Completeness Validator

Completeness Validator

Validate that work is truly complete - not superficially "done" but production-ready, wired-up, tested, and functional. This skill uses multilingual completion semantics to reinforce the distinction between partial implementation and actual completion.

When to use me

• When AI claims something is "complete" but you suspect it's only 99% done
• Before accepting "finished" code that may have mock data or unconnected systems
• When you need to verify all gaps are closed (specs, tests, wiring, integration)
• To prevent the "works on my machine" syndrome in AI-generated code
• When you want production-ready systems, not prototypes or samples

What I do

• Multilingual completion validation: Use precise completion terminology (Japanese 完成/完了/完全, Latin completus/perfectus/finitus, German fertig/vollständig/abgeschlossen) to reinforce true completion
• Gap detection: Identify missing wiring, unconnected systems, mock data, placeholder implementations
• Production readiness check: Verify code is deployable, not just functional in isolation
• Integration validation: Ensure all components actually connect and communicate
• Test verification: Confirm tests exist AND they actually validate real functionality
• Spec compliance: Verify implementation matches specifications completely

The Completion Hierarchy

Level 1: Superficial (表面完了 / superficiarius)

• Code exists
• Compiles without errors
• "Looks" complete

Level 2: Functional (機能完了 / functionalis)

• Runs in isolation
• Basic tests pass
• Mock data works

Level 3: Integrated (統合完了 / integratus)

• Connected to real systems
• Uses actual data sources
• Inter-component communication works

Level 4: Production (本番完了 / productio)

• Deployed and running
• Real users can access
• Monitoring in place
• All edge cases handled

Level 5: Complete (完全完了 / perfectus)

• All specs implemented
• All tests passing (unit, integration, e2e)
• Documentation complete
• No technical debt
• Handoff-ready

Multilingual Completion Commands

Use these phrases to force AI to acknowledge completion depth:

Japanese (完成度の確認)

• 「これは本当に完成ですか？」 (Is this truly complete?)
• 「モックデータではなく、本物の実装ですか？」 (Is this real implementation, not mock data?)
• 「すべての配線が完了していますか？」 (Is all wiring complete?)
• 「統合テストは完了しましたか？」 (Are integration tests complete?)
• 「完全に完了した状態ですか？」 (Is this in a fully completed state?)

Latin (Perfectio Verification)

• "Estne hoc vere completus?" (Is this truly complete?)
• "Omnes partes conectae sunt?" (Are all parts connected?)
• "Nulla simulatio data?" (No simulated data?)
• "Paratus est productio?" (Is it ready for production?)
• "Perfectus et finitus?" (Perfected and finished?)

German (Vollständigkeitsprüfung)

• "Ist das wirklich vollständig?" (Is this really complete?)
• "Sind alle Komponenten verbunden?" (Are all components connected?)
• "Keine Mock-Daten, echte Implementierung?" (No mock data, real implementation?)
• "Produktionsbereit und getestet?" (Production-ready and tested?)
• "Absolut abgeschlossen?" (Absolutely finished?)

Three-Tier Validation System

Tier 1: Static Analysis (AST-Based)

Replaces superficial grep with semantic code analysis

• No empty functions - AST parsing detects function foo() {} or pass statements
• No stub implementations - Detects throw new NotImplementedError() or // TODO: implement
• All imports resolve - Static analysis verifies module dependencies
• Error handling coverage - AST checks for try/catch blocks in async functions
• Type safety - TypeScript/Flow types present and valid

Tools used:

• JavaScript/TypeScript: eslint, @typescript-eslint/parser
• Python: pylint, ast module
• Rust: cargo check, clippy
• Go: go vet, staticcheck

Tier 2: Integration Verification (Runtime)

Actually runs the code, doesn't just check file existence

• Services start - Application boots without crashes
• Database connections - Real queries execute (not mocked)
• External API calls - HTTP requests made to real endpoints
• Environment validation - .env values actually work
• Health checks pass - /health or equivalent endpoint responds 200

Methods:

• Testcontainers for real databases/services
• Network capture to verify actual HTTP requests
• Smoke tests that exercise real code paths

Tier 3: Production Readiness Review (PRR)

Industry-standard gates based on research from Cortex.io

• Observability - Logs, metrics, traces configured
• On-call - Runbooks, escalation paths, pager integration
• Security - Secrets management, vulnerability scans
• CI/CD - Pipeline tested, rollback tested, canary deployment ready
• Documentation - README tested by new developer, API docs accurate
• Compliance - Required checks passing, audit trail enabled

Reference: Based on Production Readiness Review (PRR) standards used by 98% of engineering leaders

Usage

# Run full three-tier validation
@completeness-validator check

# Run specific tier only
@completeness-validator check --tier 1    # AST analysis only
@completeness-validator check --tier 2    # Runtime verification only
@completeness-validator check --tier 3    # PRR gates only

# Generate detailed completion report
@completeness-validator report

# Force AI to acknowledge completion depth
@completeness-validator confirm --depth perfectus --interactive

# Available depth levels:
#   superficialis  - Code exists (Level 1)
#   functionalis   - Runs with mocks (Level 2)
#   integratus     - Connected to real systems (Level 3)
#   productio      - Deployed and monitored (Level 4)
#   perfectus      - Complete in all aspects (Level 5)

Output Format

COMPLETENESS VALIDATION REPORT
==============================

Target Level: Production (本番完了)
Validation Time: 2024-01-15 14:30:00

CHECKS PASSED: 18/25 (72%)

❌ CRITICAL GAPS DETECTED:
   1. Database connection not configured (統合未完了)
   2. API endpoints return mock data (モックデータ使用中)
   3. Missing integration tests (統合テスト欠如)
   4. Environment variables not documented (環境変数未文書化)

⚠️  WARNINGS:
   1. 3 TODO comments found (TODOコメント残存)
   2. Error handling incomplete in 2 functions (エラーハンドリング不完全)

STATUS: NOT COMPLETE (未完成)
RECOMMENDATION: Address critical gaps before marking complete

Next Steps:
1. Configure database connection
2. Replace mock data with real implementation
3. Add integration tests
4. Document environment setup

Examples

Example 1: Detecting Mock Data

AI says: "The user authentication is complete."

You ask: 「これは本当に完成ですか？モックデータではなく、本物のデータベース接続ですか？」 (Is this truly complete? Is this real database connection, not mock data?)

Validation reveals:

• ❌ Using in-memory user store (mock)
• ❌ No database connection configured
• ❌ Password hashing not implemented
• ❌ Session management missing

Result: NOT COMPLETE (未完成)

Example 2: Verifying Integration

AI says: "The payment processing is done."

You ask: "Estne hoc vere completus? Omnes partes conectae sunt?" (Is this truly complete? Are all parts connected?)

Validation reveals:

• ✅ Payment logic implemented
• ❌ Not connected to payment gateway
• ❌ Webhook handlers missing
• ❌ Transaction rollback not implemented

Result: PARTIAL (部分完了) - Integration incomplete

Example 3: Production Readiness

AI says: "The feature is finished and ready."

You ask: "Ist das wirklich vollständig? Produktionsbereit und getestet?" (Is this really complete? Production-ready and tested?)

Validation reveals:

• ✅ All code implemented
• ✅ Unit tests passing
• ✅ Integration tests passing
• ❌ No monitoring configured
• ❌ No error alerting setup
• ❌ Performance not benchmarked

Result: FUNCTIONAL BUT NOT PRODUCTION-READY (機能完了、本番未完了)

Anti-Patterns to Detect

The "99% Complete" Trap

• Code exists but isn't wired up
• Functions implemented but not called
• Tests written but not run
• "Just needs deployment" (but deployment isn't configured)

The Mock Data Mirage

• Everything "works" with fake data
• Real data sources not connected
• External APIs not integrated
• "Works in dev" but won't work in production

The Integration Illusion

• Components exist in isolation
• No inter-component communication
• "Ready to integrate" (but not actually integrated)
• Missing configuration for connections

The Spec Gap

• Code written but doesn't match requirements
• Features missing from specification
• "Creative interpretation" of requirements
• Edge cases not handled

Notes

Research Validation

This skill's approach is grounded in peer-reviewed research:

1. Multilingual prompting effectiveness validated by arXiv 2025 research showing language alignment helps prevent LLM hallucination
2. Production Readiness Reviews (PRR) used by 98% of engineering leaders (Cortex.io 2024 State of Production Readiness Report)
3. AST-based static analysis is industry standard (AOSA Book "500 Lines or Less")

Limitations

Not a silver bullet - This skill significantly improves completion detection but cannot guarantee 100% accuracy:

• AST analysis catches empty functions but not logic errors
• Runtime tests verify basic connectivity but not all edge cases
• PRR gates check configuration but require human judgment for compliance
• Multilingual prompting helps prevent hallucination but doesn't eliminate it

Confidence levels by check type:

• High confidence (≥90%): Empty functions, missing imports, TypeScript/Python type errors, missing files
• Medium confidence (60-80%): Env vars exist (but may not work), health endpoint responds (but may be decoy)
• Low confidence (≤40%): Test output says "passing" (tests may be trivial), library exists in package.json (but may be unused)

Verification Recommendations:

1. For critical systems: Add behavioral validation (Testcontainers, real API tests)
2. For production deployment: Require human-led Production Readiness Review
3. For high-risk components: Manual code review focusing on logic and security
4. For legacy/system integration: Actual deployment to staging environment

Best used as:

1. Automated first-pass validation in CI/CD
2. Educational tool to understand completion gaps
3. Conversation starter with AI using multilingual prompts
4. Checklist to prompt human review (not replace it)

Completion is Binary

Something is either complete or not. "Mostly done" means not done.

• 99% complete = NOT COMPLETE (未完成 / Non Completus)
• Mock data = Technical debt that still needs to be done
• Integration = Mandatory for completion
• Tests with mocks ≠ System actually works
• Documentation = Part of done, not after-the-fact

When to Use Each Tier

Tier 1 (AST Analysis) - During development:

• After writing code, before commit
• In CI/CD pre-commit hooks
• Automated in pull requests

Tier 2 (Runtime) - Before merge:

• After integration complete
• Before requesting review
• In CI/CD pipeline

Tier 3 (PRR) - Before deployment:

• Before production release
• During sprint planning
• For compliance audits

Related Skills

• @testing-e2e - Verify end-to-end workflows
• @testing-integration - Verify component connections
• @gap-analysis - Identify missing requirements
• @spec-gap-analysis - Verify spec compliance
• @trust-but-verify - Independent validation

References

• See references/README.md for completion methodology
• See references/multilingual-completion.md for linguistic nuances
• See references/anti-patterns.md for common traps