Most CLAUDE.md files I see in the wild look like this:
# Rules
- Use TypeScript
- Write tests
- Be helpful
That's not a configuration. That's a wish list.
A real CLAUDE.md needs:
- Autonomy boundaries — what can the agent do without asking?
- Safety rules — what must it never do?
- Skill modules — how does it handle specific workflows?
- Multi-agent patterns — how do parallel agents avoid conflicts?
Writing all of this from scratch takes hours. So I built a CLI that does it in 30 seconds.
npx hideyoshi init
No install required. Just run it:
$ npx hideyoshi init
◆ Which AI coding tool?
│ ● Claude Code
│ ○ Cursor
│ ○ Windsurf
│ ○ Aider
│ ○ Multiple tools
◆ Primary framework?
│ ● Next.js
◆ Primary language?
│ ● TypeScript
◆ How many people work on this codebase?
│ ● 2-5 people
◆ How much autonomy should the agent have?
│ ● Moderate — commit freely, ask before deploy
◆ Will you run multiple AI agents in parallel?
│ ● Yes
◆ Which skill modules do you want?
│ ◻ Code Review
│ ◻ Bug Triage (verify-before-act)
│ ◻ Deploy Guard
│ ◻ Content Writer
It generates a complete set of files:
├── CLAUDE.md # Main agent config
├── AGENTS.md # Multi-agent safety rules
└── .agents/skills/
├── code-review/SKILL.md
├── bug-triage/SKILL.md
└── deploy-guard/SKILL.md
What the generated CLAUDE.md looks like
Here's a snippet from what it produces for a moderate-autonomy Next.js team:
# Project Agent Configuration
You are the development agent for this project.
## Autonomy Boundaries
### Do without asking
- Run tests and linting
- Format code
- Create commits within scope
- Install dependencies (retry once on failure)
### Ask before doing
- Deploy to production
- Modify CI/CD pipelines
- Change environment variables
- Delete files outside current task scope
## Coding Standards
- TypeScript (ESM) is the standard
- Next.js App Router conventions
- Verify builds pass before pushing
It's tailored to your tool, framework, team size, and autonomy level. Not a generic template.
Multi-agent safety (the part most people skip)
If you said "yes" to parallel agents, the CLI generates an AGENTS.md with rules like:
## Multi-Agent Safety
- Never use `git stash` (can destroy other agents' work)
- Never switch branches unless explicitly instructed
- Only stage YOUR changes — never `git add -A`
- Run `git pull --rebase` before pushing
- Ignore unfamiliar files (they belong to another agent)
These rules exist because I learned each one from a real incident. Force pushes, corrupted indexes, deleted configs — all from agents stepping on each other.
Skill modules
Skills are modular behavior files that teach your agent domain-specific workflows. The CLI generates them based on your selection:
Code Review skill — teaches the agent to check for security issues, style violations, and test coverage before approving.
Bug Triage skill — enforces "verify before act": reproduce the bug, identify the root cause (file + line), then fix. No guessing.
Deploy Guard skill — pre-deploy checklist: tests pass, no console errors, build succeeds, env vars present.
Each skill has a clear scope, specific do/don't rules, and an escalation path.
The free CLI vs. the full Playbook
The CLI generates starter configs — enough to get going. The full Playbook includes:
| Free CLI | Playbook | |
|---|---|---|
| CLAUDE.md template | 1 | 5+ (tool-specific) |
| Skill modules | Up to 4 | 10+ |
| Multi-agent patterns | Basic rules | Full conflict resolution |
| Safety checklists | — | Security + governance |
| Real-world examples | — | 3 complete configs |
| Trust spectrum framework | — | Full chapter |
The CLI is genuinely useful on its own. The Playbook is for teams that want the complete system.
Launch week: 50% off with code LAUNCH50 (expires March 28)
Try it: npx hideyoshi init
Questions? Find me at @hideyoshi_th.
Top comments (1)
Hello, I'm a new member who registered today. I own a company called M-GAMES. If you would like to volunteer (without pay) and help me develop and promote it, I would be grateful.