You're not avoiding the task because you need smaller steps. Autopsy runs a real differential diagnosis on why you're stuck — then prescribes the fix for your actual cause, not a generic one.
Autopsy doesn't pattern-match your task to a tip. It holds several possible causes at once, asks the one question that best separates them, and rules causes out one at a time — the same logic a diagnostician uses on symptoms.
Every diagnosis Autopsy makes comes from this fixed set — not an open-ended guess — so the reasoning stays sharp instead of drifting into vague reassurance.
This runs the real reasoning flow client-side for the demo. Swap the response function for your LangGraph endpoint when you wire the backend.