docs: GOVERNANCE LOCK — AGENTS.md is the ONLY source of guidelines

Establish unambiguous governance:
- AGENTS.md v16.0 is the ONLY document that contains engineering guidelines
- All procedures, harnesses, rules, decision frameworks live in AGENTS.md
- CLAUDE.md, GEMINI.md, and all other documents FOLLOW AGENTS.md
- Other documents ONLY reference AGENTS.md with explicit links
- If any document conflicts with AGENTS.md, AGENTS.md is authoritative

Changes:
- AGENTS.md: Add 'GOVERNANCE LOCK' section at top (5 rules, scope definition)
- CLAUDE.md: Add critical warning (context only, not guidelines)
- Rules: Never add procedures to supplementary documents

Non-negotiable enforcement:
- New guidelines → AGENTS.md only
- Found guidelines elsewhere → move to AGENTS.md, replace with reference
- Conflicting guidance → AGENTS.md wins
- Exception: Project status, architecture context, navigation (CLAUDE.md only)

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
This commit is contained in:
2026-08-15 10:55:10 +09:00
parent d090538b17
commit 9d541a5982
2 changed files with 35 additions and 12 deletions
+18 -6
View File
@@ -1,13 +1,25 @@
# CLAUDE.md
This file provides project context and architecture for Claude Code (claude.ai/code).
**⚠️ CRITICAL: This file is for PROJECT CONTEXT ONLY. It does NOT contain engineering guidelines.**
**IMPORTANT: All engineering decisions are governed by AGENTS.md v16.0. This file supplements AGENTS.md but DOES NOT override it.**
**All engineering guidelines, rules, harnesses, and procedures are in AGENTS.md v16.0 ONLY.**
**Single Source of Truth (Hierarchy):**
1. **AGENTS.md** — AI coding constitution, principles, rules (authoritative)
2. **CLAUDE.md** — Project context, status, architecture (supplementary)
3. **Project code** — Implementation (must comply with AGENTS.md)
If any section below conflicts with AGENTS.md, AGENTS.md is authoritative and this text is invalid.
**What belongs in AGENTS.md:**
- Coding principles and rules
- Development configuration (database, ports, authentication, SSH)
- Procedures and workflows
- Decision frameworks
- Anti-patterns and guardrails
**What belongs in CLAUDE.md:**
- Project status and timeline
- Architecture overview (high-level only)
- File structure and navigation
- References to AGENTS.md (with explicit links)
**When working: Always check AGENTS.md first. CLAUDE.md is supplementary context only.**
## ⚖️ Governance: AGENTS.md v16.0 Strategic Principles