diagnosing-a-failure

Use to establish why something is actually failing before fixing it -- a general-purpose, in-session debugging procedure covering reproduction, boundary-scoped evidence collection, hypothesis testing, and a disconfirmation check, ending in one Diagnosis Verdict handed back to the caller. Never writes to GitHub, never authors a fix or a durable failing test, and never decides an architecture or design question on its own -- that stays the caller's job downstream of the Verdict. Distinct from eliciting-a-design (that skill elicits an agreed shape for something not yet built; this skill investigates why something already built is misbehaving -- no shared vocabulary, no runtime contact).