grok

Use this skill when the user asks to deeply understand why a system behaves a certain way, trace an end-to-end flow, explain a confusing interaction, or prove which code path is responsible.

The pattern is two-agent adversarial investigation:

• Investigator: exhaustive evidence gathering and hypothesis testing.
• Reviewer: skeptical validation using 5 Whys until the answer directly addresses the user's question.

Entry Gate

1. Restate the exact question being answered.
2. Define what would count as a sufficient answer.
3. Identify likely evidence classes: source code, docs, tests, logs, runtime behavior, user-visible UI, prior memory, or external system state.
4. Spawn exactly two subagents when the tool surface allows it:
   • an investigator agent
   • a reviewer agent
5. Keep the immediate coordination and final answer local. Do not outsource final judgment.