compoundlisted

# Compound — Knowledge Compounding Transforms session learnings into `docs/solutions/` — persistent documentation that future agents and developers can discover and reuse. **Announce at start:** "Compounding knowledge from this session..." ## When to Use Run `/compound` after any session where: - A bug was solved with a non-trivial root cause - A non-obvious pattern or API behavior was discovered - An architectural decision was made with considered alternatives - An approach was tried and abandoned (the dead end is worth recording) - A `Harness-Delta: backlog` signal was raised by a subagent Do NOT run after every session. Only compound when something is genuinely worth preserving for future sessions. ## Workflow ### Step 1: Launch 4 parallel research subagents Dispatch all four concurrently using the Agent tool. Provide each with the session context. They read the session transcript and git diff. They return **text only** — no file writes. If a spec was active this session, pass its slug to the Decision Extractor — it also harvests the `### Alternatives considered` section of `specs/<slug>/SUMMARY.md` (rejected alternatives with reusable reasons are decisions worth compounding). Subagent prompt files are in `.claude/skills/compound/subagents/`. The Related Docs Finder needs `module` and `tags` from the Context Analyzer to assess overlap accurately. Choose one of these two approaches: **Option A — 3+1 sequential (recommended for accuracy):** 1. Dispatch Context An