investigatelisted

@markdownai v1.0 # /investigate — Problem-Solving Framework Two modes: design new solutions (brainstorm) or fix existing problems (debug). ## Brainstorm Mode — Before any plan or implementation **Use when:** Designing a new feature, solving an open problem, or deciding between approaches. ### Steps 1. **Search existing KB** — `kb_search` on the feature/problem. Read relevant atoms before forming opinions. 2. **Search prior sessions** — `soil_search` on `hanuman/atoms` with relevant keywords. Check if this was approached before. 3. **State the problem** — one sentence. What are we solving? 4. **Generate 3 approaches** — for each: name it, state the core tradeoff in one sentence. 5. **Recommend one** — which approach and why, in 2 sentences. 6. **Flag constraints** — does this touch: - MCP tools → use `_mcp.call()` subprocess client, not direct import - Session state → use `_state.py` functions, not direct file writes - Hook events → each behavior needs its own `try/except` - `settings.json` → use `install.py`, not manual edits 7. **Stop** — do not implement until USER confirms the approach. ### Rules - KB search first. Never brainstorm in a vacuum. - Three approaches minimum. Two is lazy, four is stalling. - Constraints are hard gates, not suggestions. --- ## Debug Mode — Before coding a fix **Use when:** A bug is reported, tests fail, or unexpected behavior occurs. ### Steps 1. **Search prior context** — `soil_search` on `hanuman/atoms` for the error