# Verification Evidence

> Ground completion in executed checks and regression evidence. Use after fixes, refactors, dependency changes or before a completion claim.

- Skill: `noumenonnn/verification-evidence` (Agent Skill)
- Install (CLI): `npx skillmds@latest add noumenonnn/verification-evidence`
- Raw SKILL.md: https://api.skillmd.com/api/skills/noumenonnn/verification-evidence/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: noumenonnn (https://skillmd.com/u/noumenonnn)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/noumenonnn/verification-evidence

---


# Verification Evidence

Choose checks from the affected behavior: a reproduction/targeted test, relevant static checks, then broader build/integration coverage proportionate to risk. Verify test discovery/counts and expected assertions. Do not equate process exit 0, a successful build or absence of review findings with overall correctness.

For a stable bug seam, demonstrate red before the fix and green after with a regression test. Use installed TDD/debugging discipline when appropriate, not duplicate workflows.

Builds and tests may write caches/artifacts; do not promise OS read-only execution. Do not autofix, update snapshots or install dependencies solely to make checks pass. Stop or report BLOCKED if the environment cannot execute a check. Do not race checks with another writer in the same tree.

Final evidence:
- Verified: exact commands, scope, outcomes and evidence paths.
- Failed: observed failure and relevant context.
- Blocked/not run: missing capability and why.
- Residual risk: only meaningful unverified behavior/platforms.

Use PASS only for relevant checks that actually ran successfully. Stale checks from before a subsequent edit are not current evidence without a justified impact assessment.