learning-patternslisted

Extract durable learnings from a session and propose project customizations — agent-instructions file, CONTEXT.md, ADRs, project skills, hooks. Use when the user says "learn", "extract learnings", "what did we learn", "save learnings", "adapt config", "capture domain language", or wants to encode session patterns durably. NOT for documentation edits (use documenting-code) or committing changes (use committing-code).
alexei-led/cc-thingz · ★ 33 · Code & Development · score 81

Install: claude install-skill alexei-led/cc-thingz

# Learn from Session Extract actionable, durable learnings and propose project-specific customizations. Ground every change in actual conversation or tool evidence. Ask one question at a time when confirmation is needed. ## Routing - Use only for durable project learning: reusable instructions, domain language, decisions, commands, skills, hooks. - Documentation edits route to `documenting-code`. - Do not fabricate. If evidence is missing, say extraction is blocked and ask for it. - Secrets, credentials, local paths, transient debugging details, and one-off failures are non-durable unless the user explicitly says otherwise. - Always name the exact target artifact before proposing persistence. ## Workflow If the user passed a topic, scope extraction to that topic. If the user passed `--dry-run`, run Phases 1–7 only. ## Phase 1: Discover Read existing customization files. Default portable paths (adapt to the platform's conventional locations): - `AGENTS.md` — agent instructions / project memory - `.agents/skills/*/SKILL.md` — project skills - `.agents/agents/*.md` — project subagents - `.agents/commands/*.md` — project commands - `CONTEXT.md`, `CONTEXT-MAP.md` — domain language - `docs/adr/*.md` — durable decisions - `.out-of-scope/*.md` — rejected scope with reasoning Record counts for the budget check. ## Phase 2: Extract Scan the conversation for these signal types. Instruction signals → agent-instructions file: - Corrections: "no", "wrong", "actually", "instead