Parallel Diagnosis

Two independent agents investigate a bug in parallel, then converge on a single unified diagnosis. Produces structured output that downstream skills (two-pass-review, fix-loop) can act on.

When to use

• YES: Multi-layer bugs (schema + router + model), framework quirks, uncertain library behavior, high-risk deploys, intermittent failures
• NO: Trivial bugs, typos, issues where root cause is already known, zero blast radius

Instructions

1 — Parallel diagnosis

If the problem statement is ambiguous (unclear which subsystem, multiple possible symptoms, or no file paths identified), use the AskUserQuestion tool to narrow scope before spinning up agents. Present what you understand and ask the user to confirm or clarify.