https://github.com/hiroyuki-miyauchi/axiarch
Constitution-driven governance framework for AI agents. 2,500+ universal standards, self-evolving knowledge base, and AI behavior protocols. Validated on Google Antigravity.
https://github.com/hiroyuki-miyauchi/axiarch
agents agents-md ai-coding ai-governance antigravity claude-code code-quality constitution-driven cursor developer-tools github-copilot llms-txt prompt-engineering specification-driven-development vibe-coding windsurf
Last synced: about 2 months ago
JSON representation
Constitution-driven governance framework for AI agents. 2,500+ universal standards, self-evolving knowledge base, and AI behavior protocols. Validated on Google Antigravity.
- Host: GitHub
- URL: https://github.com/hiroyuki-miyauchi/axiarch
- Owner: hiroyuki-miyauchi
- License: apache-2.0
- Created: 2025-11-23T08:32:44.000Z (8 months ago)
- Default Branch: main
- Last Pushed: 2026-05-11T00:08:55.000Z (3 months ago)
- Last Synced: 2026-05-11T02:59:25.332Z (3 months ago)
- Topics: agents, agents-md, ai-coding, ai-governance, antigravity, claude-code, code-quality, constitution-driven, cursor, developer-tools, github-copilot, llms-txt, prompt-engineering, specification-driven-development, vibe-coding, windsurf
- Language: Shell
- Homepage: https://chronoviq.com
- Size: 13.2 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Changelog: CHANGELOG.md
- Contributing: CONTRIBUTING.md
- License: LICENSE
- Code of conduct: CODE_OF_CONDUCT.md
- Codeowners: .github/CODEOWNERS
- Security: SECURITY.md
- Roadmap: ROADMAP.md
- Notice: NOTICE
- Agents: AGENTS.md
Awesome Lists containing this project
README
# ð° Axiarch
## æ²æ³é§åå AIãšãŒãžã§ã³ãã¬ããã³ã¹ãã¬ãŒã ã¯ãŒã¯
## Constitution-Driven AI Agent Governance Framework
[](https://github.com/hiroyuki-miyauchi/axiarch/releases)
[](https://opensource.org/licenses/Apache-2.0)
[](#-universal-rules-39-files--jaen)
[](#-忢±å
容--whats-included)
[](#-aiãšãŒãžã§ã³ãäºææ§--ai-agent-compatibility) [](#-aiãšãŒãžã§ã³ãäºææ§--ai-agent-compatibility)
[æ¥æ¬èª](#-axiarchã¢ã¯ã·ã¢ãŒã¯ãšã¯) ã» [English](#-what-is-axiarch-ax-ee-ark)
---
## ð¯ðµ AxiarchïŒã¢ã¯ã·ã¢ãŒã¯ïŒãšã¯
**AxiarchïŒã¢ã¯ã·ã¢ãŒã¯ïŒ** ã¯ã**æ²æ³é§ååã® AIãšãŒãžã§ã³ãã¬ããã³ã¹ãã¬ãŒã ã¯ãŒã¯ïŒConstitution-Driven AI Agent Governance FrameworkïŒ**ã§ãã
ãæ®éæ²æ³ïŒUniversalã»äžå€ïŒããšãåºæã«ãŒã«ïŒBlueprintã»å¯å€ïŒãã®æç¢ºãªè²¬ååé¢ãããã«ãããå®è¡é§åãããããã³ããïŒPromptsã»ä»»æå±€ïŒããšãã **3局統åã¬ããã³ã¹ã»ã¢ãŒããã¯ãã£** ããã Axiarch ã®äžæ žã§ããAIæ¯æŽéçºã«ããããã«ã·ããŒã·ã§ã³ãå質ããªããïŒéè¡ïŒã®ãªã¹ã¯ããã®æ§é ã«ãã£ãŠè»œæžããæçžŠè
ã®ã¹ãã«ã¬ãã«ã«äŸåããããããžã§ã¯ãå
šäœã®æäœå質ïŒQuality FloorïŒãå匷ãåºäžãããŸãã
[Google Antigravity](https://antigravity.google/) äžã§èšèšã»å®æŠæ€èšŒæžã¿ãã«ãŒã«æ¬äœã¯çŽç²ãª Markdown ã§ããã`AGENTS.md` ã¯äž»èŠãªã³ãŒãã£ã³ã°ãšãŒãžã§ã³ããåç
§ã»å¯Ÿå¿ãããªãŒãã³ãã©ãŒãããã®ããã**ä»ã® AI ãšãŒãžã§ã³ãïŒOpenAI CodexãCursorãClaude CodeãGitHub Copilot çïŒãšãäºææ§ãããèŠèŸŒã¿**ã§ã â ãã ãåäœã¯æªæ€èšŒã§ãã
### èšèšææ³
- **ããã³ããéã§ã¯ãããŸããã** **æçžŠè
ãå€ãã£ãŠãæäœå質ã®åºïŒQuality FloorïŒãå匷ãæ
ä¿ãã**ãå€å±€é²åŸ¡ã®ã¬ããã³ã¹èšèšã§ãã
- **æçžŠè
ã«äŸåããªãå質ããŒã¹ã©ã€ã³ã®æ»å®ã** ã·ãã¢ãšã³ãžãã¢ã䜿ã£ãŠãåå¿è
ã䜿ã£ãŠããæ²æ³ãåäžã®æ²æ³åºæºã匷å¶ããŸãã
- **3å±€ã¬ããã³ã¹åé¢ã** ã絶察äžå€ã®æ²æ³åºæºïŒUniversalïŒããšããããžã§ã¯ãã§æé·ããåç仿§ïŒBlueprintïŒãã®è²¬åãç©çå±€ã§åé¢ããAIã®ãã³ã³ããã¹ãå¿åŽãããã«ãŒã«åœ¢éªžåãã®ãªã¹ã¯ãæ§é çã«æã蟌ã¿ãŸãã
### ãªãå¿
èŠã
| ã¬ããã³ã¹ãªã | Axiarchãã |
|:-------------|:--------------------------|
| AIããããã£ãœããå£ããŠãããã³ãŒããçæ | Zero Bug, Zero Warning ããªã·ãŒãæ²æ³ã§åŒ·å¶ |
| å質ãæçžŠè
ã®ã¹ãã«ã«å®å
šäŸå | Universal Rulesãæäœå質ã®åºãèšå® |
| 仿§ãªãã® vibe coding | Blueprint FirstïŒã³ãŒãããå
ã«ä»æ§ãå®çŸ© |
| ã»ãã·ã§ã³éã§ç¥èŠãæ¶å€± | æèšãã°ãç¥èŠãã«ãŒã«ã«çµæ¶å |
| AIããããžã§ã¯ãæ§é ãå¹»èŠã§çæ | Boot Sequence ProtocolïŒãŸãã«ãŒã«ãèªã |
### ã¢ãŒããã¯ãã£
```text
âââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ
â 第1å±€: Universal (æ®éæ²æ³ / Immutable) â
â ââ AGENTS.md (æé«æ³èŠ / è¡åãããã³ã«) â
â ââ Universal Rules (39ãã¡ã€ã«: äžå€ã®æ®éçåºæº) â
âââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ€
â 第2å±€: Blueprint (åºæä»æ§ / Mutable) â
â ââ ãããžã§ã¯ãæŠèŠãæ©èœä»æ§ãæèšãã°çã®å¯å€ã«ãŒã« ãâ
âââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ€
â 第3å±€: Prompts (å®è¡ãšã³ãžã³ / Optional) âã
â ââ ç£æ»ã»å質æ
ä¿ã¿ã¹ã¯çšã®ä»»æããã³ããã»ãã¬ãŒã ã¯ãŒã¯ â
âââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ€
â å®è¡ããã¥ã¡ã³ã â ã¿ã¹ã¯åäœ â çæ â
â task.md, implementation_plan.md, walkthrough.md â
âââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ€
â ãã£ãŒãããã¯ã«ãŒã â çµæ¶å â ç¶ç¶ç â
â æèš â ã«ãŒã«(第2å±€) â ããè¯ãæèš â
âââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ
```
---
## ðºðž What is Axiarch (AX-ee-ark)?
**Axiarch** is a **Constitution-Driven AI Agent Governance Framework**.
It is designed to strictly govern and mitigate quality drift, hallucinations, and uncontrolled AI behavior in production development through a **Three-Layer Governance Architecture**: Layer 1 **Universal** (Immutable Constitution), Layer 2 **Blueprint** (Mutable Project State), and Layer 3 **Prompts** (Optional Execution Triggers).
Designed and validated through hundreds of real production sessions on [Google Antigravity](https://antigravity.google/). Since all rules are pure Markdown and `AGENTS.md` is an open format referenced by major coding agents, **it is expected to be compatible with other AI agents** (OpenAI Codex, Cursor, Claude Code, GitHub Copilot, etc.) â though this has not been verified.
### Core Design Philosophy
- **Not a prompt collection.** It is a **multi-layered governance architecture** that safeguards minimum quality standards **regardless of who operates the AI agent.**
- **Operator-independent quality baseline.** Whether a senior engineer or a beginner uses the AI, the constitution enforces the same constitution standards.
- **Three-layer separation.** By physically decoupling "Immutable Universal standards" from "Mutable Blueprint states", Axiarch structurally mitigates 'context amnesia' and rule atrophy.
### The Problem
| Without Governance | With Axiarch |
|:-------------------|:--------------------------|
| AI generates code that "looks right" but breaks silently | Constitution enforces Zero Bug, Zero Warning policy |
| Quality depends entirely on the operator's skill | Universal Rules set a minimum quality floor |
| No structure for specs â vibe coding | Blueprint First: specs before code, always |
| Lessons learned are lost between sessions | Lessons Log crystallizes knowledge into rules |
| AI hallucinates project structure and rules | Boot Sequence Protocol: read rules first, then act |
### Architecture
```text
âââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ
â Layer 1: Universal (Immutable Constitution) â
â ââ AGENTS.md (Supreme Law / Behavior Protocols) â
â ââ Universal Rules (39 files / Immutable Universal Standards) â
âââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ€
â Layer 2: Blueprint (Mutable Project State) â
â ââ Project Overview, Feature Specs, Lessons Log â
âââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ€
â Layer 3: Prompts (Optional Execution Framework) â
â ââ Task-specific prompts for auditing and QA execution â
âââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ€
â Execution Documents â Per-Task â Gen. â
â task.md, implementation_plan.md, walkthrough.md â
âââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ€
â Feedback Loop â Crystallization â Cont. â
â Lessons â Rules(Layer 2) â Better Lessons â
âââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ
```
---
## ð AIãšãŒãžã§ã³ãäºææ§ / AI Agent Compatibility
| ç¶æ
/ Status | Agent | Native Config | AGENTS.md |
|:--------------|:------|:-------------|:----------|
| â
**Verified** â å®åã§å®èšŒæžã¿ | **Google Antigravity** | `.agents/rules/` | â
Reads |
| â ïž **Untested** â æªæ€èšŒïŒåäœããèŠèŸŒã¿ïŒ | **OpenAI Codex** | `AGENTS.md` (native) | â
Native |
| â ïž **Untested** â æªæ€èšŒïŒåäœããèŠèŸŒã¿ïŒ | **Cursor** | `.cursor/rules/*.mdc` | â
Reads |
| â ïž **Untested** â æªæ€èšŒïŒåäœããèŠèŸŒã¿ïŒ | **GitHub Copilot** | `.github/copilot-instructions.md` | â
Reads |
| â ïž **Untested** â æªæ€èšŒïŒåäœããèŠèŸŒã¿ïŒ | **Claude Code** | `CLAUDE.md` | â
Reads |
| â ïž **Untested** â æªæ€èšŒïŒåäœããèŠèŸŒã¿ïŒ | **Windsurf** | `.windsurfrules` | â
Reads |
| â ïž **Untested** â æªæ€èšŒïŒåäœããèŠèŸŒã¿ïŒ | **Aider / Zed / Other** | Various | â
Reads |
> [!NOTE]
> **JA**: åAIãšãŒãžã§ã³ãã«ã¯**åºæã®èšå®ãã£ã¬ã¯ããª**ããããŸãïŒäŸ: Cursor㯠`.cursor/rules/`ãCopilot㯠`.github/copilot-instructions.md`ïŒã`AGENTS.md` ã¯**äž»èŠãšãŒãžã§ã³ããèªããå
±éæ²æ³**ã§ãããåããŒã«ã®ãã€ãã£ãèšå®ïŒäŸïŒ`.cursor/rules/`ã`.github/copilot-instructions.md`ïŒãš**䜵çš**ãããŸããããŒã«åºæã®ã»ããã¢ããã®ä»£æ¿ã§ã¯ãããŸãããOpenAI Codex㯠`AGENTS.md` èªäœããã€ãã£ãèšå®ãšããŠæ¡çšããŠããããã远å ã®ãã€ã³ã¿ãŒãã¡ã€ã«ã¯äžèŠã§ãã
>
> **EN**: Each AI agent has its **own native configuration directory** (e.g., `.cursor/rules/` for Cursor, `.github/copilot-instructions.md` for Copilot). `AGENTS.md` is a **shared constitution readable by major agents**, designed to complement each tool's native config (e.g., `.cursor/rules/`, `.github/copilot-instructions.md`). It is NOT a replacement for tool-specific setup. OpenAI Codex natively uses `AGENTS.md` as its configuration format, so no additional pointer file is needed.
> [!IMPORTANT]
> **JA**: æ¬ã¬ããã³ã¹ã»ã¢ãŒããã¯ãã£ã¯ **[Google Antigravity](https://antigravity.google/)** äžã§èšèšã»å®æŠæ€èšŒããããã®ã§ããã«ãŒã«æ¬äœã¯çŽç²ãªMarkdownã§ãããAIã¢ãã«ã®æŠå¿µçãªä»çµã¿ã¯ããŒã«éã§å
±éã§ããããä»ã®ãšãŒãžã§ã³ãã§ãåäœããèŠèŸŒã¿ã§ããã**ä»ããŒã«ã§ã®åäœã¯æ€èšŒããŠããŸããããå©çšã¯èªå·±è²¬ä»»ã§ãé¡ãããŸããAntigravity以å€ã®ç°å¢ã§ã®åäœä¿èšŒã¯ããããŸããã**
>
> **EN**: This governance architecture was designed and validated through hundreds of real production sessions on **Google Antigravity**. Since all rules are pure Markdown and the underlying AI model concepts are fundamentally the same across tools, they are expected to work on other agents â but **this has not been verified. Use at your own risk. No guarantees are provided for non-Antigravity environments.**
> [!TIP]
> **JA**: ãã©ã«ãå `axiarch-rules/` ã¯åºèªãåæ ããŠããŸãããã«ãŒã«ãã¡ã€ã«èªäœã¯çŽç²ãªMarkdownã§ãããããŒã«åºæã®äŸåã¯ãããŸããã
>
> **EN**: The folder name `axiarch-rules/` reflects its origin, but the rule files themselves are pure Markdown with zero tool-specific dependencies.
---
## ðР忢±å
容 / What's Included
### ðïž æé«æ³èŠ / Supreme Law
| File | JA | EN |
|:-----|:---|:---|
| `AGENTS.md` | AIè¡åæ²æ³ïŒGoogle Antigravityã§èšèšãä»ãšãŒãžã§ã³ãã§ãåäœèŠèŸŒã¿ïŒ | AI Behavior Constitution (designed on Google Antigravity, compatible with other agents) |
### ð Universal Rules (39 files à JA/EN)
> Universal Rules ã¯ãããžã§ã¯ãã§äœ¿ãå¯èœæ§ã®ããå
šæè¡é åã®ãã¹ããã©ã¯ãã£ã¹ã©ã€ãã©ãªã§ããAI㯠LOADING_PROTOCOL ã«åŸããã¿ã¹ã¯ã«å¿
èŠãªãã¡ã€ã«ã®ã¿ãéžæçã«ããŒãããŸãã䜿ããªãæè¡ã®ã«ãŒã«ãååšããŠãåé¡ãããŸãããããèªäœããå°æ¥ã®æ¡çšæãæªç¥ã®æè¡ã«çŽé¢ãããšãã®å質æ
ä¿ã®æºæ³ã§ãã
>
> Universal Rules is a comprehensive best-practice library across all major technology domains. The AI selectively loads only what each task requires, following LOADING_PROTOCOL. Rules for technologies your project doesn't currently use cause no harm â they are there when you need them.
| Category | Files | JA | EN |
|:---------|:------|:---|:---|
| Core & Mindset | 1 | éçºå²åŠãSupreme Directive | Development philosophy, Supreme Directive |
| Product & Business | 9 | ãããã¯ãæŠç¥ãåžå Žæ€èšŒãGTMãåçãäŸ¡æ Œãã°ããŒã¹ããã©ã³ããASOãIR | Product strategy, market validation, GTM, revenue, pricing, growth, brand, ASO, IR |
| Design & UX | 1 | ãã¶ã€ã³ã·ã¹ãã ãA11y | Design system, accessibility |
| Engineering | 10 | ã³ãŒãå質(80§)ãAPIãSupabaseãWebãCMSãFlutterãNativeãFirebaseãAWSã**Git Workflow** | Code quality (80§), API, Supabase, Web, CMS, Flutter, Native, Firebase, AWS, **Git Workflow** |
| AI & Data | 2 | AIãšã³ãžãã¢ãªã³ã°ãããŒã¿åæ | AI Engineering, Data Analytics |
| Operations | 7 | å
éšããŒã«ãå¶æ¥ã»BizDevãHRãCXãSREãã€ã³ã·ãã³ããããŒãããŒã·ãã | Internal tools, sales/BizDev, HR, CX, SRE, incident response, partnerships |
| Security & Legal | 4 | ã»ãã¥ãªãã£ãããŒã¿ã¬ããã³ã¹ãOSSã³ã³ãã©ã€ã¢ã³ã¹ãç¥çè²¡ç£ | Security, data governance, OSS compliance, IP |
| QA & FinOps | 2 | ãã¹ãæŠç¥ãã¯ã©ãŠãFinOps | Test strategy, Cloud FinOps |
| Global & Governance | 3 | i18nãã¬ããã³ã¹ãèšèªãããã³ã« | i18n, governance, language protocol |
### ð Blueprint (ãããžã§ã¯ãåºæãã³ãã¬ãŒã / Project-Specific Templates)
| File | JA | EN |
|:-----|:---|:---|
| `core/000_project_overview.md` | ãããžã§ã¯ãæŠèŠãã³ãã¬ãŒã | Project overview template |
| `core/010_project_lessons_log.md` | æèšãã° | Lessons log |
| `core/998_feature_spec_template.md` | **æ©èœä»æ§ãã³ãã¬ãŒãïŒBlueprint Firstã®æ žïŒ** â åãå
¥ãæ¡ä»¶(Given/When/Then)ãæ©èœåäœã§å¿
é å | **Feature spec template (Blueprint First core)** â Acceptance Criteria (Given/When/Then) required per feature |
| `core/999_project_specific_template.md` | ãããžã§ã¯ãåºæã«ãŒã«ãã³ãã¬ãŒã | Project-specific rule template |
### ð§ åºç€ / Infrastructure
| File | JA | EN |
|:-----|:---|:---|
| `LOADING_PROTOCOL.md` | 5ã¹ãããã®ã«ãŒã«ããŒãæé | 5-step rule loading protocol |
| `CRYSTALLIZATION_PROTOCOL.md` | æèšã®èªåçµæ¶åãããã³ã« | Lesson auto-crystallization protocol |
| `INDEX.md` | å
šã«ãŒã«ã®è©³çŽ°çŽ¢åŒ | Detailed index of all rules |
| `compliance_matrix.md` | èŠä»¶å¯Ÿç
§è¡š | Compliance matrix |
### ð¯ ããã³ããé / Prompt Library
> ããã³ããã¯çšéå¥ãã©ã«ãïŒ`develop/`, `audit/`, `govern/`, `operate/`ïŒã«æŽçãããŠããŸãã
> Prompts are organized into role-based folders (`develop/`, `audit/`, `govern/`, `operate/`).
**ð develop/ â éçºã»å®è¡ / Development & Execution**
| File | JA | EN |
|:-----|:---|:---|
| `develop/feature_development.md` | æ°æ©èœå®è£
ã»æ¢åæ¹ä¿®ã»ãã°ä¿®æ£ã»æ²æ³ç£æ»ãç¶²çŸ
çã«å®è¡ããããã³ãã | Comprehensive prompt for new feature implementation, improvement, bug fixing, and compliance auditing |
| `develop/refactoring_audit.md` | æ¢åã³ãŒãã®åäœãä¿ã£ããŸãŸæ§é ã»åå®å
šã»DRYååãæ¥µéãŸã§æ¹åããéç Žå£çãªãã¡ã¯ã¿ãªã³ã°ç£æ»ããã³ãã | Non-destructive refactoring audit â elevate structure, type safety, and DRY principles without changing existing behavior |
| `develop/push_execute.md` | å質ã²ãŒãã»DBæŽåæ§ç¢ºèªã»ãã©ã³ãæŠç¥éµå®ãçµãGit Pushå®è¡ããã³ãã | Quality gate, DB integrity check, branch strategy compliance, and Atomic Push execution |
| `develop/ci_fix.md` | CI/CD倱ææã®ãšã©ãŒåçŸã»æ ¹æ¬åå åæã»ä¿®æ£ã»ã«ãŒã«éå
ãäžè²«å®è¡ããããã³ãã | CI/CD failure error reproduction, root cause analysis, fix, and rule feedback |
**ð audit/ â åè³ªã»æŽåæ§ç£æ» / Quality & Integrity Auditing**
| File | JA | EN |
|:-----|:---|:---|
| `audit/fullstack_qa_audit.md` | ã»ãã¥ãªãã£ã»ãã©ã€ãã·ãŒæå€§åãçé ã«ã·ãªã³ã³ãã¬ãŒåºæº6æ±ã»åªå
床ä»ãå ±åïŒCritical/High/MediumïŒã»ROIææ¡ã»Domain Distributionã«ããç¥èŠéå
ãå«ãçµ±åç£æ»ããã³ãã | Full-Stack QA & Strategic audit â 6-Pillar Silicon Valley Standard with priority reporting, ROI proposals, Domain Distribution knowledge feedback |
| `audit/api_architecture_audit.md` | APIèšèšã»DTO矩åã»ãŒããã©ã¹ãã»ãªã ããã£ãã«å¯Ÿå¿ã軞ãšããå
šæ¹äœæ§é ç£æ»ããã³ãã | Omni-directional structural audit â API design, DTO obligations, Zero Trust, and omnichannel readiness |
| `audit/data_integrity_audit.md` | JSONéããã»Hybrid Syncã»Split Brainã»ããªããå®è£
ãæ€ç¥ããããŒã¿æŽåæ§ç£æ»ããã³ãã | Data integrity audit â JSON dump detection, Hybrid Sync / Split Brain elimination, and facade detection |
| `audit/system_integrity_audit.md` | åå®å
šæ§ã»API/DBåæã»ããªããæ€ç¥ã»ããŒã¿ããã¿ã€ãºæŠç¥ã軞ãšããã·ã¹ãã å
šäœæŽåæ§ç£æ»ããã³ãã | System integrity audit â type safety, API/DB sync, facade detection, and data monetization readiness |
| `audit/deep_optimization_audit.md` | ã¡ãã£ã¢/LCP/SSRæé©åæŒãã®æ ¹æ¬åå ç¹å®ã»è§£æ¶ã軞ãšããã·ã¹ãã å
šäœã®æ·±å±€æé©åç£æ»ããã³ãã | Deep optimization audit â media/LCP/SSR gap root cause detection, elimination, and full-system integrity |
**âïž govern/ â ã³ã³ãã©ã€ã¢ã³ã¹ã»ã¬ããã³ã¹ / Compliance & Governance**
| File | JA | EN |
|:-----|:---|:---|
| `govern/compliance_inspector_audit.md` | Universal/Blueprintæ³ãžã®å®å
šæºæ ã匷å¶ãã8ã€ã®éå€§æ²æ³éåãã¬ãŒã ã¯ãŒã¯ã«åºã¥ã深局ã³ã³ãã©ã€ã¢ã³ã¹ç£æ»ããã³ãã | Deep constitutional compliance audit â 8 Major Constitutional Violations framework |
| `govern/constitution_compliance_audit.md` | 7ã€ã®éå€§æ²æ³éåïŒã¢ãŒããã¯ãã£ã»åçåçµ±åã»åå®å
šã»æé©åã»ããªããã»æ ¹æ¬åå ïŒã軞ãšããæ·±å±€æ²æ³éµå®ã¹ãã£ã³ããã³ãã | Constitutional compliance scan â 7 Major Violations framework |
| `govern/governance_auditor.md` | 8ã€ã®æ±ïŒSecurity/Business/Legal/AI/Architecture/ä¿å®æ§/UX/PerformanceïŒã§è¡ãå
šæ¹äœã¬ããã³ã¹ç£æ»ããã³ãã | Holistic governance audit â 8-Pillar framework with structured report format |
| `govern/blueprint_governance_audit.md` | éçºç¥èŠãBlueprintã«ãŒã«ã«çµæ¶åããç¶²çŸ
çç£æ»ããã³ãã | Comprehensive audit prompt to crystallize development insights into Blueprint rules |
| `govern/localization_audit.md` | Lazy Englishæé€ã»UIæ¥æ¬èªåç100%ã»LTVã»AI/GEOã»æ³åã®å
šæ¹äœããŒã«ãªãŒãŒã·ã§ã³ç£æ»ããã³ãã | Localization audit â eliminating Lazy non-English UI text, achieving 100% English UI |
**ð¡ïž operate/ â ã€ã³ã·ãã³ãã»åå
¥ / Incident Response & Onboarding**
| File | JA | EN |
|:-----|:---|:---|
| `operate/onboarding_audit.md` | æ°ã»ãã·ã§ã³/ã¡ã³ããŒåå æã«ã³ãŒãããŒã¹ãæ·±ãçè§£ãã¢ãŒããã¯ãã£ã»å°é·ã»æåã®ã¢ã¯ã·ã§ã³ãææ¡ããåå
¥ç£æ»ããã³ãã | Codebase onboarding audit â deeply understand architecture, landmines, and first actions |
| `operate/incident_response.md` | æ¬çªé害ã®ããªã¢ãŒãžã»5 Whysæ ¹æ¬åå åæã»ç·æ¥ä¿®æ£ã»ãã¹ãã¢ãŒãã ã»åçºé²æ¢ã«ãŒã«éå
ãŸã§äžè²«å®è¡ããSREå°çšããã³ãã | SRE-focused prompt â triage, 5 Whys RCA, emergency fix, post-mortem, and recurrence-prevention rule crystallization |
> [!TIP]
> **JA**: `axiarch-prompts/` ã¯ã«ãŒã«ãšã¯ç¬ç«ãã**åå©çšå¯èœãªããã³ãããã³ãã¬ãŒãé**ã§ããAIãšãŒãžã§ã³ãã«ç¹å®ã®é«å質ã¿ã¹ã¯ãå®è¡ãããéã«äœ¿çšããŸãã
>
> **EN**: `axiarch-prompts/` is a **reusable prompt template library** independent of rules. Use them to instruct AI agents to execute specific high-quality tasks.
---
## â¡ ã¯ã€ãã¯ã¹ã¿ãŒã / Quick Start
### å¿
é ãã¡ã€ã«äžèЧ / Required Files
> [!TIP]
> **JA**: ãããžã§ã¯ãã«ã³ããŒããã®ã¯ **2ã€ã ã**ã§ãããªããžããªå
ã®ãã®ä»ã®ãã¡ã€ã«ïŒ`CHANGELOG.md`, `CONTRIBUTING.md` çïŒã¯**ãã®ãªããžããªèªäœã®ç®¡ççš**ã§ãããããªãã®ãããžã§ã¯ãã«ã¯äžèŠã§ãã
>
> **EN**: You only need to copy **2 items** to your project. Other files in this repository (`CHANGELOG.md`, `CONTRIBUTING.md`, etc.) are for **managing this repository itself** and are NOT needed in your project.
| ãã¡ã€ã« / File | å¿
é ïŒ / Required? | 説æ / Description |
|:---------------|:-------------------|:-------------------|
| `AGENTS.md` | â
**å¿
é ** / **Required** | AIè¡åæ²æ³ãå
šãšãŒãžã§ã³ãå
±é / AI constitution. Universal for all agents |
| `axiarch-rules/` | â
**å¿
é ** / **Required** | ã«ãŒã«æ¬äœïŒUniversal + BlueprintïŒ / Rule definitions |
| `.agents/rules/prompt_pointer.md` | ð¶ **Antigravity ã®ã¿** / **Antigravity only** | Antigravityåºæã®ãã€ã³ã¿ãŒ / Antigravity-specific pointer |
| `.cursor/rules/axiarch.mdc` | ð¶ **Cursor ã®ã¿** / **Cursor only** | Cursoråºæã®ãã€ã³ã¿ãŒã`init.sh` ã§èªåã³ã㌠/ Cursor-specific pointer. Auto-copied by `init.sh` |
| `.github/copilot-instructions.md` | ð¶ **Copilot ã®ã¿** / **Copilot only** | Copilotåºæã®ãã€ã³ã¿ãŒã`init.sh` ã§èªåã³ã㌠/ Copilot-specific pointer. Auto-copied by `init.sh` |
| `.windsurfrules` | ð¶ **Windsurf ã®ã¿** / **Windsurf only** | Windsurfåºæã®ãã€ã³ã¿ãŒã`init.sh` ã§èªåã³ã㌠/ Windsurf-specific pointer. Auto-copied by `init.sh` |
| `CLAUDE.md` | ð¶ **Claude Code ã®ã¿** / **Claude Code only** | Claude Codeåºæã®ãã€ã³ã¿ãŒã`init.sh` ã§èªåã³ã㌠/ Claude Code-specific pointer. Auto-copied by `init.sh` |
| `.claude/settings.json` | ð¶ **Claude Code ã®ã¿** / **Claude Code only** | `UserPromptSubmit` 匷å¶å·è¡ããã¯ã`init.sh` ã§èªåã³ã㌠/ Enforcement hook. Auto-copied by `init.sh` |
| `scripts/` | ð· **æšå¥š** / **Recommended** | 蚺æã»ãã«ã¹ãã§ãã¯ã¹ã¯ãªããéïŒ`check-axiarch-health.sh` ã§å
šãããã³ã«éµå®ãäžçºèšºæã`check-git-config-clean.sh` ã§ `.git/config` å¥å
šæ§ãã§ãã¯ïŒã`init.sh` ã§èªåã³ã㌠/ Diagnostic & health check scripts (`check-axiarch-health.sh` for full-protocol compliance one-shot, `check-git-config-clean.sh` for `.git/config` integrity). Auto-copied by `init.sh` |
| `axiarch-prompts/` | ð· **ä»»æ** / **Optional** | ããã³ãããã³ãã¬ãŒãé / Prompt template library |
| `init.sh` | ð· **ä»»æïŒæšå¥šïŒ** / **Optional (Recommended)** | 察話åŒã»ããã¢ããã¹ã¯ãªãããèšèª/ãšãŒãžã§ã³ãéžæããã¡ã€ã«ã³ããŒã次ã®ã¹ããããèªåå / Interactive setup script. Automates language/agent selection, file copy, and next-step guidance |
| `CHANGELOG.md` | â äžèŠ / Not needed | ãªããžããªç®¡ççš / For this repo only |
| `CONTRIBUTING.md` | â äžèŠ / Not needed | ãªããžããªç®¡ççš / For this repo only |
| `SECURITY.md` | â äžèŠ / Not needed | ãªããžããªç®¡ççš / For this repo only |
| `CODE_OF_CONDUCT.md` | â äžèŠ / Not needed | ãªããžããªç®¡ççš / For this repo only |
| `LICENSE` / `NOTICE` | â äžèŠ / Not needed | ãªããžããªç®¡ççš / For this repo only |
| `.github/` | â äžèŠ / Not needed | Issue/PRãã³ãã¬ãŒãããªããžããªç®¡ççš / Issue/PR templates. For this repo only |
| `.gitignore` | â äžèŠ / Not needed | ãªããžããªç®¡ççš / For this repo only |
### ãšãŒãžã§ã³ãå¥ã»ããã¢ãã / Agent-Specific Setup
| æé / Step | Antigravity | Codex | Cursor | Claude Code | Copilot | Windsurf |
|:-----------|:------------|:------|:-------|:------------|:--------|:---------|
| 1. `AGENTS.md` + `axiarch-rules/` ãã³ããŒïŒ`axiarch-prompts/` ã¯ä»»æïŒ | â
| â
| â
| â
| â
| â
|
| 2. `.agents/rules/prompt_pointer.md` ãé
眮 | â
**å¿
é ** | â äžèŠ | â äžèŠ | â äžèŠ | â äžèŠ | â äžèŠ |
| 3. `CLAUDE.md` + `.claude/settings.json` é
眮 | â äžèŠ | â äžèŠ | â äžèŠ | â
`init.sh` èªå / Auto via `init.sh` | â äžèŠ | â äžèŠ |
| 4. 远å èšå® | â | â (AGENTS.md = native) | ä»»æ: `.cursor/rules/*.mdc` | â | ä»»æ: `.github/copilot-instructions.md` | ä»»æ: `.windsurfrules` |
### ð¡ïž Claude Code 匷å¶å·è¡æ©æ§ / Enforcement Mechanism
> **JA**: Claude Code æ¡çšãããžã§ã¯ãã«ã¯ `.claude/settings.json` ã忢±ããã`UserPromptSubmit` ããã¯ã**æ¯ãŠãŒã¶ãŒããã³ããéä¿¡æ**ã« AGENTS.md ãããã³ã«ã®æé»å®è¡ã匷å¶ããŸããAI ãã軜ãäŒè©±ã ããããš LOADING_PROTOCOL ãã¹ãããããåé¡ãç©ççã«é²æ¢ããä»çµã¿ã§ãã
>
> **EN**: Claude Code projects ship with `.claude/settings.json` containing a `UserPromptSubmit` hook that compels implicit AGENTS.md protocol execution **on every user prompt**. This physically prevents the AI from skipping LOADING_PROTOCOL just because a prompt feels "casual."
| ãã¡ã€ã« / File | åœ¹å² / Role | ã³ããã / Commit |
|:----------------|:------------|:------------------|
| `.claude/settings.json` | ããŒã å
±æã® Axiarch 匷å¶ãã㯠/ Team-shared Axiarch enforcement hook | â
**å¿
é ** / **Required** |
| `.claude/settings.local.json` | åäººã®æš©éã»èš±å¯èšå® / Personal permissions | â gitignored |
| `.claude/worktrees/`, `.claude/projects/` | Claude Code ã»ãã·ã§ã³ããŒã¿ / Claude Code session data | â gitignored |
> [!CAUTION]
> **JA**: ãã®ããã¯ã®**åé€ã»ç¡å¹åã¯ãæ²æ³æ¹æ£ãã¬ãã«ã®ç Žå£ç倿Ž**ã§ããããªãŒããŒã®æç€ºçæ¿èªãå¿
èŠã§ãã詳现㯠`axiarch-rules/{lang}/LOADING_PROTOCOL.md` ã®ã匷å¶å·è¡æ©æ§ãã»ã¯ã·ã§ã³ãåç
§ã
>
> **EN**: **Removing or disabling this hook is a constitution-amending destructive change** requiring explicit owner approval. See "Enforcement Mechanism" in `axiarch-rules/{lang}/LOADING_PROTOCOL.md`.
#### ð ãã©ãã«ã·ã¥ãŒã / Troubleshooting
> **JA**: ãããã¯ãåããŠããªãæ°ããããå Žåã`bash scripts/check-axiarch-health.sh` ãå®è¡ããŠãã ããã**ããã¯æå¹åã»çºç«å±¥æŽã»AI éµå®ïŒtask.md ããŒãå±¥æŽïŒãäžçºèšºæ**ããŸãã`init.sh` çµç±ã§èªåé
åžããã axiarch æšæºããŒã«ã§ãã
>
> **EN**: When you suspect "the hook isn't firing", run `bash scripts/check-axiarch-health.sh`. **One-shot diagnosis** of hook enablement, firing history, and AI adherence (task.md load logs). Distributed automatically via `init.sh`.
```bash
bash scripts/check-axiarch-health.sh
# Or from another directory:
bash /path/to/project/scripts/check-axiarch-health.sh /path/to/project
```
> [!NOTE]
> **JA**: å
¬åŒä»æ§: [Hooks](https://code.claude.com/docs/en/hooks) / [Permissions](https://code.claude.com/docs/en/permissions)ã`permissions.allow` ã« `Bash(echo *)` ã远å ããå¿
èŠã¯**ãããŸãã**ïŒhook command 㯠permission çµè·¯å€ïŒã
>
> **EN**: Official spec: [Hooks](https://code.claude.com/docs/en/hooks) / [Permissions](https://code.claude.com/docs/en/permissions). Adding `Bash(echo *)` to `permissions.allow` is **NOT required** (hook commands are spawned outside the permission pipeline).
### 1. ãããžã§ã¯ãã«ã³ã㌠/ Copy to your project
> [!TIP]
> **JA**: `init.sh` ã䜿ããšå¯Ÿè©±åŒã§èšèªã»ãšãŒãžã§ã³ããéžæããŠãã¡ã€ã«ãèªåã³ããŒã§ããŸããæåã»ããã¢ããã®ä»£ããã«äœ¿çšå¯èœã§ãã
>
> **EN**: Use `init.sh` for an interactive setup that automatically selects language/agent and copies files. It can be used instead of manual setup.
```bash
# æšå¥š: init.sh ã§èªåã»ããã¢ãã / Recommended: Auto-setup with init.sh
curl -sSL https://raw.githubusercontent.com/hiroyuki-miyauchi/axiarch/main/init.sh | bash
# ãŸãã¯æåã§ã³ã㌠/ Or copy manually:
# å¿
é ã®2ã€ã ãã³ã㌠/ Copy only the 2 required items
cp AGENTS.md /path/to/your/project/
cp -r axiarch-rules /path/to/your/project/
# æšå¥šïŒèšºæã»ãã«ã¹ãã§ãã¯ã¹ã¯ãªããé / Recommended: diagnostic & health-check scripts
cp -r scripts /path/to/your/project/
# ä»»æïŒããã³ããéãã³ã㌠/ Optional: copy prompt library
cp -r axiarch-prompts /path/to/your/project/
```
### 2. ãšãŒãžã§ã³ãåºæã®èšå® / Agent-specific config
```bash
# === Google Antigravity ===
# .agents/rules/ ã¯Antigravityåºæãèªåèªã¿èŸŒã¿å¯Ÿè±¡ãªã®ã§ãã€ã³ã¿ãŒãé
眮ã
# .agents/rules/ is Antigravity-specific. Place a pointer for auto-loading.
mkdir -p /path/to/your/project/.agents/rules
cp .agents/rules/prompt_pointer.md /path/to/your/project/.agents/rules/
# === OpenAI Codex ===
# Codexã¯AGENTS.mdããã€ãã£ãèšå®ãšããŠèªãããã远å èšå®ã¯äžèŠã§ãã
# Codex reads AGENTS.md natively â no additional setup needed.
# (AGENTS.md is already copied in Step 1)
# === Claude Code ===
# Claude Codeã¯CLAUDE.mdããã€ãã£ãã«èªãã®ã§ãã€ã³ã¿ãŒãã³ããŒã
# 匷å¶å·è¡ãã㯠(.claude/settings.json) ãé
眮ããã
# Claude Code reads CLAUDE.md natively â copy the pointer file.
# Also place the enforcement hook (.claude/settings.json).
cp CLAUDE.md /path/to/your/project/CLAUDE.md
mkdir -p /path/to/your/project/.claude
cp .claude/settings.json /path/to/your/project/.claude/settings.json
# === Cursor ===
# ãã€ãã£ãèšå®ãã¡ã€ã«ã忢±ãããŠããŸããinit.sh ã§èªåã³ããŒãããŸãã
# A native config file is included. It is auto-copied by init.sh.
# æåã»ããã¢ããã®å Žå / For manual setup:
mkdir -p /path/to/your/project/.cursor/rules
cp .cursor/rules/axiarch.mdc /path/to/your/project/.cursor/rules/
# === GitHub Copilot ===
# ãã€ãã£ãèšå®ãã¡ã€ã«ã忢±ãããŠããŸããinit.sh ã§èªåã³ããŒãããŸãã
# A native config file is included. It is auto-copied by init.sh.
# æåã»ããã¢ããã®å Žå / For manual setup:
mkdir -p /path/to/your/project/.github
cp .github/copilot-instructions.md /path/to/your/project/.github/
# === Windsurf ===
# ãã€ãã£ãèšå®ãã¡ã€ã«ã忢±ãããŠããŸããinit.sh ã§èªåã³ããŒãããŸãã
# A native config file is included. It is auto-copied by init.sh.
# æåã»ããã¢ããã®å Žå / For manual setup:
cp .windsurfrules /path/to/your/project/
```
> [!CAUTION]
> **JA**: `.agents/rules/` 㯠**Antigravityåºæ**ã®ãã£ã¬ã¯ããªã§ããCodexãCursorãClaude CodeãGitHub Copilotã§ã¯äžèŠã§ããç¹ã«Codex㯠`AGENTS.md` èªäœããã€ãã£ãèšå®ãªã®ã§ãStep 1ã®2ç¹ã³ããŒã ãã§åäœããŸããåããŒã«ã«ã¯åºæã®èšå®ãã£ã¬ã¯ããªããããŸãïŒäžè¡šåç
§ïŒãAntigravityã®å Žåããã€ã³ã¿ãŒã®ã¿é
眮ããã«ãŒã«æ¬äœã¯ `axiarch-rules/` ã«äžå
管çã
>
> **EN**: `.agents/rules/` is **Antigravity-specific**. It is NOT needed for Codex, Cursor, Claude Code, or GitHub Copilot. Codex in particular uses `AGENTS.md` as its native config, so the 2-item copy in Step 1 is all you need. Each tool has its own native configuration directory (see table above). For Antigravity, only place the pointer here â rule definitions live in `axiarch-rules/`.
### 3. åæå / Initialize
```bash
# AGENTS.md ãç·šé â Project Native Language ã Japanese ãŸã㯠English ã«èšå®
# Edit AGENTS.md â Set Project Native Language to Japanese or English
# 䜿çšããªãèšèªãã£ã¬ã¯ããªãåé€ / Delete unused language directory:
rm -rf axiarch-rules/en # For Japanese projects
# OR
rm -rf axiarch-rules/ja # For English projects
# axiarch-prompts/ ãã³ããŒããå Žå / If you copied axiarch-prompts/:
rm -rf axiarch-prompts/en # For Japanese projects
# OR
rm -rf axiarch-prompts/ja # For English projects
```
### 4. èšå®ãšéçº / Configure & Develop
| Step | JA | EN |
|:-----|:---|:---|
| 1 | `{lang}/blueprint/core/000_project_overview.md` ããããžã§ã¯ãã«åãããŠç·šé | Edit `{lang}/blueprint/core/000_project_overview.md` for your project |
| 2 | æ°æ©èœã¯ `core/998_feature_spec_template.md` ã察å¿ãããã¡ã€ã³ãã©ã«ãã«ã³ã㌠| For new features, copy `core/998_feature_spec_template.md` to the corresponding domain folder |
| 3 | **ã³ãŒããæžãåã«åãå
¥ãæ¡ä»¶ãæžã**ïŒBlueprint FirstïŒ | **Write Acceptance Criteria before writing code** (Blueprint First) |
| 4 | éçºéå§ â AIãšãŒãžã§ã³ãã¯æ²æ³ã«åŸã | Start development â AI agents will follow the constitution |
### å°å
¥åŸã®ãã£ã¬ã¯ããªæ§æ / Post-Setup Directory Structure
```text
your-project/
âââ AGENTS.md â å¿
é ïŒæé«æ³èŠ / Required: Supreme Law
âââ CLAUDE.md â Claude Code ã®ã¿ïŒãã€ã³ã¿ãŒïŒ / Claude Code only (pointer)
âââ .claude/ â Claude Code ã®ã¿ / Claude Code only
â âââ settings.json â UserPromptSubmit 匷å¶å·è¡ããã¯ïŒã³ãããå¿
é ïŒ / Enforcement hook (must commit)
âââ .agents/ â Antigravity ã®ã¿ / Antigravity only
â âââ rules/
â âââ prompt_pointer.md â ãã€ã³ã¿ãŒ / Pointer
âââ axiarch-rules/ â å¿
é ïŒã«ãŒã«æ¬äœ / Required: Rule Definitions
â âââ ja/ (or en/) â èšèªéžæ / Language selected
â âââ INDEX.md
â âââ LOADING_PROTOCOL.md
â âââ CRYSTALLIZATION_PROTOCOL.md
â âââ universal/ â äžå€ / Immutable
â âââ blueprint/ â ãããžã§ã¯ãåºæ / Project-Specific
âââ axiarch-prompts/ â ä»»æïŒããã³ããé / Optional: Prompt Library
â âââ ja/ â æ¥æ¬èªç / Japanese
â â âââ develop/ â éçºã»å®è¡ / Development & Execution
â â âââ audit/ â åè³ªç£æ» / Quality Auditing
â â âââ govern/ â ã¬ããã³ã¹ / Governance
â â âââ operate/ â ã€ã³ã·ãã³ãã»åå
¥ / Incidents & Onboarding
â âââ en/ â è±èªç / English
â âââ develop/
â âââ audit/
â âââ govern/
â âââ operate/
âââ src/ â ãããžã§ã¯ãã³ãŒã / Your Code
```
---
## ðïž èæ¯ / Background
### ð¯ðµ ãªãAxiarchãäœã£ãã
AxiarchïŒã¢ã¯ã·ã¢ãŒã¯ïŒã¯ãAIãšã®ååã«ãããå質課é¡ãžã®å±æ©æããçãŸããŸããã
AIãšãŒãžã§ã³ããç»å Žãã以åãããçæAIïŒChatGPTçïŒãæ°èŠäºæ¥ã®æ§æ³ãå®åã®äžã§æŽ»çšããŠããŸãããã³ãŒããæžãããŠã¿ãããšããããŸããããæ¬æ Œçãªéçºã«äœ¿ã£ãããã§ã¯ãããŸãããããã§ããAIãšåãåãäžã§ã²ãšã€ã ã確信ããŠããããšããããŸãã â **ãã«ã·ããŒã·ã§ã³ãã³ã³ããã¹ãã®åªå€±ãå質ã®ãã¬ãæç¢ºãªã«ãŒã«ãªãã«AIã䜿ãç¶ããã°ãå質ã¯éãã«å£åããç«ãŠçŽããé£ãããªãã**
2025幎ã[Google Antigravity](https://antigravity.google/)ã®ãããªAIãšãŒãžã§ã³ããç»å ŽãããšããçŽæããŸããïŒ**ãã¬ããã³ã¹æ§é ãªãã«ããã䜿ãå§ãããåãè¿ããã€ããªãã**ãšãéã«èšãã°ã**æåãããæ²æ³ããæŽåããã°ãå質ãåçã«åºäžãã§ãã**ã¯ãã ãšã
ããã§ãéçºãå§ãã**åã«**匷åºãªã¬ããã³ã¹ã®æ§ç¯ã«çæããŸããããããŠãã®ã¢ãŒããã¯ãã£ãæºããŠå®éã®ãããã¯ã·ã§ã³éçºã«æå
¥ããæ°çŸã»ãã·ã§ã³ã®å®çžŸãçµãŠãã©ãã·ã¥ã¢ããããŠãããŸããã
**ãã®éçšã§ãã«ãŒã«ãªãã§ã¯è§£æ±ºäžå¯èœãªãã¿ãŒã³ãæç¢ºã«ãªããŸããïŒ**
- **ã³ã³ããã¹ãå¥å¿çã** ã»ãã·ã§ã³éã§ã¢ãŒããã¯ãã£ã®æ±ºå®ãå¿ãããåœåèŠçŽãããªãããããã»ãã¥ãªãã£ãã¿ãŒã³ãéè¡ãããä¿®æ£æžã¿ã®ãã°ãåå°å
¥ãããã
- **æçžŠè
äŸåã®å質ã** æ£ç¢ºãªæç€ºãåºãã°åªããææç©ãåºããææ§ãªæç€ºã§ã¯ãããªããŒã·ã§ã³ãçç¥ããåå®å
šãç¡èŠãã宿ãªã·ã§ãŒãã«ããã«éããã
- **Vibe Coding ã®åŒåã** æç€ºçãªå¶çŽããªããã°ããã¹ãŠã®ã»ãã·ã§ã³ã "vibe coding" ã«åŒãå¯ãããã â æ£ããèŠãããã確ç«ããããã¿ãŒã³ãæé»ã«éåããã³ãŒãã
- **ç¥èŠã®èžçºã** èŠåŽããŠåŸãæèšã倱ãããåãåé¡ãäœåºŠãåçºèŠãããããã
æ¥çå
šäœãåãåé¡ã«çŽé¢ããŠããŸãããAIãçæããã³ãŒããã»ãã¥ãªãã£è匱æ§ã®äž»èŠãªåå ãšãªãã€ã€ãããšã®å ±åã"AI slop"çŸè±¡ â äœå質ã»éè€ã³ãŒãã®æ°Ÿæ¿«ãã³ã³ããã¹ããŠã£ã³ããŠã®å£åã«ããäžè²«æ§ã®åªå€±ãããšãåå¥ã¹ãããã®æåçãé«ããšãããã«ãã¹ãããã®ã¯ãŒã¯ãããŒã§ã¯å€±æçã倧ããèšãäžãããŸãã
**æ°ã¥ãïŒ** åé¡ã¯AIã¢ãã«ã§ã¯ãªãã*ã¬ããã³ã¹ã®äžåš*ã§ãããã»ãã·ã§ã³ãæçžŠè
ãã³ã³ããã¹ããªã»ãããè¶
ããŠçãæ®ããæ°žç¶çã§åŒ·å¶åã®ããã¬ããã³ã¹æ§é â ãæ²æ³ã â ãååšããªãã£ãã®ã§ãã
ãã®åé¡ã«å¯Ÿããäžã€ã®å®è·µçã¢ãããŒããšããŠæ§ç¯ããã®ã Axiarch ã§ããäœè
èªèº«ãããã³ããšã³ããšã³ãžãã¢ãšããŠã®çµéšã¯ãããã®ã®ãããã¯ãšã³ããã€ã³ãã©ã¯æªçµéšããã®ã¹ã¿ãŒãã§ãããã ããããã**ç¹å®ã®èšèªããã¬ãŒã ã¯ãŒã¯ã«äŸåããããšã³ãžãã¢çµéšã®æ·±ãã«é¢ããããAIãšååãããã¹ãŠã®éçºè
â éãšã³ãžãã¢ãå人éçºè
ãå«ã â ãæäœéã®å質åºäžããå®çŸã§ãã**ãã¬ãŒã ã¯ãŒã¯ãç®æããŸããã
äžçäžã®éçºè
ãAIãšãŒãžã§ã³ãã®æ©æµãæå€§éã«åããããããã«ãç¹ã«AI掻çšã«ãããŠãŸã çºå±éäžã«ããæ¥æ¬ããã®çºä¿¡ãšããŠãèªåœã®AI掻çšä¿é²ã«ãè²¢ç®ã§ããã°ãšèããŠããŸãããããŠäœããããã®åãçµã¿èªäœãèªèº«ã®ç¥èŠãæ·±ããããã»ã¹ã§ããããŸãã
ææïŒ2,500以äžã®æ²æ³åºæºãç¶²çŸ
ãã39ã®Universalã«ãŒã«ãã¡ã€ã«ããã«ã·ããŒã·ã§ã³ãªã¹ã¯ã軜æžãã5ã¹ãããã®Boot Sequence Protocolãæèšãèªåçã«ã«ãŒã«ã«å€æããCrystallization Protocolããã¹ãŠããã£ãäžã€ã® `AGENTS.md` æ²æ³ãéããŠåŒ·å¶ãããŸãã
### ðºðž Why Axiarch Was Built
Axiarch was born from a conviction that AI without governance will inevitably fail.
Before AI agents existed, working with generative AI (ChatGPT, etc.) for business planning and day-to-day work revealed a persistent truth â **hallucinations, context loss, and inconsistent quality are inevitable without explicit rules.** AI is powerful, but using it without governance means quality will quietly erode in ways that become increasingly hard to course-correct.
When AI agents like [Google Antigravity](https://antigravity.google/) emerged in 2025, the intuition was immediate: **"Starting without a governance structure would be irreversible."** Conversely, **establishing a "constitution" from day one could dramatically raise the quality floor.**
So the governance architecture was built **before** development began. Then it was deployed into real production development and refined through hundreds of real production sessions.
**Through that process, patterns that were impossible to solve without governance became clear:**
- **Context amnesia.** The AI forgot architectural decisions between sessions. Naming conventions drifted. Security patterns regressed. Fixed bugs were re-introduced.
- **Operator-dependent quality.** When instructions were precise, output was excellent. When vague, the AI took shortcuts â skipping validation, ignoring type safety, reaching for easy escapes.
- **Vibe coding gravity.** Without explicit constraints, every session tended toward "vibe coding" â code that looked correct but silently violated established patterns.
- **Knowledge evaporation.** Hard-won lessons were lost and had to be re-discovered repeatedly.
The industry was facing the same problem at scale. AI-generated code becoming a common vector for security vulnerabilities. The "AI slop" phenomenon â proliferation of low-quality, duplicated code â accelerating technical debt. Context window degradation causing models to lose coherence. High individual step reliability compounding into significant failure rates across multi-step workflows.
**The realization:** The problem wasn't the AI model. It was the absence of *governance*. There was no "constitution" â no persistent, enforceable set of rules that survived across sessions, operators, and context resets.
Axiarch was built as one practical approach to this challenge. The author's own background â front-end engineering experience, but no prior back-end or infrastructure expertise â shaped the core design principle: **language-agnostic, framework-agnostic, and accessible regardless of engineering experience. A Constitution-Driven AI Agent Governance Framework that enables any developer working with AI â including non-engineers and solo developers â to raise their minimum quality floor.**
The goal extends beyond personal use: to contribute to the global adoption of AI-assisted development, and to help accelerate AI utilization worldwide. The result: 39 Universal Rule files covering 3,000+ constitution standards, a 5-step Boot Sequence Protocol that reduces startup hallucination risk, a Crystallization Protocol that automatically converts lessons into rules. All enforced through a single `AGENTS.md` constitution that any AI agent can read.
---
## ð¡ èšèšææ³ / Philosophy
**Axiarch** ã¯ããæ®éæ²æ³ïŒUniversalïŒããšãåºæä»æ§ïŒBlueprintïŒãã®æç¢ºãªè²¬ååé¢ã«ãã**2倧ã³ã¢å±€**ãšã確å®ãªå®è¡ãæ
ä¿ãã**æå¿ã®ç¬¬3å±€ïŒPromptsïŒ**ãããªã**ã3局統åã¬ããã³ã¹ã»ã¢ãŒããã¯ãã£ïŒThree-Layer Governance ArchitectureïŒã**ã®äžã«æ§ç¯ãããŠããŸãã
**Axiarch** is built on a **"Three-Layer Governance Architecture"**, strictly separating responsibilities into two core layersâthe **Immutable Constitution (Universal)** and the **Mutable Project State (Blueprint)**âplus a third **Optional Execution Layer (Prompts)**.
### ðïž The Three-Layer Architecture (3å±€ã¬ããã³ã¹æ§é )
#### Layer 1: Universal (äžå€å±€ / Immutable Constitution)
- **圹å²**: æä»£ããããžã§ã¯ããå€ãã£ãŠãæºãããªãããœãããŠã§ã¢éçºããã³ãããã¯ãéçšã®çµ¶å¯Ÿçåºæºãšå¶çŽããå®çŸ©ãã岩ç€å±€ïŒ`AGENTS.md` + `universal/`ïŒã
- **ç¹æ§**: **Read-OnlyïŒåŒ·å¶ä¿è·ïŒ**ãAIãšãŒãžã§ã³ãã«ããç¬æã®å€æŽãåºãçŠãããäžå€ã®é åãã§ããAIç¹æã®ç¬èªè§£éïŒVibe CodingïŒãã»ãã¥ãªãã£è匱åã®ãªã¹ã¯ãæ§é çã«è»œæžããŸãã
#### Layer 2: Blueprint (å¯å€å±€ / Mutable Project State)
- **圹å²**: ãããžã§ã¯ãã®äºæ¥ç®çãåºæã®æ©èœä»æ§ããããŠéçºäžã«åŸããããå®è·µã®æèšããèç©ã»çµæ¶åããå±€ïŒ`blueprint/`ïŒã
- **ç¹æ§**: **Read-WriteïŒåçæé·ïŒ**ãAIãšãŒãžã§ã³ãèªèº«ãç¶ç¶çã«ä»æ§ãæèšãæžã蟌ã¿ãèªåŸããŒããéããŠæèã埩å
ããããã®ãçããèšæ¶é åãã§ãããæšæ¥ãŸã§ã®åæãå¿ããïŒContext AmnesiaïŒããªã¹ã¯ã倧å¹
ã«è»œæžããç¶ç¶çãªå質ã®åºäžãïŒQuality Floorã®åäžïŒããããããŸãã
#### Layer 3: Prompts (å®å
šä»»æå±€ / Purely Optional Execution Engine)
- **圹å²**: Layer 1ãš2ã®ã«ãŒã«ããç¹å®ã¿ã¹ã¯ïŒã»ãã¥ãªãã£ç£æ»ãèšèšæé©åãã€ã³ã·ãã³ã察å¿ãªã©ïŒãžåŒ·å¶çã«åŒã³åºãé©çšããããšã³ãžã³ïŒ`axiarch-prompts/`ïŒã
- **ç¹æ§**: **OptionalïŒå®å
šä»»æïŒ**ãAxiarchã®æ žã¯Layer 1ãš2ã®éµå®ã§ãããLayer 3ãå°å
¥ããªããŠããã¬ãŒã ã¯ãŒã¯èªäœã¯æ©èœããŸããå¿
èŠã«å¿ããŠç¹å®ã®é«å質ã¿ã¹ã¯ãå®è¡ããããã®ãªãã·ã§ãã«ãªãã©ã°ã€ã³ãšããŠæ©èœããŸãã
---
### åå / Principles
| åå / Principle | JA | EN |
|:-----------------|:---|:---|
| **Blueprint First** | å€§èŠæš¡å€æŽã¯å®è£
åã«ä»æ§æŽæ°ã矩åå | Specs before code â major changes require spec updates before implementation |
| **Multi-step refinement** | INDEX â ã«ãŒã«ããŒã â Blueprint â ã¿ã¹ã¯ â èšç» â å®è£
â Walkthrough â æèš | INDEX â Rule Load â Blueprint â Task â Plan â Implement â Walkthrough â Lessons |
| **Strong guardrails** | 16ã®çŠæ¢äºé
+ 5ã€ã®ãããã³ã« + 3å±€Git Hooks | 16 prohibitions + 5 protocols + 3-layer Git Hooks |
| **Continuous learning** | ã¿ã¹ã¯å®äºåŸã«æèšãã«ãŒã«ã«çµæ¶å | Lessons crystallized into rules after every task |
| **Quality floor** | æçžŠè
ã®ã¹ãã«ã«é¢ãããæäœåè³ªåºæºãåŒ·å¶ | Minimum standards enforced regardless of operator skill |
> **JA**: AxiarchïŒã¢ã¯ã·ã¢ãŒã¯ïŒã®æ žå¿ç䟡å€ã¯ãåãªãããã³ããã®éåäœã§ã¯ãªãããã®ã**æ®éïŒLayer 1ïŒãšå¯å€ïŒLayer 2ïŒã®æç¢ºãªè²¬ååé¢**ããšã**ããã³ããïŒLayer 3ïŒã«ããå®è¡é§å**ããšãã3å±€æ§é ã«ããç¶æ
管çã¢ãŒããã¯ãã£ã«ãããŸãããBlueprint FirstïŒå®è£
åã®ä»æ§å®çŸ©ã®çŸ©ååïŒããšãæ²æ³ããé£åãããããšã§ãAIã®ãã«ã·ããŒã·ã§ã³ãå質ããªãããé²ãããã®å
ç¢ãªã¬ããã³ã¹åºç€ãšãªããéçºããéçšãŸã§ã®å
šäœå質ã®åºäžãããããããŸãã
>
> **EN**: Axiarch's core value lies not in being a mere prompt collection, but in its state-management architecture built upon "the **Clear Separation of Immutable (Layer 1) and Mutable (Layer 2)**" and "the **Execution-driven Prompts (Layer 3)**". By integrating "Blueprint First" with the Constitution, it serves as a robust governance foundation that structurally mitigates AI hallucinations and quality drift, thereby raising the overall quality floor from development to operations.
---
## ð€ ã³ã³ããªãã¥ãŒã / Contributing
**JA**: ã³ã³ããªãã¥ãŒããæè¿ããŸãããã«ãªã¯ãšã¹ãã®åã«ããŸãIssueã§å€æŽææ¡ãè°è«ããŠãã ããã
**EN**: Contributions are welcome. Please open an issue to discuss proposed changes before submitting a pull request.
> [!IMPORTANT]
> **JA**: Universal Rules (`{lang}/universal/`) ã¯**æ²æ³**ã§ãã倿Žã«ã¯æç€ºçãªãæ²æ³æ¹æ£ãã®æ¿èªãå¿
èŠã§ããBlueprintãã³ãã¬ãŒããåºç€ãã¡ã€ã«ã¯éåžžã®ã³ã³ããªãã¥ãŒããåãä»ããŸãã
>
> **EN**: Universal Rules (`{lang}/universal/`) are the **Constitution**. Modifications require explicit "Amend Constitution" approval. Blueprint templates and infrastructure files accept standard contributions.
---
## ð ã©ã€ã»ã³ã¹ / License
**JA**: ãã®ãããžã§ã¯ã㯠**Apache License 2.0** ã®äžã§ã©ã€ã»ã³ã¹ãããŠããŸãã詳现㯠[LICENSE](LICENSE) ãã¡ã€ã«ãåç
§ããŠãã ããã
**EN**: This project is licensed under the **Apache License 2.0** â see the [LICENSE](LICENSE) file for details.
---
## èè
/ Author
**Hiroyuki Miyauchi** â [LinkedIn](https://www.linkedin.com/in/hiroyuki-miyauchi/) / [chronoviq.com](https://chronoviq.com/)
**JA**: ãã°å ±åãæ©èœãªã¯ãšã¹ã㯠[GitHub Issues](https://github.com/hiroyuki-miyauchi/axiarch/issues) ããå©çšãã ããã
**EN**: For bug reports and feature requests, use [GitHub Issues](https://github.com/hiroyuki-miyauchi/axiarch/issues).