verify

# /verify - Verification Runner

You run all verification methods from a definition file against the current codebase state. You are called by /do, not directly by users.

## Input

`$ARGUMENTS` = "<definition-file-path> <execution-log-path>"

Example: `/tmp/define-123.md /tmp/do-log-123.md`

## Process

### 1. Parse Inputs

Read both files:
- Definition file: extract all criteria and their verification methods
- Execution log: understand what was attempted (for context in subagent checks)

### 2. Categorize Criteria

Group by verification method:
- **bash**: Shell commands (npm test, tsc, lint, etc.)
- **subagent**: Code pattern checks via Task tool
- **manual**: Require human verification

### 3. Run Bash Verifications

For each bash criterion:
```bash
# Run with higher timeout (5 minutes default)
# Retry once on transient failures (network, flaky tests)
```

Capture:
- Exit code
- stdout/stderr
- Specific failure location if detectable

### 4. Run Subagent Verifications

Launch in parallel waves (max 5 concurrent):

```
Use the Task tool to check criterion AC-N:
Task("vibe-experimental", "criteria-checker", "Check AC-N: [description]. Context: [relevant files]. Checks: [from definition]")
```

For each subagent:
- Provide criterion definition
- Provide relevant code files
- Provide execution log excerpt
- Request pass/fail with specific file:line issues

### 5. Collect Results

Build results structure:
```
{
  "passed": ["AC-1", "AC-2", ...],
  "failed": [
    {
      "id": "AC-3",
      "method": "bash",
      "command": "npm test",
      "location": "src/foo.test.ts:45",
      "expected": "'queued'",
      "actual": "'sent'"
    },
    {
      "id": "AC-7",
      "method": "subagent",
      "location": "src/handlers/notify.ts:23",
      "issue": "Raw Error() throw, expected AppError"
    }
  ],
  "manual": ["AC-10", "AC-11"]
}
```

### 6. Decision Logic

```
if any automated failed:
    → Return failures only (hide manual)
    → /do continues working

elif all