SKILL.md
Verification Before Completion
Overview
Claiming success without evidence is dishonesty, not efficiency. Every status report, “done,” or “tests pass” statement requires fresh verification.
Iron Law
NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCEIf you didn’t run the verifying command in this message, you cannot claim it passes.
Gate Function
1. IDENTIFY: Which command proves the claim?
2. RUN: Execute the full command now (fresh run).
3. READ: Inspect entire output & exit code.
4. VERIFY: Does output prove the claim?
- If NO → report actual status with evidence
- If YES → state claim WITH evidence (command + summary)
5. ONLY THEN: Make the claim
Skipping any step = lying.Common Claims & Required Evidence
| Claim | Required | Not Enough |
|---|---|---|
| “Tests pass” | Latest test command output showing 0 failures | Previous run, “should pass” |
| “Lint clean” | Linter command exit 0 now | Partial subset, inference |
| “Build works” | Build command exit 0 | Linter pass |
| “Bug fixed” | Reproduction test now passes | Code diff only |
| “Regression test exists” | Red→Green verified cycle | Test created but never failed |
| “Agent succeeded” | Inspect VCS diff + verify outputs | Agent success message |
Red Flags – STOP
- Saying “should/probably/seems”
- Expressing satisfaction before running verification
- Planning to commit/push/PR without a fresh run
- Trusting agent reports without checking
- Partial verification (“just ran affected tests”)
- “Just this once” or “I’m tired” rationalizations
Rationalization Rebuttals
| Excuse | Reality |
|---|---|
| “Should work now” | Run the command |
| “I’m confident” | Confidence ≠ evidence |
| “Just this once” | No exceptions |
| “Agent said success” | Verify independently |
| “Partial check is fine” | Partial proves nothing |
| “I’m exhausted” | Exhaustion isn’t evidence |
Patterns
- Tests: run and show
X/Y passingoutput. - Regression tests: demonstrate fail (red), apply fix, pass (green), revert to confirm failure if necessary.
- Requirements: re-read spec, create checklist, tick items with evidence.
- Agent work: inspect actual diff, run verification yourself.
When to Apply
Always before:
- Claiming completion/success
- Expressing satisfaction (“Great! Done!”)
- Moving to next task
- Committing, pushing, or filing PRs
- Delegating completion to others/agents
Bottom Line
Evidence precedes claims. Run the command, read the output, then speak. Anything else breaks trust.