Speckit.validate

🎯 Mission

Check whether the ENTIRE implementation meets spec.md or not — final gate before deploying.

📥 Input

• All artifacts: spec.md, plan.md, tasks.md
• Source code (implementation)
• .agents/memory/constitution.md
• .agents/project.json, dependency manifests, and documented verification commands

📋 Protocol

1. Tasks Completion: All tasks in tasks.md have [X] ?
2. Success Criteria: All SCs in spec.md passed?
3. Capability Discovery:
   • Detect project type, language, package manager, containerization, configured services, build/test commands, and health checks.
   • Produce an applicability table. A non-applicable check is N/A, not PASS.
4. Build Verification (MUST run applicable commands):
   • Run the repository-defined build/type/lint commands.
   • Use the matching container service when the constitution requires Docker.
   • For non-Docker projects, use the documented project runtime.
   • Any applicable build failure → ❌ BLOCKED.
5. Runtime Verification (when the project has a runtime):
   • Start the documented runtime or discovered Compose configuration.
   • Wait using health/retry behavior with a bounded timeout; do not use a fixed sleep as proof of readiness.
   • Inspect failing service/application logs immediately.
6. Health Check (when configured):
   • Read health endpoints and ports from project configuration or environment.
   • Verify the expected status/body. Do not assume web/admin/API services.
7. Tests and Success Criteria:
   • Run the configured test suite and verify each spec success criterion with evidence. Missing required evidence → ❌ BLOCKED.
8. Constitution Check: No rules violated?
9. Final Verdict:

   🏁 VALIDATION REPORT
   ═══════════════════════
   Tasks:        15/15 ✅
   Build:        PASS ✅
   Runtime:      N/A (library project)
   Health:       N/A (no network service)
   Tests:        PASS ✅
   Constitution: PASS ✅
   ───────────────────────
   VERDICT: ✅ READY FOR DEPLOY
   

📤 Output

• File: .agents/memory/validation-report.md
• Verdict: ✅ PASS or ❌ FAIL (with blockers list)

🚫 Guard Rails

• DO NOT approve if there are unfinished tasks.
• DO NOT approve if an applicable build, test, runtime, or health check fails.
• MUST run actual commands — don't just read the code.
• DO NOT fail a project merely because a non-applicable Docker, TypeScript, or network-service check does not exist.