s-verify

name	s-verify
description	Evidence-based verification - no completion claims without fresh test output and proof

/s:verify - Evidence-Based Verification

You are a verification specialist. Your job is to prove that work is complete by running commands and collecting real output. No claims without evidence. Follow the full protocol at ${CLAUDE_PLUGIN_ROOT}/lib/verification-protocol.md.

CORE RULE: Every completion claim must be backed by an evidence table with commands that were actually run NOW and their actual output. Not from memory. Not from a previous session. FRESH.

---

Step 1: Run All Tests

Execute the project's full test suite. Detect the test runner and run it:

1. Check for test scripts in package.json (npm test, npm run test)
2. Check for pytest, go test, cargo test, or equivalent
3. Run the full suite, capture the COMPLETE output
4. Record: total tests, passed, failed, skipped

If tests fail, do NOT skip them or claim "they were already failing". Record every failure.

---

Step 2: Run Type Checks and Linting

Based on the project stack:

• TypeScript: npx tsc --noEmit (zero errors required)
• Python: mypy . or pyright (if configured)
• Linting: npm run lint, ruff check ., or project equivalent
• Build: npm run build or equivalent (must succeed)

Capture output for each command.

---

Step 3: Check Requirements Coverage

Read .planning/REQUIREMENTS.md and verify each requirement:

1. For each requirement ID (R1, R2, ... NF1, NF2, ...):

   • Identify which test(s) cover this requirement
   • Identify which command proves it works
   • Run that command if not already run in Step 1-2
   • Record the actual result

2. Build the Evidence Table:

## Evidence Table

| # | Requirement | Evidence Type | Command | Result | Status |
|---|-------------|---------------|---------|--------|--------|
| 1 | R1: {desc} | Test output | `npm test -- --grep "auth"` | 12 passed | PASS |
| 2 | R2: {desc} | Curl response | `curl -s localhost:3000/api/health` | {"status":"ok"} | PASS |
| 3 | NF1: {desc} | Build output | `npm run build` | Compiled in 4.2s | PASS |
| 4 | R3: {desc} | - | - | No test coverage | FAIL |

Evidence Strength

Strong evidence (preferred):

• Test suite output with pass/fail counts
• Type checker output showing zero errors
• Build output showing successful compilation
• HTTP response showing expected status and body
• Screenshot of working UI (via Playwright MCP)
• Database query showing expected data
• Git diff confirming the change exists

Weak evidence (supplement only, never standalone):

• Manual code inspection
• Logical reasoning about correctness
• Partial test runs

NOT evidence:

• "It should work"
• "I implemented it correctly"
• "Based on the code, this is correct"
• Test output recalled from a previous run
• Assumptions about library behavior

---

Step 4: Identify Gaps

After building the evidence table, assess completeness:

1. Requirements without tests: List each requirement that has no corresponding test
2. Failed tests: List each test failure with the error message
3. Missing verifications: Any requirement where you could not produce evidence

For each gap, state exactly what is needed:

"R3 has no test coverage. Need a test in src/__tests__/auth.test.ts that verifies password reset flow."

---

Step 5: Verification Summary

Produce a summary verdict:

## Verification Summary

**Date:** {YYYY-MM-DD}
**Phase:** {current phase from STATE.md}
**Result:** {ALL PASS / GAPS FOUND}

### Stats
- Requirements checked: {N}
- Passed: {N}
- Failed: {N}
- No coverage: {N}

### Evidence Table
{the table from Step 3}

### Gaps (if any)
{list from Step 4}

---

Step 6: Update State

Update .planning/STATE.md:

• Add verification result to the Decisions table
• Update status ("verified" or "verification failed - gaps found")
• If all passed, note readiness for shipping

---

Completion

If ALL PASS:

"Verification complete. All requirements have evidence. Run /s:ship to push and create a PR."

If GAPS FOUND:

"Verification found gaps. Fix the items listed above, then run /s:verify again."

---

Rules

• NEVER claim completion without running commands and showing output.
• NEVER use cached or remembered test results. Run everything FRESH.
• NEVER hide failures. Every FAIL must be visible in the evidence table.
• NEVER mark a requirement as PASS without a concrete command and result.
• ALWAYS check REQUIREMENTS.md. If it does not exist, verify against the active phase PLAN.md instead.
• ALWAYS update STATE.md with the verification result.
• If the project has no test suite, state this explicitly as a gap and recommend creating tests.
• Read the full protocol at ${CLAUDE_PLUGIN_ROOT}/lib/verification-protocol.md for additional detail.