Workflow

Minimal Loop

install -> configure -> generate a report or query -> capture key sessions

1. Install the plugin as tracework@tracework.
2. Run /tracework:cold-start-interview.
3. Run /tracework:daily, /tracework:weekly, or a concrete /tracework:query.
4. End key sessions with 收工 or /tracework:capture so future reports and decision queries have stronger evidence.

This proves the core value quickly: reports can be useful immediately, while captured sessions preserve the why that git history cannot recover.

Reuse Map

Capture -> raw/weeks/{week}/{slug}.json
Query   <- raw/decisions/ + raw/weeks/
Recall  <- raw/weeks/ + raw/artifacts/ + raw/decisions/
Daily   <- raw/weeks/ + fallback git coverage
Weekly  <- raw/weeks/ + fallback git coverage
Monthly <- Daily Note.md + matching raw entries
Roadmap <- raw entries + decision indexes

Skills are independent. There is no required pipeline. The shared storage convention lets later views reuse earlier evidence.

Daily and weekly can run with no raw entries by using git fallback coverage. Those outputs must stay limited: they can describe activity and progress, but they must not invent motivation, trade-offs, risks, or verified impact.

Progressive Closure

Briefs and reviews should roll work upward while preserving a route back down:

upward: raw entries -> decisions -> work streams -> outcomes / risks / next steps
downward: outcome claim -> stream -> decision -> raw entry / evidence ref

Weekly and monthly outputs can use report-local labels:

• O#: outcome, progress, risk, or decision
• W#: supporting work stream
• D#: decision or tradeoff
• E#: evidence audit

These labels make a report inspectable. They do not change the raw schema.

Zero-Config

Without a vault, capture can still return a structured recap in the conversation. Configure a vault when you want quiet writes, cross-session recall, decision query, and higher-quality brief/review outputs.

Evidence Rules

• A source ref is provenance, not proof by itself.
• Git-only work is fallback evidence.
• Activity counts do not prove outcomes.
• Unsupported queries should return an evidence gap.
• New facts should be appended as new raw entries, not written into old records.