version: "1.0.0" name: audit description: Audit the OS in one composite report. Say "audit the OS", "check OS health", "full audit", or "is the OS healthy" (or run /founder-os:audit). Runs readiness, lint, wiki graph state, brain staleness, and voice completeness together. Read-only. why: "Gives you one honest number for OS health so you know what to fix instead of guessing which layer is causing bad output." enhance: "Run after a sprint roll so cadence files are fresh - stale daily-anchors and empty brain log are the two checks most likely to drag your score down." allowed-tools: ["Read", "Glob", "Grep", "Bash", "Agent"] mcp_requirements: []

Audit

Runs on: reasoning - reads your files and reasons; any capable agent can run this.

This skill returns one Founder OS health report. It composes five checks that can also run on their own.

Pre-Flight

If core/identity.md does not exist, stop with: Founder OS not set up here. Run /founder-os:setup first.

Checks

Run all five checks. Sequential by default; parallel only if explicitly available.

Parallel path (preferred when the Agent tool is exposed in this Claude Code build): dispatch five sub-agents in a single message, one per check, each with a focused prompt that names the check and the expected output shape. Then merge the five results into the composite block.

Sequential path (fallback - applies if the Agent tool is not available, if a sub-agent fails, or if you are unsure): read skills/<check>/SKILL.md for each of the five checks below and apply its logic inline against the live OS. Mark the report parallel: unavailable.

For checks 4 and 5 there is no separate skill file - apply the rules below inline regardless of path.

1. readiness-check - read skills/readiness-check/SKILL.md and apply. Returns score and next 3 moves.
2. lint - read skills/lint/SKILL.md and apply. Returns broken links, orphans, stale content, provenance gaps, possible contradictions.
3. wiki-build state - read brain/relations.yaml and report counts of curated edges and auto-generated wikilinks. Do NOT run wiki-build itself unless the user has approved the write at the gate above. If brain/relations.yaml does not exist, return wiki: not initialised - run /founder-os:wiki-build to create it.
4. Brain staleness - count entries past Decay after, flags older than Week 3, and rants in brain/rants/ older than 30 days where processed: false.
5. Voice completeness - check whether core/voice-profile.yml and core/brand-profile.yml exist and have been filled past their template defaults.

Wiki Write Gate

Audit is read-only by default. wiki-build writes to brain/relations.yaml, so ask before running it:

Audit can refresh the wiki graph before scoring. This writes only to brain/relations.yaml. Run wiki-build first? yes / no

If the user says no, read the existing graph and mark wiki freshness as not refreshed.

Composite Output

Render only the composite report. Do not paste raw sub-check outputs.

FOUNDER OS HEALTH - <YYYY-MM-DD>
---
Readiness: <score>/100 - <one-line headline>
Wiki: <links count>, <orphans count>, <broken refs count>
Brain: <decay-due count>, <Week 3+ flags>, <stale rants>
Voice: <state>
Parallel: available | unavailable
---
Top 3 actions to lift health:
1. <action>
2. <action>
3. <action>

Brain Staleness Rules

• Decay after: date in the past counts as decay-due.
• Any flag marked Week 3, Week 3+, or older than 21 days counts as Week 3+.
• Any file in brain/rants/ older than 30 days with processed: false counts as stale.

Voice Completeness Rules

Voice is incomplete if:

• core/voice-profile.yml is missing.
• It still contains placeholder brackets or template examples.
• It has fewer than 2 writing samples referenced.

Brand is incomplete if:

• core/brand-profile.yml is missing.
• Colors, fonts, logo, or document style are still blank placeholders.

Rules

• Read-only unless the user approves wiki-build.
• The composite report is the only output.

<!-- private-tag: not applicable: audit reads and reports; any brain/ reference is to files being read, not user-speech writes -->