@kweiza/harness

Claude Code harness setup CLI for Kweiza projects. Automates project initialization with company-wide standards, stack-specific presets, and Anthropic's recommended harness engineering practices.

Quick Start

npx @kweiza/harness init my-project

Features

• One-command setup — Interactive CLI generates CLAUDE.md, .claude/rules/, hooks, and more
• 9 stack presets — Next.js, Vite, FastAPI, Express, Figma Plugin, AI Agent, ComfyUI, Fullstack Platform, Docker
• Composable presets — Combine multiple presets (e.g., nextjs + docker)
• Anthropic best practices — Follows official harness engineering recommendations:
  • CLAUDE.md for project overview (< 200 lines)
  • .claude/rules/ for modular, path-scoped rules
  • .claude/settings.json for hooks (100% enforcement)
• Company standards — Git workflow, code quality, testing, security, documentation rules baked in
• Updatable — Sync projects with latest standards via harness update

Installation

# Use directly with npx (no install needed)
npx @kweiza/harness init

# Or install globally
npm install -g @kweiza/harness

Commands

harness init [path]

Initialize a new project with Kweiza harness standards.

npx @kweiza/harness init my-app

Interactive prompts will ask for:

• Project name
• Presets to apply (multi-select)
• Whether to initialize git

harness add <preset>

Add a preset to an existing project. Merges into existing CLAUDE.md, settings, and rules.

cd my-existing-project
npx @kweiza/harness add fastapi

harness update

Sync your project with the latest harness standards. Only updates CLAUDE.md, settings, and rules — never touches your scaffold files.

npx @kweiza/harness update

harness list

Show all available presets.

npx @kweiza/harness list

Available Presets

Preset	Stack	Language
nextjs	Next.js web app	TypeScript
vite	Vite web app	TypeScript
fastapi	FastAPI backend	Python
express	Express backend	TypeScript
figma-plugin	Figma plugin	TypeScript
ai-agent	Background AI agent	Python
comfyui	ComfyUI custom nodes	Python
fullstack-platform	Node-based generative AI platform	TS + Python
docker	Docker containerization	—

What Gets Generated

When you run harness init, the following structure is created in your project:

your-project/
├── CLAUDE.md                      # Project overview + build commands
├── .claude/
│   ├── settings.json              # Hooks (lint, format — enforced on every commit)
│   └── rules/
│       ├── git-workflow.md        # Branch naming, conventional commits
│       ├── code-quality.md        # Type safety, SRP, no magic numbers
│       ├── testing.md             # TDD, regression tests, CI gates
│       ├── security.md            # Secrets, OWASP, input validation
│       ├── documentation.md       # README, API docs
│       └── <preset>.md            # Stack-specific rules (path-scoped)
└── .gitignore

How It Works

Merge Strategy

Presets are composable. When you select multiple presets:

• CLAUDE.md — Base content at top, each preset appended as a section
• settings.json — Hooks merged into a single array (duplicates removed)
• rules/ — All rule files copied (base + each preset). Warns if > 20 files (Anthropic limit)
• scaffold files — Copied without overwriting existing files

Version Tracking

Harness writes version info to .claude/settings.local.json (gitignored). This lets harness update know which presets were applied and when.

Contributing

See CONTRIBUTING.md for development setup and contribution guidelines.

License

MIT