Skills

Lode ships eight skills. The primary path is capture -> query: preserve one session's decision evidence, then replay one decision with citations. The other skills compound that record into session-start context, roadmap narrative, and reports.

Cold Start Interview

Trigger: /lode:cold-start-interview

First-run setup. Creates ~/.lode/config.yaml with:

• Knowledge vault path (where Lode stores its data)
• Project identity and slug
• Report language (English, Chinese, or mixed)
• Weekly report mode and team context

Run this once. After that, all other skills read config automatically.

Query

Trigger: /lode:query, why did we choose this?, 为什么当时这么选

The primary value path: targeted decision replay. Reads {vault}/raw/decisions/ first, then falls back to raw weekly entries to answer specific project-history questions:

• Why a path was chosen
• What alternatives were rejected or deferred
• Which open questions should be revisited
• What impact a decision had

The skill returns compact nodes with raw provenance (source_entry_refs) and direct evidence links (evidence_refs or typed source_refs) kept distinct. Provenance says where a claim was recorded; it does not by itself verify that claim. Missing evidence is reported explicitly, and attached references remain visible for inspection; Lode refuses to answer when it has no supporting record.

Capture

Trigger: /lode:capture, /lode:capture checkpoint, 收工, done, 今天到这, checkpoint

Adaptive-depth session recap and checkpoint capture. Classifies your session as one of five archetypes:

• Decision — design choices, tradeoffs, rejected paths
• Build — features implemented, contracts changed
• Investigation — explorations, findings, dead ends
• Repair — bugs fixed, root causes, mitigations
• Maintenance — refactors, upgrades, cleanup

Writes report-worthy signals (work_stream, state change, status, impact, decisions, risks, contracts, and evidence) rather than process logs. A Fruit Check prevents commits, line counts, task counts, or vague completion claims from becoming outcomes without an observable state change, impact, honest status, and source path. Capture also generates lightweight sync suggestions for architecture docs, plans, and README files.

Zero-config mode: outputs structured Markdown directly in the conversation without writing to any vault.

Vault mode writes quietly by default. Use checkpoint capture during long work to save decisions, stage progress, risks, or next-step entry points without filling the main conversation with a full recap.

Recall

Trigger: /lode:recall, 开工, session start, 继续上次

Session-start context recall for projects with history. Reads recent raw entries, decision indexes, and artifact indexes from your knowledge vault to surface:

• Recent decisions and their rationale
• Open risks and unresolved questions
• Abandoned alternatives worth remembering
• Stale intent artifacts that may need attention

Works without a vault — falls back to conversation-only context.

Daily

Trigger: /lode:daily, 更新日报, 日报, daily note

Updates Obsidian daily notes from raw entries and git history. Aggregates what happened across sessions into a single day view.

Weekly

Trigger: /lode:weekly, 周报, weekly PPT

Builds a weekly outline around at most three headline outcomes or progress claims. Each report uses a local 3+1 chain: O# outcome/progress, W# work stream, D# decision/tradeoff, and E# evidence audit. Raw entries remain the primary semantic source; git logs are fallback and coverage evidence only. Fallback-only work stays limited progress, never verified delivery.

Monthly

Trigger: /lode:monthly, 月度回顾, 月报, monthly review

Generates an outcomes-first monthly review from daily notes, with optional raw evidence enrichment and the same report-local O#/W#/D#/E# traceability used by weekly reports. Task counts, active days, category totals, and code-size figures stay in a coverage appendix and are never promoted into outcomes. Risks, evidence gaps, next-month handoff, and candidate rules remain visible. Python scripts handle parsing and aggregation for determinism.

Roadmap

Trigger: /lode:roadmap, 决策路线图, decision roadmap

Generates a narrative decision roadmap from accumulated raw entries. Tracks:

• Decisions made and their context
• Accumulating risks that haven't been resolved
• Recurring open questions
• How thinking evolved over time
• Confidence, evidence references, and evidence gaps for each decision point

The roadmap is not a feature list — it tells the story of what happened and why.