Workflow

Primary Loop

install
  -> try: write weekly / write daily
  -> when you need multi-day memory: configure vault and project groups
  -> wrap up after key sessions
  -> write daily / weekly / monthly
  -> drill down only when questioned or resuming

1. Install tracework@tracework.
2. Try /tracework:weekly or /tracework:daily first; without a vault, the report returns in the conversation.
3. Run /tracework:cold-start-interview when you want durable storage and strict reporting groups.
4. End key sessions with 收工 / “wrap up” or /tracework:capture so reports can explain intent, risks, trade-offs, and evidence boundaries.
5. Use /tracework:daily, /tracework:weekly, and /tracework:monthly for high-frequency management closure.
6. Optionally enable metadata-only session scanning and run /tracework:capture day once at day end to recover missed sessions.
7. Use Query, Recall, or Roadmap only when work is questioned, resumed, or reviewed over a long horizon.

User Actions, Not a Command Table

What you say	What runs
wrap up / 收工	capture
write daily / weekly / monthly	daily / weekly / monthly
why did we choose this / continue last time	query / recall

See Skills for the full command surface.

Scope Partition

work     -> work projects only
personal -> personal projects only
all      -> separate group narratives in one private view

Partition happens before selection. Weekly then resolves goal sources and accounts for actual change, variance, and next commitments without forcing a headline count. Its brief body keeps only information that changes management judgment, action, or confidence; accountability, coverage, and provenance stay in the appendix.

Reuse Map

Capture -> raw/weeks/{week}/{slug}.json
Capture Day -> scoped session index + local transcripts -> raw entries
Daily   <- raw entries + limited git fallback
Weekly  <- raw entries + limited git fallback
Monthly <- raw entries + Daily/Weekly editorial context
Query   <- derived decisions + raw entries
Recall  <- raw entries + artifact/decision navigation
Roadmap <- decision evidence pack

Raw entries remain semantic truth. Report-local evidence ids belong to the appendix, not the raw schema or spoken narrative.