Skills

Full command reference. For first use, prefer Quick Start: try a weekly or daily report before configuring a vault.

Tracework is a local reporting plugin for Codex and Claude Code with five skills: capture keeps key facts; daily, weekly, and monthly produce progress reports; cold-start-interview configures storage and project groups.

Reports select projects before reading effective records, keeping work and personal groups separate. Monthly reviews use raw records and current risk states; Daily and Weekly reports provide prior judgments, so a complete Daily archive is optional. If some projects cannot be read, Tracework returns a conversation draft with coverage gaps and saves it only on explicit request.

High-Frequency Reporting

Daily

/tracework:daily [work|personal|all] writes one daily judgment per reporting group, necessary progress, and the next gate in short prose. One supported change is enough. Protected per-date/per-group updates preserve user edits; a conflict returns a draft instead of overwriting the file.

Weekly

Three modes:

• quick: 这周做了啥 / quick weekly — 1–7 conversation bullets plus carried-forward items; no file write.
• brief (default): /tracework:weekly / write the weekly report — objective-anchored management brief covering actual change, variance, and next commitments. Its body is decision-complete; its appendix preserves accountability, work coverage, and provenance.
• slides: explicit weekly PPT / slides wording — a standalone PPT-ready Markdown Deck with only necessary main-deck pages; unusual length triggers compression review, and zero admitted pages return an empty state.

Without explicit framing, Slides serves same-department colleagues at a weekly meeting, assumes they know the project background but not this week's changes, and ends with next-week closure. It starts from why the work exists, selects the weekly result or final choice, reopens sources only for selected claims, and uses cognitive roles only when a complex merge/split decision needs them. Public output contains claim-led slide content and a compact Evidence Appendix; it contains no production guidelines or page-level source packets.

Design rationale and mechanism split only when they use different grounded content, serve the same Deck Thesis, and the mechanism depends on the design premise. Complementary proof objects may share a page when each has a clear evidence responsibility for one claim.

Monthly

/tracework:monthly [work|personal|all] uses raw entries as semantic truth and Daily/Weekly as prior human judgments. It produces phase arcs, recurring risks, and next-month closure targets.

Reports can use current visible conversation facts when project, group and work date are known. They deduplicate these with raw/git and preserve conflicts. This does not capture facts into long-term memory or scan other conversations.

Stored factual mistakes can be corrected through Capture. Daily, Weekly, and Monthly then use the corrected fact in its original work period; explicit as-of preserves the knowledge boundary. Accepted risks and unresolved conflicts stay visible. Existing reports change only when you request a refresh. Targeted recovery reads only the project/date you select and leaves temporary reads unsaved.

Evidence Foundation

Capture

/tracework:capture, checkpoint wording, or 收工 stores adaptive-depth raw facts. It does not pre-write separate Daily, Weekly, and Monthly prose.

/tracework:capture day [YYYY-MM-DD] [scope] incrementally recovers durable facts from opt-in local session manifests. Scope partition happens before transcript reading, and transcript bodies are never copied into the vault.

Setup

Cold Start

Configures the vault, project identity, and reporting group with the minimum required questions. Metadata-only session scanning remains an explicit opt-in.