ruvnet/metaharness

🛠️ The meta-harness for AI agents — scaffold your own focused, branded agent harness with its own npx CLI, MCP server, memory, learning loop, and witness-signed releases. Works with Claude Code, Codex, pi.dev, Hermes, OpenClaw, and RVM (hardware-isolated sandbox).

What it solves

MetaHarness is a factory for agent frameworks. It allows users to quickly generate a custom, repo-aware AI agent harness—including a CLI, coding agents, MCP servers, and governance policies—tailored to a specific GitHub repository or a blank slate.

How it works

It analyzes a repository's file layout to generate recommended agents, skills, and tools. The resulting harness is an npm-publishable package that the user owns and can customize. It supports ten different agent hosts (such as Claude Code, OpenAI Codex, and GitHub Actions) and uses a default-deny security posture for its MCP (Model Context Protocol) adapters.

Who it’s for

Developers and organizations who want a dedicated, versioned AI agent tailored to their specific codebase without building a framework from scratch.

Highlights

  • Rapid Scaffolding: Generate a custom harness in under 60 seconds via browser or CLI.
  • Darwin Mode: A self-improving mechanism where the harness mutates its own configuration and tests changes in a sandbox to measurably improve performance.
  • Multi-Host Support: Compatible with a wide range of hosts including Claude Code, Prime Agent, and GitHub Actions.
  • Model Routing: Includes a router to direct requests to the cheapest model capable of clearing a quality bar, reducing costs.
  • Weight-EFT: A tool to LoRA-tune open-source models (like GLM/Qwen) using the harness's gold-resolved archives to reduce reliance on frontier models.
  • Security Focused: Features a static-only scan (harness mcp-scan) to flag dangerous permissions and unguarded secrets.

Related

  • Project
    Darwin-Agent/HarnessX

    HarnessX is a composable agent harness foundry that decouples model selection from behavior logic, allowing developers to evolve agent performance through automated harness optimization and RL fine-tuning.

  • Project
    hoangnb24/repository-harness

    A tool that turns software repositories into agent-ready workspaces by providing a structured protocol and navigation tools to help AI coding agents follow authoritative project context.

  • Dispatch
    Beyond the Model: The Rise of Harness Engineering for AI Agents

    Explore the concept of Harness Engineering, a systematic approach to making AI coding agents reliable through environment design, state management, and explicit constraints.

  • Dispatch
    Understanding Agent Harnesses in AI

    An agent harness is a software environment that provides system prompts, tools, and agentic loops to transform a raw AI model into a functional AI agent.

  • Project
    autonomous-ai/openharness

    A unified command center for orchestrating multiple AI coding agents across various machines with end-to-end encryption and support for specialized toolchains.