project-ContemPlace: Command for Claude Code

Audit captures

Analyze recent real-world captures against the product's design philosophy. Surface gaps between what the system promises and what it delivers, then recommend concrete improvements.

Arguments

$ARGUMENTS — number of recent fragments to analyze, or a tag filter like "tag:bookbinding" (fetches up to 15 matching notes). Default behavior: all fragments since the last audit date (up to 50). If no prior audit exists, default to 10. An explicit number overrides the smart default.

Workflow

Phase 1: Load everything in parallel

Maximize the first parallel batch. Launch all of these simultaneously — the ethos reads and the data fetches have no dependencies on each other:

Prior audit context (do this BEFORE the parallel batch — it determines the sample size): 0. Find the last audit date. MEMORY.md may be truncated past line 200, so do NOT rely on it being fully loaded. Instead:

• Use Glob to find all audit_*.md files in the memory directory
• Read the most recent one (sorted by filename date) to get: date, sample size, key findings, open recommendations
• This determines the smart default: fetch all fragments since that audit date (up to 50)
• If no audit files exist, this is the first audit — default to 10

Ethos reads:

1. Read docs/philosophy.md — the 10 core principles
2. Read docs/capture-agent.md — capture pipeline behavior, title model, linking logic, traceability rules
3. Read mcp/src/capture.ts — the live SYSTEM_FRAME (structural contract the LLM actually receives)

Live capture voice (authoritative source — the migration seed can drift): 4. Fetch from the DB via Supabase REST API:

export $(grep -E '^(SUPABASE_URL|SUPABASE_SERVICE_ROLE_KEY)=' .dev.vars | xargs) && \
curl -s "${SUPABASE_URL}/rest/v1/capture_profiles?name=eq.default&select=capture_voice" \
  -H "apikey: ${SUPABASE_SERVICE_ROLE_KEY}" \
  -H "Authorization: Bearer ${SUPABASE_SERVICE_ROLE_KEY}"

Fragment data: 5. list_recent with the smart default limit (all since last audit, up to 50) or the explicit argument to get the sample set. Then filter by created_at to include only fragments newer than the last audit date. If the argument is an explicit number, use that directly without date filtering.