Validation Phase Skill

ROOT AGENT ONLY - Spawns agents, runs only from root Claude Code agent.

Purpose: Validate implementation quality before reporting Trigger: After execution-phase completes Input: workerResults, complexity (int), prdPath, scope Output: {status, confidenceScore?, buildChecks, reviewFindings}

---

Workflow

1. Run confidence-check (if complexity ≥15)

• 5-point weighted assessment: duplicates, patterns, architecture, examples, requirements
• Score ≥90% → Proceed
• Score 70-89% → Flag concerns
• Score <70% → Escalate

2. Verify Acceptance Criteria (AC) completion

• Scan story files in PRD for AC checkbox format: - [x] (completed) or - [ ] (incomplete)
• If any AC marked incomplete (- [ ]):
  • Report list of stories with incomplete AC
  • CRITICAL block - must complete AC before proceeding
• If all AC checked (- [x]):
  • Continue to quality checks

3. Run code-quality-validator agent

• Scales checks by change size (Small/Medium/Large)
• Small (1-3 files): pnpm build
• Medium (4-10 files): pnpm build && lint && prettier
• Large (11+ files): pnpm build && lint && prettier && test

4. Run reviewer agent

• Code patterns, SOLID principles, security
• Error handling, naming consistency, PRD compliance
• Return quality assessment + recommendations

---

Validation Strategy

Change Size	Files	Checks Run
Small	1-3	Build only
Medium	4-10	Build + Lint + Prettier
Large	11+	Build + Lint + Prettier + Tests

---

Failure Handling

Failure Type	Action
Incomplete AC	CRITICAL - must complete before proceeding
Build failure	CRITICAL - must fix before proceeding
Lint/Prettier warnings	Non-blocking, provide recommendations
Test failures	Ask user: fix, continue with warnings, or skip

---

Output Example

{
  "status": "pass",
  "confidenceScore": 92,
  "acVerification": {
    "status": "pass",
    "checked": 8,
    "incompleteStories": []
  },
  "buildChecks": {
    "build": { "status": "pass" },
    "lint": { "status": "pass" },
    "prettier": { "status": "pass" },
    "tests": { "status": "pass" }
  },
  "reviewFindings": [
    {
      "severity": "info",
      "file": "src/auth.service.ts",
      "message": "Add JSDoc comments"
    }
  ],
  "recommendations": ["Add integration tests for edge cases"]
}

AC Failure Example:

{
  "status": "blocked",
  "acVerification": {
    "status": "incomplete",
    "checked": 8,
    "incompleteStories": [
      "prj-epc-002: User login - AC 2, 4 incomplete",
      "prj-epc-005: Password reset - AC 1 incomplete"
    ]
  },
  "message": "Cannot proceed: 2 stories have incomplete acceptance criteria"
}

---

Integration

Called by:

• /audit command (after execution-phase)
• /build command (after execution-phase)
• /ms command (for complexity ≥15)

Calls:

• confidence-check skill (if complexity ≥15)
• code-quality-validator agent
• reviewer agent
• AskUserQuestion tool (on test failures)

Next step: report-phase (final report + sign-off)