---

name: verification-before-completion description: Evidence-based completion claims - BLOCKS claiming tests pass, builds succeed, features work, or tasks are done without running verification commands first. Enforces TDD for implementation tasks. Triggers on done, complete, fixed, passes, works, ready, finished, verified, all good, ship it, commit, PR, push, merge, implement, create feature, build.

Verification Before Completion

The Iron Law

NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE

Never claim work is complete, fixed, passing, or ready without:

1. Running the verification command
2. Reading the actual output
3. Confirming it matches your claim

TDD Enforcement (Implementation Tasks)

For ANY implementation task (new feature, component, handler, schema):

Phase	Requirement	Evidence
1. TEST FIRST	Write failing test BEFORE implementation	Test file exists, test fails with expected reason
2. IMPLEMENT	Write minimal code to pass test	Test now passes
3. REFACTOR	Clean up while tests pass	Tests still pass
4. VERIFY	Run full verification suite	Build + Types + Tests pass

TDD is NOT optional for:

• New components (React components, hooks)
• New data provider handlers
• New Zod schemas
• New service classes
• Bug fixes (regression test first)

TDD Red Flags - STOP if you're about to:

• Write implementation code without a test file
• Say "I'll add tests later"
• Skip the failing test phase
• Implement multiple features before testing

Manual E2E for UI Changes

For ANY UI-facing change, offer Manual E2E Testing:

When implementing UI changes (forms, lists, slide-overs, buttons), remind user:

Manual E2E Available: This UI change can be verified using Claude Chrome. See docs/tests/e2e/ for existing test prompts or I can generate a new one.

Would you like me to generate a Claude Chrome test prompt for this feature?

When to generate E2E prompts:

• New forms (CRUD create/edit)
• New validation rules
• UI workflow changes
• Accessibility improvements
• Any user-facing feature

The 5-Step Gate

Before ANY completion claim:

Step	Action	Required Output
1. IDENTIFY	What command proves your claim?	npm run build, npx tsc --noEmit, etc.
2. RUN	Execute the command fresh	Full terminal output
3. READ	Review complete output	Exit codes, error counts
4. VERIFY	Does output match claim?	YES → proceed, NO → fix first
5. CLAIM	State result WITH evidence	"Build passes (exit 0, 0 errors)"

Skip any step = lying, not verifying

Crispy CRM Verification Commands

Claim	Required Command	Success Criteria
Build passes	npm run build	Exit 0, no errors
Types are correct	npx tsc --noEmit	Exit 0, no type errors
Tests pass	npm test	All tests pass, 0 failures
E2E works	npx playwright test	All scenarios pass
Feature complete	Checklist verification	All requirements checked
Ready to commit	Build + Types + Tests	All three pass

Red Flags - STOP

If you're about to say ANY of these, STOP and verify:

Red Flag	What To Do Instead
"Should work now"	RUN npm run build
"Types look correct"	RUN npx tsc --noEmit
"Tests should pass"	RUN npm test
"I'm confident this works"	Confidence ≠ evidence
"The feature is complete"	Verify against requirements
"Ready for commit/PR"	Run full verification suite
"Looks good"	Show actual output

Rationalization Prevention

Excuse	Reality
"Should work now"	RUN the command
"I'm confident"	Confidence ≠ evidence
"Just this once"	No exceptions
"Linter passed"	Linter ≠ TypeScript compiler
"I checked the code"	Reading ≠ running
"It worked before"	Before ≠ now
"Partial check is enough"	Partial proves nothing

Correct Patterns

Build Verification

✅ CORRECT:
[Run: npm run build]
[See: exit 0, no errors]
"Build passes - exit 0, compiled successfully"

❌ WRONG:
"Build should pass now"
"The code looks correct"
"I fixed the error so it works"

TypeScript Verification

✅ CORRECT:
[Run: npx tsc --noEmit]
[See: no output (means success)]
"TypeScript check passes - 0 errors"

❌ WRONG:
"Types look correct"
"Fixed the type error"
"Should compile now"

Feature Completion

✅ CORRECT:
Requirements checklist:
- [x] User can create contact (verified via test)
- [x] Validation shows errors (verified via build)
- [x] Form saves to database (verified via E2E)
"Feature complete - all 3 requirements verified"

❌ WRONG:
"Feature is complete"
"All requirements implemented"
"Ready for review"

Git Operations

✅ CORRECT:
[Run: npm run build] → exit 0
[Run: npx tsc --noEmit] → 0 errors
"Ready to commit - build and types verified"

❌ WRONG:
"Ready to commit"
"Can push now"
"PR is ready"

When This Applies

ALWAYS before:

• Claiming tests pass
• Claiming build succeeds
• Claiming types are correct
• Claiming bug is fixed
• Claiming feature is complete
• Claiming ready for commit/PR/push
• Any variation of "done", "works", "fixed"

Rule applies to:

• Exact phrases ("tests pass")
• Paraphrases ("specs are green")
• Implications ("we're good to go")
• Any positive statement about work state

The Bottom Line

No shortcuts for verification. Run the command. Read the output. THEN claim the result. This is non-negotiable.

Evidence before claims, always.