Tracework

Turn agent sessions into evidence-backed work memory.

Documentation · Showcase · 中文

Tracework gives agent work a trace that can be reported, questioned, and carried forward.

It can write daily, weekly, and monthly work reports from the evidence already available, then improve those reports and future decision queries when sessions are captured into local raw records.

What Tracework Does

Tracework can:

• Write workplace daily reports and weekly/monthly reviews from raw entries, or from git as limited fallback coverage
• Capture session decisions, rejected paths, risks, artifacts, and next steps
• Answer why a path was chosen when local evidence supports it
• Recall useful context before a later session
• Keep Markdown and JSON records in your own knowledge vault

The core loop stays small:

report or query current work -> capture key session signals -> improve later reports and decisions

Decision replay is the trust mechanism: a later agent or reader can drill from a claim back to raw entries, rejected alternatives, risks, and source refs. If the record does not support an answer, Tracework should say so instead of inventing history.

Tracework is not a meeting-notes tool, approval workflow, performance packaging tool, employee-monitoring surface, or generic office suite. Activity counts are coverage metadata, not proof of outcomes.

Skills

Command	When	Output
/tracework:cold-start-interview	First run	Configures the local vault and project profile
/tracework:capture	Wrap-up or checkpoint	Dynamically captures lite, standard, or deep session memory
/tracework:recall	Session start	Recalls recent decisions, risks, open questions, and relevant artifacts
/tracework:query	Targeted follow-up	Answers "why did we choose this?" with cited local evidence
/tracework:weekly	Weekly	Rolls raw session records into a brief-ready outline
/tracework:monthly	Monthly	Builds a monthly review from workplace daily reports and raw evidence
/tracework:roadmap	Phase review	Synthesizes a narrative decision history
/tracework:daily	Daily	Writes workplace daily reports from raw entries, using git only for coverage gaps

The direct-value loop is intentionally small:

/tracework:cold-start-interview once
/tracework:daily, /tracework:weekly, or /tracework:monthly when you need a report
/tracework:query when someone needs the why
收工 or /tracework:capture after key work to improve future evidence
开工 or /tracework:recall when durable memory exists

Daily and weekly can still produce git-only limited reports when raw entries are missing. Capture is the quality multiplier: it preserves the decision, risk, artifact, and evidence boundaries that git cannot explain.

Install

Codex

codex plugin marketplace add KKenny0/Tracework
codex plugin add tracework@tracework

Update:

codex plugin marketplace upgrade tracework
codex plugin add tracework@tracework

Claude Code

claude plugin marketplace add KKenny0/Tracework
claude plugin install tracework@tracework

Update:

claude plugin marketplace update tracework
claude plugin update tracework@tracework

Storage

• Config: ~/.tracework/config.yaml or {project}/.tracework/config.yaml
• Raw entries: {vault}/raw/weeks/{week}/{slug}.json
• Decision indexes: {vault}/raw/decisions/{slug}.json
• Human-readable outputs: {vault}/Daily Note.md and {vault}/Work Diary/

The public product, command, config, and schema namespace is tracework. Tracework does not use legacy storage fallbacks; configure knowledge_vault in one of the config files above.

Development

npm --prefix cli run build
npm --prefix cli run copy-skills
npm --prefix cli run check-skills
npm --prefix cli run test
npm --prefix site run build

After editing .codex-plugin/, skills/, or assets/, run:

npm --prefix cli run copy-skills
npm --prefix cli run check-skills

Before releasing a user-visible plugin update:

1. Bump the plugin version in .codex-plugin/plugin.json, .claude-plugin/plugin.json, and .claude-plugin/marketplace.json.
2. Run npm --prefix cli run copy-skills so plugins/tracework mirrors the source manifests, skills, and assets.
3. Run npm --prefix cli run check-skills and npm --prefix cli run test.
4. Run claude plugin validate .claude-plugin/plugin.json and claude plugin validate .claude-plugin/marketplace.json.
5. Run claude plugin tag --dry-run . from a clean worktree before creating the release tag.

Core docs:

• Configuration
• Data model
• Artifact governance

License

MIT