Summary

This repository curates and ranks over 100 AI agent harnesses, orchestration frameworks, and techniques, providing a structured resource for developers and researchers. It addresses the critical need to evaluate harness quality, not just model capability, for deploying reliable agentic systems.

How It Works

An "agent harness" is defined as the runtime that transforms a model's reasoning into sustained, tool-using, error-recovering, goal-directed behavior. Architecturally, it acts as a mediator between raw AI capability and the environment, managing scheduling, permissions, memory, and policy enforcement. This project ranks harnesses by relevance, GitHub stars, adoption surface area, autonomy, and recovery, highlighting harness quality as key to production agent success.

Quick Start & Requirements

• Primary install: uvx agent-harnesses-mcp (requires uv or uvx).
• Machine-readable data: harnesses.json, llms.txt, and an MCP server are available for programmatic access.
• Contribution guidelines are available in CONTRIBUTING.md.

Highlighted Details

• Comprehensive Catalog: Features 100+ AI agent harnesses, categorized by function (e.g., Coding Agents, Frameworks, Multi-agent Orchestration).
• Data-Driven Ranking: Projects are ranked by GitHub stars (captured 2026-07-05) and evaluated on simplicity vs. capability, autonomy, and durability.
• Machine-Readable Outputs: