Verify Complete — Verification Gate

INTERNAL SKILL — This skill is now part of the /weave orchestrator. Use /weave instead for the full graph-first workflow. Direct invocation is deprecated and may be removed in a future release.

Trigger: Before running wv done on any non-trivial node.

Purpose: Address failure mode #3 (hallucination without verification). Ensures claims are backed by evidence, not assumptions.

Instructions

When invoked with a node ID (/wv-verify-complete <wv-id>):

1. Identify Verification Method

Choose the strongest applicable method:

2. Execute Verification

Run the appropriate command(s):

# Examples by project type
pnpm test              # JavaScript/TypeScript
pytest                 # Python
cargo test             # Rust
go test ./...          # Go
pnpm run build         # Build check
pnpm run typecheck     # Type check
ruff check .           # Python lint

3. Capture Evidence

Record verification in node metadata:

wv update <id> --metadata='{
  "verification": {
    "method": "test|build|typecheck|lint|runtime|diff",
    "command": "command that was run",
    "result": "pass|fail",
    "evidence": "key output snippet or observation",
    "verified_at": "ISO timestamp"
  }
}'

4. Gate Decision

• •PASS: Verification succeeded → proceed to wv done
• •FAIL: Verification failed → fix issues, re-verify
• •NO TESTS: Document manual verification steps explicitly

Output Format

## Verification: wv-XXXXXX

**Method:** test **Command:** `pnpm test src/auth` **Result:** PASS

**Evidence:**

```text
PASS src/auth/login.test.ts ✓ should authenticate valid credentials (12ms) ✓ should reject invalid
password (8ms)

Test Suites: 1 passed, 1 total Tests: 2 passed, 2 total
```

Verified at: 2026-02-03T08:15:00Z

NEVER

• •Close a node claiming "should work" without running verification
• •Accept "no errors in editor" as sufficient evidence
• •Skip verification for "trivial" changes that touch shared code
• •Claim tests pass without actually running them

ALWAYS

• •Run at least one executable check (test/build/lint/typecheck)
• •If no automated checks exist, document manual verification steps
• •Include command output or observations as evidence
• •Record verification in node metadata before closing

Escalation

If verification is not possible:

1. •Document why (no tests, environment issue, etc.)
2. •Add metadata.verification.method = "deferred"
3. •Create follow-up node: wv add "Add tests for <area>" --metadata='{"type":"task"}'
4. •Proceed with explicit acknowledgment of verification gap

Integration

This skill is called by:

• •/fix-issue — Step 5 (Verify) should invoke /wv-verify-complete
• •Pre-close gate hook — Blocks wv done if metadata.verification missing
• •Manual invocation when completing any significant work

Examples

Example 1: Test Verification

/wv-verify-complete wv-a1b2

# Agent runs: pnpm test src/utils/parser.test.ts
# Output: 5 tests passed
# Updates metadata with verification evidence
# Proceeds to wv done

Example 2: Build Verification (No Tests)

/wv-verify-complete wv-c3d4

# No tests for this area
# Agent runs: pnpm run build
# Output: Build successful, no errors
# Agent runs: pnpm run typecheck
# Output: No type errors
# Records both as verification evidence

Example 3: Manual Verification

/wv-verify-complete wv-e5f6

# UI change, no automated tests
# Agent documents:
#   - Opened browser to /dashboard
#   - Clicked "Export" button
#   - CSV downloaded with correct data
#   - Verified 3 rows matched database
# Records as runtime verification with steps

Metadata Schema

{
  "verification": {
    "method": "test",
    "command": "pnpm test src/auth",
    "result": "pass",
    "evidence": "5 tests passed in 1.2s",
    "verified_at": "2026-02-03T08:15:00Z",
    "files_checked": ["src/auth/login.ts", "src/auth/login.test.ts"]
  }
}

Related Skills

• •/ship-it — Define done criteria before implementation
• •/fix-issue — Calls wv-verify-complete as part of workflow
• •/close-session — Final verification before session end