# Verification Before Completion

> Requires fresh verification evidence before claiming work complete. Use this skill when about to say done, open a PR, or hand off, before asserting tests pass or behavior is correct. Do not use when/for choosing merge vs discard options after verification (use branch-completion) or mid-debug evidence gathering (use debugging-methodology).

- Skill: `jagreehal/verification-before-completion` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add jagreehal/verification-before-completion`
- Raw SKILL.md: https://api.skillmd.com/api/skills/jagreehal/verification-before-completion/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: jagreehal (https://skillmd.com/u/jagreehal)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/jagreehal/verification-before-completion

---


# Verification Before Completion

## Critical rules

- `NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE`.
- Agent confidence is not evidence. Only fresh command output is.
- MUST: run the proving command this turn; show actual output and exit code/failure count; re-verify after any code change; verify all requirements, not only tests.
- NEVER: "should"/"probably"/"seems to" for verified status; celebrate before running; commit/push/PR without verification; trust sub-agent success; rely on partial or previous runs.
- Red flags: satisfaction language before commands, "just this once", any success wording without a fresh run.
- Before matching claims to commands, read [references/claim-matrix.md](references/claim-matrix.md).

## Workflow

1. IDENTIFY the command that proves the claim.
2. RUN it fully and freshly this turn.
3. READ full output, exit code, failure count.
4. VERIFY: output confirms the claim? If no → state actual status with evidence. If yes → state claim with evidence.
5. ONLY THEN make the claim. Skip any step = lying, not verifying.

## Resources

- [references/claim-matrix.md](references/claim-matrix.md) — claim→command table, correct patterns, TDD red-green, rationalizations. Read before claiming status.

## Validation

- [ ] Exact proving command identified
- [ ] Command run in full this turn
- [ ] Output and exit code/failure count read
- [ ] Output confirms the claim
- [ ] No "should"/"probably"/"seems to" for verified status
- [ ] Sub-agent reports independently verified (diff or fresh output)

## Constraints

- No "trivial enough to skip" exception.
- Related: `tdd-workflow`, `debugging-methodology`, `testing-strategy`, `session-continuity`, `branch-completion`.