revfactory/harness

A meta-skill that designs domain-specific agent teams, defines specialized agents, and generates the skills they use.

What it solves

Harness provides a way to automatically design and deploy multi-agent teams for Claude Code. Instead of manually configuring agents and their roles, users can provide a domain description, and the tool generates the necessary agent definitions, skills, and orchestration logic to handle complex tasks efficiently.

How it works

It functions as a "team-architecture factory" that analyzes a project's domain and control-flow requirements to select the best execution mode and architecture pattern. It generates specific artifacts in the .claude/ directory, including agent definitions (who does what) and skills (how they do it), and provides a CLAUDE.md pointer for triggering.

It supports three primary execution modes:

  1. Workflow orchestration: Deterministic scripts for fan-outs and verification loops.
  2. Persistent agent collaboration: Named agents that maintain context across turns.
  3. Sub-agent delegation: Lightweight, one-shot parallel dispatch.

Who it’s for

Developers using Claude Code who want to move beyond single-agent interactions to structured, multi-agent teams capable of handling larger-scale or more complex software engineering workflows.

Highlights

  • Automated Team Design: Supports six architecture patterns (e.g., Pipeline, Expert Pool, Supervisor) to map tasks to the best agent structure.
  • Dynamic Model Selection: Assigns model tiers (fable, opus, sonnet) based on task complexity and latency needs rather than using a blanket pin.
  • Evolution Mechanism: The /harness:evolve skill allows users to feed usage feedback back into the system to improve agents and skills over time.
  • Workflow-native Quality Patterns: Includes built-in patterns for adversarial verification, judge panels, and completeness critics to reduce errors.

Related

  • Project
    shareAI-lab/learn-claude-code

    A 17‑lesson tutorial that teaches you how to build the surrounding “harness” code for an LLM‑based coding agent (Claude). It walks through adding tools, permissions, context management, memory, task scheduling, multi‑agent coordination, and goal‑driven termination, all around a simple agent‑loop that lets the model decide when to call tools.

  • Project
    Donchitos/Claude-Code-Game-Studios

    A comprehensive template for Claude Code that organizes AI agents into a professional game development studio hierarchy to enforce structure, quality, and professional workflows.

  • Project
    rsmdt/the-startup

    A multi-agent AI framework for Claude Code that implements a spec-driven development workflow to turn ideas into shipped code via structured planning and parallel specialist agents.

  • Project
    muratcankoylan/Agent-Skills-for-Context-Engineering

    A collection of open-source Agent Skills focused on context engineering and harness engineering to help developers build production-grade AI agent systems.

  • Dispatch
    Scaling AI Coding: How Claude Code Navigates Large Codebases

    An exploration of agentic search and the 'harness' architecture used by Claude Code to manage multi-million line monorepos and legacy systems without relying on traditional RAG indexing.