bitfrog-harness-init

One command to make any repo agent-ready.

/bitfrog-harness-init brainstorms your project deeply, then generates a minimal harness engineering configuration — so AI agents can write reliable code in your codebase.

What it generates

your-project/
├── CLAUDE.md                      # AI workbook — project structure, module boundaries, test strategy, workflows
├── Plans.md                       # Development plan
├── feedback-log.md                # Error log — 3 similar issues → new rule
├── .claude/
│   ├── settings.json              # Hooks — auto-detect boundary violations, block sensitive files, pre-commit verify
│   ├── agents/
│   │   ├── worker.md              # Dev agent — test-first, feedback loop, forbidden actions
│   │   └── reviewer.md            # Review agent — read-only, PASS/FAIL with grep commands
│   └── memory/
│       └── decisions.md           # Architecture decisions — auto-loaded each session

How it works

1. Explores — reads your project files, git log, directory structure
2. Brainstorms — asks 8-11 questions one at a time to understand your project goals, module boundaries, test strategy, known pitfalls
3. Confirms — presents the harness design section by section for your approval
4. Generates — writes all config files tailored to your specific project

It never guesses. It asks first.

Install — 10 seconds

Open Claude Code and paste this:

Install bitfrog-harness-init: run git clone --single-branch --depth 1 https://github.com/rainyulei/bitfrog-harness-init.git ~/.claude/skills/bitfrog-harness-init then tell me it's ready and I can use /bitfrog-harness-init in any project.

That's it. Claude clones the skill and you're good to go.

Add to your repo so teammates get it (optional)

Add bitfrog-harness-init to this project: run mkdir -p .claude/skills && cp -Rf ~/.claude/skills/bitfrog-harness-init .claude/skills/bitfrog-harness-init && rm -rf .claude/skills/bitfrog-harness-init/.git then tell me it's ready.

Files get committed to your repo. git clone just works for teammates.

Alternative: plugin marketplace

/plugin marketplace add rainyulei/bitfrog-harness-init
/plugin install bitfrog-harness-init@rainyulei-bitfrog-harness-init

Usage

In any project directory:

/bitfrog-harness-init

Then answer the questions. That's it.

What is Harness Engineering?

Three layers of working with AI:

Layer	What it controls	Example
Prompt Engineering	How you talk to AI	"Write a function that..."
Context Engineering	What AI can see	CLAUDE.md, docs, codebase
Harness Engineering	What happens after AI acts	Hooks, tests, feedback loops, agent review

Harness Engineering is the third layer — it's not about making AI smarter, it's about making AI reliable.

This plugin generates the harness layer for your project.

Principles

• Brainstorm first — won't generate config until it understands your project
• Respect existing config — extends your CLAUDE.md, merges into settings.json
• Minimal viable harness — no rules the project doesn't need
• Hook errors must be useful — what was violated → why the rule exists → how to fix
• Evolution over perfection — start simple, feedback-log drives improvements

Reference implementation

See storyflow for a complete example of harness engineering applied to a real project.

License

MIT