I build AI Dev OS — an open framework that turns tacit developer knowledge into explicit, enforceable rules for AI coding assistants.
The problem: AI generates code that looks correct but violates your team's conventions, security practices, and architectural patterns. Loading more guidelines into context actually degrades output quality.
The solution: A layered rule architecture (Lifespan Layers) with a two-tier context strategy — 3-5 static rules during generation + comprehensive check & fix after generation. Benchmark: 96.9/100.
npx ai-dev-os init| Repository | What it does |
|---|---|
| ai-dev-os | Core framework — Lifespan Layers, Specificity Cascade, theory |
| rules-typescript | TypeScript / Next.js coding guidelines (L1–L3) |
| rules-python | Python / FastAPI coding guidelines (L1–L3) |
| plugin-claude-code | Claude Code — Skills, Agents, Hooks |
| plugin-cursor | Cursor — .mdc rules |
| plugin-kiro | Kiro — Steering Rules, Hooks |
| cli | npx ai-dev-os init — one-command setup |
| benchmark | Quantitative impact data — guideline effect on code quality |
75% of rules survive tool migrations. Switch between Claude Code, Cursor, and Kiro freely.
- AI DEV OS book (Japanese) — Zenn
- Practical tutorials (Japanese) — Qiita