Verification Before Completion
Core Principle: No completion claims without fresh verification evidence.
Unverified claims break trust and ship broken code: undefined functions that crash production, incomplete features missing requirements, lost time on rework after false completion. The cost of running one command is trivial compared to the cost of a wrong claim.
The Verification Gate
Before any claim of success, completion, or satisfaction:
- IDENTIFY - What command proves this claim?
- RUN - Execute the full verification command (fresh, complete)
- READ - Check full output, exit code, failure counts
- VERIFY - Does output confirm the claim?
- NO - State actual status with evidence
- YES - State claim WITH evidence from step 2-3
When This Applies
Before:
- Claiming "tests pass", "build succeeds", "linter clean", "bug fixed"
- Expressing satisfaction ("Great!", "Done!", "Perfect!")
- Using qualifiers ("should work", "probably fixed", "seems to")
- Committing, creating PRs, marking tasks complete
- Marking a multi-file implementation as complete (dispatch
ce:code-revieweragent) - Moving to next task or delegating work
Common Verification Requirements
| Claim | Required Evidence | Not Sufficient |
|---|---|---|
| Tests pass | yarn test output: 0 failures | Previous run, "looks correct" |
| Build succeeds | Build command: exit 0 | Linter clean, "should work" |
| Bug fixed | Test reproducing bug: now passes | Code changed, assumed fix |
| Linter clean | Linter output: 0 errors | Partial check, spot test |
| Regression test works | Red→Green cycle verified | Test passes once |
| Agent task complete | VCS diff shows expected changes | Agent reports "success" |
| Work is complete | Code review via ce:code-reviewer with no unresolved Critical issues | Self-review, "looks good to me" |
Red Flags
Stop and verify if you're about to:
- Use hedging language ("should", "probably", "seems to")
- Express satisfaction before running verification
- Trust agent/tool success reports without independent verification
- Rely on partial checks or previous runs
Key Examples
Regression test (TDD Red-Green):
Write test → Run (fail) → Fix code → Run (pass) → Revert fix → Run (fail) → Restore → Run (pass)
vs. "I've written a regression test" (without verifying red-green cycle)
Build vs Linter:
Run `npm run build` → See "exit 0" → Claim "build passes"
vs. Run linter → Claim "build will pass" (linter ≠ compiler)
Agent delegation:
Agent reports success → Check `git diff` → Verify changes → Report actual state
vs. Trust agent's success message without verification