What problem does it solve? Teams and AI agents often accept completion claims based on stale reports, partial evidence, or a subagent's word, leading to false confidence that work is done. This Skill forces every explicit completion or success claim to be checked against freshly reproduced evidence before it is trusted. ## Core Features & Use Cases - Claim-First Verification: Restates the claim in its narrowest form, then demands evidence covering exactly that claim — not a broader or smaller one. - Three-Way Verdict: Returns VERIFIED, NOT_VERIFIED, or MANUAL_VERIFY_REQUIRED, keeping 'could not check' strictly separate from 'checked'. - Consequence-Free Operation: Writes nothing, mutates nothing, and never advances a Spec — completion recording is explicitly delegated to a separate validation workflow. - Use Case: A subagent reports that a bug fix passed all tests. Before accepting that, run this Skill to re-execute the tests yourself and confirm the evidence covers the exact claim being made. ## Quick Start Use the sb-verify-completion skill to verify the claim that the recent fix passes all tests before we mark the task done.
