What problem does it solve? Finished tasks often claim success without evidence: tests were never re-run, acceptance criteria quietly changed mid-task, or implemented code is a stub that nothing calls. This Skill audits a completed task against its own task file and produces a per-criterion verdict backed by evidence. ## Core Features & Use Cases - Goal-backward criterion checks: Each acceptance criterion is verified in three steps — the artifact exists, is substantive (no stubs or TODO placeholders), and is wired into the codebase — with path:line evidence recorded per step. - Criteria drift detection: Compares the task file's git history to find criteria that were changed, added, or removed during the work, and flags any change the execution log does not explain. - Fresh test execution and independent review: Re-runs all related tests and the repo check command, then dispatches spec-blind validator and reviewer subagents over the full diff. - Use Case: Before completing a multi-phase task like AUTH-001, run verification to confirm every acceptance criterion is met, tests pass fresh, and no scope creep occurred — then append the verdict to the task's execution log. ## Quick Start Ask the agent to verify task AUTH-001 to check whether its acceptance criteria were actually implemented correctly.
