メニュー
How to propose and query distilled lessons (the experiential memory). Only project-steward persists; every other agent proposes/queries. Use when you discover a durable lesson or need prior experience for the current task.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
Shared write-side contract every domain AUTHOR imports — how to implement reliably, the smallest-diff rule, validation gates, and the never-approve-your-own-work rule. The write-side mirror of token-efficiency. Load at the start of any implementation task.
Shared context-discipline contract every agent imports — context-source priority, read budget, confidence-stop, the severity rubric, and terse output. Load at the start of any review or implementation task.
General repo rules enforced by ESLint/TypeScript/CI — PRs-only, never bypass lint, import ordering, type imports, path privacy, commit format. Load for any code change.
Browser-extension + bridge standards — MV3 (Chrome + Firefox), least-privilege permissions, the loopback pairing-token auth model, the shared TS↔Rust wire protocol, and the Chrome Web Store + Firefox AMO store-policy + pre-submission checklist. Load for changes under apps/extension/**, extension_bridge/, and the extension protocol.
React renderer standards — ports & adapters (service hooks), design tokens, motion tokens, @ajh/ui primitives, i18n, feature isolation, React Query, a11y. Load for changes under apps/desktop/src/renderer/**.
Rust/Tauri backend standards — L0–L3 layers, centralized platform/net/error ownership, Rust-first business logic, registries, error handling. Load for changes under apps/desktop/src-tauri/src/**.
| name | lessons |
| description | How to propose and query distilled lessons (the experiential memory). Only project-steward persists; every other agent proposes/queries. Use when you discover a durable lesson or need prior experience for the current task. |
Store: .claude/memory/lessons.jsonl (local, capped 200, archived on overflow). Only project-steward writes.
node .claude/hooks/lessons.mjs query --domain export
node .claude/hooks/lessons.mjs query --category Performance
node .claude/hooks/lessons.mjs query --text "pagination"
Returns only matching ≤5-line lessons — never bulk-loaded. The review-gate auto-queries by touched domain/globs and folds matches into its prompt.
Surface as: LESSON · <category> · Context: … · Decision: … · Outcome: … (≤5 lines). project-steward dedupes & persists.
Tag the memory type in --tags — a free-form convention (not enforced by lessons.mjs), following the 2026 agent-memory taxonomy: episodic (a specific past event/fix), semantic (a durable fact/convention), or procedural (a workflow / review convention / test command / tool-use habit). Procedural lessons are the highest-leverage — capture "how we do X here."
Architecture decision · Failed approach · Proven approach · Performance · Security · ATS · Scraping · AI-provider · Export · Testing discovery.
Context · Decision · Outcome, ≤5 lines.docs/knowledge/decision-records/) is removed from the log — the ADR becomes its single source.