What problem does it solve? Debugging complex, flaky, or recurring bugs often devolves into guess-and-check patching that masks root causes and introduces regressions. This Skill enforces a disciplined four-phase methodology — root cause investigation, pattern analysis, hypothesis testing, and verified implementation — so fixes are based on evidence rather than speculation. ## Core Features & Use Cases - Tight Feedback Loops: Builds fast, deterministic reproduction commands (failing tests, curl scripts, bisection harnesses, fuzz loops) that go red on the exact symptom before any fix is attempted. - Ranked Hypothesis Testing: Forms 3-5 falsifiable hypotheses, tests one variable at a time, and stops after repeated failures to question the underlying architecture. - Multi-Component Evidence Gathering: Instruments component boundaries, traces data flow upstream, and supports delegated sub-agent investigation for distributed systems. - Use Case: A pytest case fails intermittently and two previous fixes did not stick. The agent raises the reproduction rate, isolates the root cause to a specific code path, writes a regression test, and verifies the fix against the full suite. ## Quick Start Ask the agent to debug a failing test or production error by finding the root cause first, reproducing it with a minimal command, and only then proposing a verified fix.
