What problem does it solve? Debugging sessions often stall on guesswork, missed silent failures, and leftover instrumentation that pollutes the repository. This Skill enforces a disciplined, evidence-driven loop that finds root causes from observed runtime state instead of plausible stories, and guarantees a clean working tree when done. ## Core Features & Use Cases - Hypothesis-driven investigation: Form at least three orthogonal hypotheses, investigate them in parallel, and escalate to hard-reasoning only after two failed evidence rounds. - Runtime and tooling references: Dedicated guides for Python, Node.js, Rust, Go, native binaries, and bundled-JS binaries (Bun SEA, Electron, PyInstaller), plus specialist tools like Playwright, Ghidra, pwndbg, pwntools, and a DAP client. - Flaky-test triage: A three-run signature protocol classifies intermittent, order-dependent, and CI-only failures before any fix is attempted. - Use Case: A developer sees HTTP 200 responses with empty bodies in production. The Skill walks them through environment assessment, parallel hypothesis testing, root-cause confirmation via toggle proof, a failing-first test, manual QA with real requests, and artifact cleanup. ## Quick Start Ask the agent to debug why your application crashes or returns empty responses, and it will load the matching runtime reference and start the hypothesis-driven investigation loop.