harness:init

Generate project files and configurations for a Claude Code Harness.

3|1|Updated Apr 5, 2026
One-click install
npx skills add https://github.com/huangbaixun/harness-engineering --skill harness-init-huangbaixun
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: harness:init
Source: https://github.com/huangbaixun/harness-engineering/tree/main/skills/init
Command: npx skills add https://github.com/huangbaixun/harness-engineering --skill harness-init-huangbaixun

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

It solves the problem of starting a new Claude Code project without a consistent, reliable AI “Harness” that enforces rules, restores context across sessions, and wires in safe workflows from day one.

Core Features & Use Cases

  • Project bootstrapping (6-layer Harness): Generates the required foundation files (CLAUDE.md, .claude/settings.json, hooks, init.sh, and core skill directories) so the agent behaves consistently and safely.
  • Guardrails via hooks + concise memory: Creates deterministic hook scripts for enforcement while keeping CLAUDE.md within the <=60-line principle and pushing detail into docs/.
  • Setup for plan→implement→verify loops: Installs workflow Skills for writing plans, test-driven development, and pre-completion verification.
  • Cross-session progress tracking: Initializes claude-progress.json and docs/features.json (when applicable) along with an archiving strategy to avoid unbounded token growth.

Quick Start

Use the harness:init Skill to initialize a new project by telling Claude Code: Help me initialize this project's Harness.

Frequently Asked Questions about harness:init

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I initialize a Claude Code project with session hooks and progress tracking?

To initialize a Claude Code project, use the harness setup to generate CLAUDE.md, session hooks, and claude-progress.json, ensuring the AI agent operates with deterministic enforcement and bounded context across sessions.

What is an AI agent harness and when do I need it for project initialization?

An AI agent harness is a six-layer foundation of configuration files and hooks that enforces rules and restores context. You need it when starting a new project to ensure consistent, safe agent behavior from day one.

Does initializing project guardrails overwrite my existing CLAUDE.md and architecture docs?

No, the initialization process generates required project files and configurations without overwriting existing content unless explicitly confirmed, protecting your current CLAUDE.md and architecture docs.

Why does my CLAUDE.md file need to be under 60 lines for AI agent workflows?

Keeping CLAUDE.md under 60 lines maintains bounded context for the AI agent. This concise memory principle pushes detailed information into docs/ to prevent unbounded token growth during sessions.

Can I set up plan, implement, and verify loops for multi-agent teams?

Yes, the harness setup installs workflow Skills for writing plans, test-driven development, and pre-completion verification, supporting cross-session progress tracking for both single- and multi-agent teams.

What's the best way to prevent unbounded token growth in Claude Code sessions?

The best way is initializing claude-progress.json and docs/features.json with an archiving strategy. This cross-session progress tracking limits token growth by bounding context within the AI agent harness.