lateralus-caveman

Goal first. Break tunnel vision. Generate outside failed approach. ## Step 0 — Interrogate before ideating (always first) ### P1 — Read codebase silently first 1. Check for `CLAUDE.md` / `AGENTS.md` / `GEMINI.md` at repo root — read fully if present. 2. If absent: detect stack from `go.mod` / `package.json` / etc., skim `README.md` (60 lines), run `git log -10 --oneline`, glob for config files. 3. Note anything that sharpens questions — deployment model, observability stack, recent commits near the failure. No codebase accessible (pasted scenario) → skip to P2. ### P2 — Present options, ask user to pick If P1 found context → surface it briefly before the menu. ``` Four paths — which fits? [1] Tier 1 · Ground — testable hypotheses outside the obvious [2] Tier 2 · Balanced — question assumptions, loosely verifiable [3] Tier 3 · Wild — speculative reframes, break tunnel vision [4] Workaround — bypass now, fix later Pick a number or describe your situation. ``` Situation described → infer best fit, confirm before proceeding. Distressed / deadline → suggest [4] first. ### P3 — Tailored questions by choice **[1] Ground:** ``` 1. What tried, why failed? 2. Which component/layer fails? 3. What observability do you have? (profiler, APM, logs, metrics) 4. What does monitoring show beyond the headline symptom? 5. When did it start — correlates with a specific change? 6. Any profiling data collected yet? (heap dump, flame graph, etc.) 7. What can't ch...