Verification Before Completion

Core Principle

"Evidence before claims, always."

Run FRESH verification commands before asserting work is complete.

The 5-Step Gate

1. Identify the proof command (test, build, lint, etc.)
2. Execute it fully (no partial runs)
3. Read complete output AND exit codes
4. Confirm output matches the claim
5. State result with evidence

What DOESN'T Count as Verification

• Previous test runs ("it passed before")
• "Should pass" reasoning
• Partial checks or extrapolation
• Agent success reports without independent verification
• Code changes without testing original symptom

What DOES Count

• Fresh command execution
• Exit code 0 confirmation
• 0 failures, 0 errors in output
• Red-green regression (fail → fix → pass cycle)

Verification Commands by Project

Python/pytest:

pytest tests/ -v --tb=short
echo "Exit code: $?"

Node.js:

npm test
echo "Exit code: $?"

Build verification:

npm run build  # or python -m build
echo "Exit code: $?"

Red Flags (DO NOT proceed)

• Using words: "should," "probably," "seems to"
• Feeling tired and wanting to be done
• Relying on partial evidence

$ARGUMENTS