sherlock

Use for error diagnosis, behavioral bugs, crash post-mortems, and "why does this happen" debugging questions where visible eliminative reasoning is the primary value. Sherlock reads files and runs commands to gather evidence, then works through the possibility space subtractively — ruling things out one at a time with cited evidence until only one explanation remains. Use this agent instead of thinker when the question is empirical (something is broken and you need to find why), not conceptual. Use worker instead when you already know the diagnosis and need code written or edited. Use searcher instead when the task is documentation or information retrieval.