs (e.g. grep -c "^HC_") ≥1 — NEVER print values
Battle-tested-then-scheduled3No bike-method-phase: 1 skill referenced by a scheduled job/routine. Violation = 0

Gap ranking — leverage = points lost × multiplier

Gap classMultiplier
Broken delivery pipeline (brief not sending, OAuth expiry blown mid-cycle)4x
Tier-1 data domain unreachable3x
All connections read-only / no write path2x
Scheduled job silently broken (stale state artifact, failing runs)2x
Unregistered domain/agent/loop (registry drift)2x
Phase-1 skill on a schedule2x
No delegation wiring / new skill missing Verification+Self-improvement1.5x
Connected tool missing reference_doc1.5x
Stale status.md, stale last_verified, MEMORY.md bloat, everything else1x

Sort descending, take top 5, each with a one-line concrete fix pointing at files/skills that actually exist.

Report format (chat + artifact)

# OS Audit — {date}
**Score: {total}/100** ({stage})   Stages: 0-39 Foundation | 40-69 Built | 70-89 Compounding | 90-100 Autonomous

Context        {##bar}  {n}/25   Connections  {##bar}  {n}/25
Capabilities   {##bar}  {n}/25   Cadence      {##bar}  {n}/25
(bar = one # per 5 pts; label per C: Strong ≥20, Solid 15-19, Thin 8-14, Missing <8)

## Strengths
- {2-3 bullets from highest-scoring criteria}

## Top 5 Gaps (by leverage)
1. **{gap}** (-{pts} × {mult}) → {one-line fix}
...

Δ vs {previous audit date}: {+/-n} total ({C deltas})   [omit on first run]

Rules

  • READ-ONLY except state/audits/YYYY-MM-DD-os-audit.md. Never modify AGENTS.md, configs, skills, state files, or knowledge. Never trigger sends, syncs, or scheduled-job restarts.
  • Never print secrets — env-file checks are presence-only (grep -c).
  • Be honest, not generous. Show the math when a C loses points.
  • Frontmatter-only skill scans; don't read every skill body.
  • Don't recommend tools/skills that don't exist in this repo.

Verification (MANDATORY)

Before declaring this skill done (and again after edits):

  • Cold test: fresh session, say "os audit". It must produce the full scoreboard + write state/audits/{today}-os-audit.md without nudging, in roughly 60s.
  • Idempotency test: run twice same day — second run overwrites the same file cleanly (no -2.md suffix, no duplicate append) and computes Δ against the prior different-day audit only.
  • Spot-check one criterion per C against ground truth (e.g., count connections[] tier-1 domains by hand and compare).

Self-improvement (MANDATORY)

  • After each run: if Alex corrected a score or a detection misfired (e.g., penalized a bulk-migration status.md timestamp despite git activity), this skill is wrong, not Alex. Propose the rubric edit immediately, or log {"skill": "os-audit", "correction": "...", "date": "YYYY-MM-DD"} to state/skill-feedback.jsonl.
  • If the same correction happens twice, the fix is MANDATORY before the next run.

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.