Summary

This repository curates resources for "agent harness engineering," focusing on creating reliable environments, constraints, and feedback loops for AI coding agents. It targets engineers and researchers aiming to scale AI-driven code generation, providing a structured approach to making agents dependable for production use with minimal human code writing.

How It Works

The core concept is the "agent harness"—infrastructure wrapping LLM coding agents, managing sessions, context, tools, and recovery. Key principles include human steering, repository knowledge as the source of truth, mechanical architectural enforcement, and agent legibility. This approach emphasizes the harness's critical role over the model itself for reliable, large-scale AI code production.

Quick Start & Requirements

This is a curated list of tools and resources, not a runnable project. It categorizes and links to components like full-lifecycle platforms (Chorus), orchestrators (Vibe Kanban), task runners (Symphony), agent runtimes (OpenClaw), coding agents (Claude Code, Codex), and standards (MCP). Setup and requirements depend entirely on the individual linked projects.

Highlighted Details

The list offers a comprehensive taxonomy of the agent harness ecosystem, categorized into:

• Platforms & Orchestrators: Solutions for end-to-end development and parallel agent execution.
• Task Runners & Frameworks: Tools bridging issues to PRs and building custom harnesses.
• Runtimes & Memory: Infrastructure for persistent agent capabilities.
• Coding Agents: Specific LLMs for code generation (Claude Code, Codex, OpenCode).
• Standards & Protocols: Emerging conventions for agent communication and configuration (MCP, ACP).
• Seminal References: Foundational articles and talks from OpenAI, Anthropic, and industry experts.

Maintenance & Community

Presented as a community-driven effort with an invitation for contributions. Specific maintainer details, sponsorships, or dedicated community channels are not listed. The focus is on collaboratively expanding the curated knowledge base.

Licensing & Compatibility

No specific license is declared for the curated list itself. Licensing and compatibility terms are determined by the individual projects linked within. Users must consult each referenced tool's license for commercial use or integration.

Limitations & Caveats

As a curated list, this repository is a pointer to other projects, not a unified system. Its value depends on the maturity and documentation of linked components. The absence of a declared license for the list necessitates careful review of each referenced item. The rapidly evolving field requires ongoing community maintenance to remain current.