apw-find-root-causelisted

Use this skill when something is observably broken — a defect, a failure, a wrong result — and you need to find and *prove* its root cause before a fix is attempted. It exists to fight one specific failure: **patching the symptom, or committing to the first plausible cause without proving it**. Under pressure to resolve the issue, a model latches onto the earliest explanation that fits, then "fixes" one level above the real defect — so the symptom returns, or a workaround hides a problem that was never understood. This skill replaces the first plausible story with a proven causal chain: capture the symptom, reproduce it, hold *competing* hypotheses, and converge on the cause through an iterative, evidence-driven loop with the user. This skill is **diagnostic-only**. It ends at a proven (or best-supported) root cause and explicitly does **not** design or apply the fix — that is handed off to `apw-implement` (optionally via `apw-plan`). It is distinct from `apw-research`: research is open-ended investigation of an unknown space, while this skill diagnoses a *known, observed* defect. It sits **upstream** of the fix — it is how you prove what is wrong before anyone changes code. Follow the steps below strictly and in order. ## Step 1 — Capture the symptom Before forming any theory, pin down exactly what is broken — and separate the symptom from its cause: - **What happens vs what's expected** — the precise observed behavior and the behavior it should have. - **Where and when