What problem does it solve? AI assistants and developers often claim work is done, fixed, or working based on stale test runs or assumptions, leading to broken code being pushed, PRs opened on failing builds, and unverified handoffs. This Skill enforces an evidence gate: no completion claim is permitted without a fresh run of the project's verify command and a zero exit code. ## Core Features & Use Cases - Evidence Gate: Blocks done/works/fixed claims, git push, PR creation, and session handoffs until the verify command runs fresh and exits 0. - Automatic Verify Command Detection: Resolves the right command via a priority chain — explicit declaration in CLAUDE.md/KNOWLEDGE.md, Makefile/justfile targets, package.json scripts, CI workflow run steps, or tool-presence fallback (bats, shellcheck, pytest, cargo, go). - Accountability Tracking: Converts unproven claims into Critical [VERIFY] TODOs in TODOS.md, routes test failures to [TEST]/[BUG] workflows, and writes [UX] checklists for machine-unverifiable results. - Use Case: Before pushing a bug fix, the gate runs bats --tap tests/ fresh, reads exit 0 with 82 passing tests, and only then permits the claim — citing the evidence. ## Quick Start Ask the AI to verify the current changes with the project's verify command before claiming the fix is done.
