EDAF v1.0 - Claude Code Configuration

Language Preferences

This file is auto-generated by /setup command. Do not edit manually - run /setup again to change preferences.

Current Settings:

• Documentation Language: English
• Terminal Output Language: English
• Save Dual Language Docs: No

EDAF 7-Phase Gate System

When implementing features, fixing bugs, or making changes, automatically follow this workflow:

Triggered by natural language requests for implementation work (no need to say "EDAF") Detailed workflows: .claude/skills/edaf-orchestration/PHASE{1-7}-*.md

Quick Reference

EDAF Execution Pattern

For each phase:

1. Execute → Run agent/worker to generate artifact
2. Evaluate → Run ALL evaluators in parallel (use Task tool)
3. Check results:
   • ✅ ALL pass (≥ threshold) → Proceed to next phase
   • ❌ ANY fail (< threshold) → Feedback loop:
     1. Read evaluator reports for specific feedback
     2. Revise artifact based on feedback
     3. Re-run ALL evaluators (not just failed ones)
     4. Repeat until ALL pass (unlimited iterations)

This feedback loop is EDAF's core quality mechanism.

Artifacts by Phase:

• Phase 1: .steering/{date}-{feature}/idea.md (requirements)
• Phase 2: .steering/{date}-{feature}/design.md (technical design)
• Phase 3: .steering/{date}-{feature}/tasks.md (task plan)
• Phase 4: Source code (implementation)
• Phase 5: .steering/{date}-{feature}/reports/ (evaluation reports)
• Phase 6: docs/ (permanent documentation updates)
• Phase 7: Deployment configs

Permanent Documentation (docs/):

• product-requirements.md, functional-design.md, development-guidelines.md
• repository-structure.md, architecture.md, glossary.md

Critical Rules

1. NEVER skip phases
2. ALWAYS run evaluators in parallel (use Task tool)
3. ALWAYS iterate until ALL evaluators pass (no exceptions)
4. IF any evaluator fails:
   • Read evaluator report for specific feedback
   • Revise artifact based on feedback
   • Re-run ALL evaluators (not just failed ones)
   • Repeat until ALL pass (unlimited iterations)
5. Phase 1 is mandatory for new features (requirements gathering)
6. Phase 4 quality-gate is ultra-strict (10.0 = zero lint errors/warnings + all tests pass)
7. UI verification required if frontend modified (Phase 5)

Component Discovery

All components are auto-discovered from file system. No manual listing needed.

Locations:

• Agents: .claude/agents/*.md + .claude/agents/workers/*.md
• Evaluators: .claude/agents/evaluators/phase{1-7}-*/*.md
• Skills: .claude/skills/*/SKILL.md (coding standards, workflows, user-invocable commands like /setup, /review-standards)
• Config: .claude/edaf-config.yml, .claude/agent-models.yml

Component Count:

• 9 Agents (requirements-gatherer, designer, planner, 4 workers, documentation-worker, ui-verification-worker)
• 40 Evaluators (7 per phase for phases 1-3,6; 8 for phase 5; 1 for phase 4; 5 for phase 7)
• Total: 49 components

Phase 5 Note: Includes standards-compliance-evaluator as an independent gate for project coding standards.

Instructions for Claude Code

Terminal Output Language

Respond in ENGLISH for all output.

Documentation Language

Generate documentation in ENGLISH.

Agent Behavior

• Workers: Follow project coding standards in .claude/skills/
• Evaluators: Output in terminal language, generate reports in documentation language
• All agents: Read detailed phase instructions in .claude/skills/edaf-orchestration/

Setup

For initial project setup, see README.md for /setup command instructions.

Last Updated: Auto-generated by /setup command Configuration: .claude/edaf-config.yml