What problem does it solve? Teams often declare work done based on weak or mismatched evidence, such as treating a successful compile as proof of runtime behavior or calling a build production-ready after unit tests pass. This Skill enforces claim/evidence alignment so every acceptance statement is backed by the correct class of evidence and nothing is silently upgraded. ## Core Features & Use Cases - Claim-to-Evidence Matching: Restates each claim as a testable statement, identifies the required evidence class (compile, unit, runtime, visual, security, signing, human-observed), and inventories what evidence actually exists. - Bounded Status Reporting: Returns explicit PASS, PARTIAL, FAIL, BLOCKED, or UNVERIFIED results with what is proven, what is not proven, and any conflicting evidence preserved. - Multi-Gate Decomposition: Splits compound claims like "production-ready" into separate gates (compile, tests, runtime, installer, recovery, security, signing, owner acceptance) so one passing test cannot stand in for the whole claim. - Use Case: After a bug fix, ask whether the transport no longer overwrites an active user seek without regressing protected behavior. The Skill runs or inspects the smallest applicable test, checks the protected state, and reports a bounded verdict. ## Quick Start Ask the AI to verify whether the claim "the installer preserves user data during repair" is supported by the available evidence and to report a PASS, PARTIAL, FAIL, BLOCKED, or UNVERIFIED status.
