Workflow

The Replay Loop

Lode is organized around the shortest loop that proves its value:

install → demo → capture one session → query one decision

Each step maps to a natural moment in your workflow:

1. Install — add the plugin and verify the CLI if needed.
2. Demo — run node examples/decision-replay-demo.mjs to see the evidence-pack shape.
3. Wrap up one real session — 收工 or /lode:capture classifies the session and captures decisions, risks, abandoned alternatives, artifact changes, and source references.
4. Query one decision — run /lode:query when an agent needs cited evidence for why a project chose a path.
5. Compound later — use /lode:recall, /lode:roadmap, /lode:daily, /lode:weekly, or /lode:monthly after history exists.

Compounding Layers

Lode is not a strict pipeline. Skills are independently triggered, but they reuse each other's artifacts when available.

Capture → raw/weeks/{week}/{slug}.json    → raw entry + artifact index
Query   ← raw/decisions/ + raw/weeks/      → cited decision replay pack
Recall ← raw/weeks/ + raw/artifacts/      → session context
Roadmap ← raw/weeks/ + raw/decisions/      → decision narrative
Daily   ← raw/weeks/ JSON + git log       → Daily Note.md
Weekly  ← raw/weeks/ + git coverage       → weekly outline
Monthly ← Daily Note.md                   → monthly review + candidate rules

The key insight: raw entries from capture are report-worthy. They carry archetype-specific fields (decision rationale, repair root causes, investigation findings) so /lode:query, /lode:roadmap, and periodic reports can explain what happened without a second write pass.

Daily, weekly, and monthly outputs are compounding layers. They should improve as decision evidence accumulates, but they are not required for the first useful Lode experience.

Reporting Roll-up

Lode uses one reversible path for reporting: roll session signals upward into outcomes, then let every important claim drill back down to decisions and evidence.

Roll upward: raw entries → decisions/tradeoffs → work streams → outcomes/progress
Verify downward: outcomes/progress → supporting work → decisions/tradeoffs → raw/source/evidence refs

• capture records concrete changes, impact, status, work streams, and evidence references.
• weekly and monthly promote only claims that pass the outcome check; ongoing work, risks, exploration, and maintenance keep their honest status.
• query uses cited local records to explain why a claim exists, what was rejected, and which evidence or evidence gaps remain.
• Raw entries remain the source of truth. Reports are traceable reading views; they do not create facts retroactively.

Commit counts, changed lines, task counts, work days, and log volume measure activity or coverage. None proves an outcome by itself.

Storage Convention

Data lives in two layers within your knowledge vault:

• Raw layer (raw/) — immutable intermediate data: weekly entries, decision replay indexes, artifact indexes, signals, skeletons
• Wiki layer — human-readable outputs: Daily Note, weekly outline, monthly review, decision roadmap

Your knowledge vault is a git repo (typically an Obsidian vault), enabling cross-machine sync via git push/pull.

Zero-Config Mode

No vault? No problem. capture outputs structured Markdown directly in the conversation, so you get value from the first session without setup. Configure a vault when you want /lode:recall, /lode:query, and reports to reuse durable history.

When you're ready for persistent storage, run /lode:cold-start-interview to configure a vault path. Existing conversation-only sessions remain useful — they just don't compound into reports.