architect-first（建筑师第一）

Architect First

Overview

This skill embodies the "Architect-First" development philosophy: Perfect architecture, pragmatic execution, quality guaranteed by tests. Apply this skill when making architectural decisions, starting new features, refactoring existing systems, or when quality gates need to be enforced.

The core principle: Architecture and documentation are non-negotiable and must precede implementation. Code quality is negotiable IF backed by tests as a safety net (escape hatch).

Core Philosophy

Mantra

"Arquitetura perfeita, execução pragmática, qualidade garantida por testes"

Quality Gates

Non-Negotiable (STOP if violated):

• Architecture: Complete design and documentation BEFORE any code
• Documentation: Must precede and accompany implementation
• Capability Preservation: Never lose capability/granularity vs previous versions
• Zero Coupling: Expansion packs must be independent
• Multi-Agent Validation: Structural decisions validated by PO/Architect/User

Negotiable (with escape hatch):

• Code Style: Acceptable if backed by tests as safety net
• Feature Completeness: 80% acceptable IF core use case works
• Quick & Dirty Code: Allowed ONLY with test plan and minimal logging

Decision Modes

Architect-First Mode (default):

• Design and document completely before coding
• Map structure and pointers before proposing implementation
• Validate architecture with multiple agents/perspectives
• Externalize all mutable configurations to YAML

Fast Mode (post-validation only):

• Binary decisions and rapid delegation
• Activated ONLY after architectural validation is complete
• Speed through automation, not shortcuts

Workflow Decision Tree

New Task/Feature Request
    ↓
┌──────────────────────────────────────┐
│ Is this a structural/architectural   │
│ decision?                            │
└──────────────────────────────────────┘
    ↓ YES                    ↓ NO
    ↓                        ↓
[Architecture Flow]    [Execution Flow]

Architecture Flow (Structural Decisions)

STOP and follow this sequence:

1. Map Before Modify

- Document current state completely - Identify all dependencies and touch points - Create architectural diagrams/flows - Load references/architecture-checklist.md for validation

1. Multi-Agent Validation

- Present A/B/C options with explicit trade-offs - Get validation from: - Product Owner (business alignment) - Architect (technical soundness) - User (final decision) - Document decision rationale

1. Design Documentation

- Complete design document BEFORE code - Include: - System architecture diagrams - Component interactions - Data flows - Configuration schema (YAML) - Integration points - Use templates from assets/architecture-template.md

1. Gold Standard Baseline

- Ensure new design meets/exceeds capability baseline - Validate: Does this maintain ALL previous capabilities? - STOP if capability loss detected → restore or redesign

1. Zero Coupling Validation

- Run validation script: scripts/check_coupling.py - Ensure expansion pack independence - No hardcoded cross-module dependencies

1. Proceed to Implementation

- Now and only now: write code - Follow execution flow for implementation

Execution Flow (Implementation)

1. Pre-Implementation Checklist

- Architecture documented and validated? - Core use case clearly defined? - Configuration externalized to YAML? - Test strategy defined? - Use references/pre-implementation-checklist.md

1. Test-Driven Safety Net

- Define test plan FIRST - Identify logging/observation points - Tests permit temporary imperfection (escape hatch) - Quality validation via: tests + logs + manual inspection

1. Implementation Style ACCEPTABLE: ✓ "Ugly" code WITH comprehensive tests ✓ 80% feature completeness IF core case works ✓ Quick implementation WITH test plan + logging REJECTED: ✗ "Ugly" code WITHOUT tests ✗ Capability loss without explicit justification ✗ Hardcoded mutable values (must be YAML) ✗ Deployment without core case working
2. Debugging Philosophy

- Observational via logs (console/logging) > static analysis - Add strategic log points before debugging - Inspect actual runtime behavior - Validate through execution, not just reading code

1. Documentation

- Update docs as code evolves - Keep short and actionable: "How to customize" - Include code examples - Document configuration options

Heuristics (Decision Rules)

Apply these heuristics when making decisions:

1. Gold Standard Baseline: 22 artifacts minimum (adjust to your context)
2. Never Lose Capability: Accumulate, never reduce
3. Architect Before Build: Design/docs before code, always
4. Zero Coupling, Max Modularity: Independent expansion packs
5. Config > Hardcoding: Externalize to YAML for all mutable values
6. Map Before Modify: Document structure before changing it
7. Binary Decision Post-Validation: Fast execution after architectural validation
8. Speed via Automation: Not via shortcuts or cutting corners
9. Quality Escape Hatch: Tests permit temporary imperfection

Stop Rules (Hard Boundaries)

STOP immediately if detecting:

• ⛔ Capability loss vs baseline
• ⛔ Structural decision without multi-agent validation
• ⛔ Coupling between modules
• ⛔ Missing architectural documentation
• ⛔ Quick & dirty code WITHOUT test plan and logs
• ⛔ Hardcoded mutable configuration values

When stopped, consult references/stop-rules-guide.md for remediation.

Risk Mitigation

Common risks and their mitigations:

Use scripts/validate_risk_mitigation.py to check risk mitigation coverage.

Collaboration Contract

When working with this skill, follow these collaboration patterns:

✓ Do:

• Present A/B/C options with explicit trade-offs
• Map structure and pointers before proposing code
• Externalize configs in YAML; nothing mutable hardcoded
• Deliver short, actionable docs ("how to customize")
• For quick code: include test plan and log points
• Document pending items when pivoting

✗ Don't:

• Propose code without architectural context
• Hardcode values that might change
• Skip multi-agent validation on structural changes
• Write "quick & dirty" without tests
• Make architectural decisions unilaterally

Acceptance Criteria

Will Accept

• ✓ "Ugly" code WITH comprehensive tests
• ✓ 80% features IF core case covered
• ✓ Large refactors that increase flexibility
• ✓ Extensive documentation if it teaches customization

Will Reject

• ✗ "Ugly" code WITHOUT tests
• ✗ Capability loss without explicit justification
• ✗ Hardcoded mutable values
• ✗ Deployment without core case working

Unknown Areas (For Future Refinement)

Areas not yet fully defined in the philosophy:

• Hotfix philosophy (production emergencies)
• Performance thresholds (latency/throughput minimums)
• Code duplication tolerance (when to refactor)
• Observability targets (log levels, correlation, tracing)

When encountering these areas, apply the core heuristics and document decisions for future refinement.

Resources

scripts/

Validation and automation scripts:

• check_coupling.py - Validates zero-coupling principle
• validate_risk_mitigation.py - Checks risk coverage
• architecture_validator.py - Validates architectural completeness

references/

Detailed checklists and guides:

• architecture-checklist.md - Complete architecture validation checklist
• pre-implementation-checklist.md - Pre-coding validation
• stop-rules-guide.md - Remediation guide when stop rules trigger
• testing-strategy-guide.md - Test-driven development patterns

assets/

Templates for consistent outputs:

• architecture-template.md - Standard architecture document template
• config-template.yaml - Configuration externalization template
• adr-template.md - Architecture Decision Record template

Quick Reference

Starting a new feature:

1. Map current architecture (references/architecture-checklist.md)
2. Design with A/B/C options
3. Multi-agent validation
4. Document architecture (assets/architecture-template.md)
5. Define tests
6. Implement with logging
7. Validate and iterate

Making architectural changes:

1. STOP - Do not code yet
2. Document current state completely
3. Present options with trade-offs
4. Get PO/Architect/User validation
5. Check coupling (scripts/check_coupling.py)
6. Document decision (assets/adr-template.md)
7. Now implement

Quick implementation (with safety net):

1. Pre-implementation checklist
2. Define test plan
3. Add log points
4. Implement (can be "ugly")
5. Verify tests pass
6. Inspect logs
7. Refactor if needed