MetaHarness
ruvnet/metaharness
Generator that mints a branded, repo-aware agent harness with its own npx CLI, MCP server and memory
MetaHarness is a generator that builds a custom AI agent harness for a repository, packaged as an npm-publishable zip with its own CLI and MCP server. The output can target Claude Code, Codex, pi.dev, Hermes, OpenClaw or RVM.
What MetaHarness does
MetaHarness takes a GitHub URL or a blank slate and produces a repo-aware agent harness: a CLI, a coding agent, a local MCP server, project-scoped memory, skills generated from the actual file layout, a governance policy and release verification with witness-signed provenance. You run it with npx metaharness or in the browser Studio, where the README says nothing leaves your machine. The output is an npm-publishable .zip carrying your name, your branding and your own npx CLI.
The README describes the tool as a factory for agent frameworks rather than another framework, and says the model is replaceable while the harness is the product. The recommended agents, skills, slash commands and MCP tools it produces can drop into Hermes among several hosts. Newer parts, including a governed variation runtime called AVO and an experimental ARC-AGI-3 harness exposed as a ChatGPT remote MCP server, are labelled private and experimental, and the README states that no ARC performance claim is made yet.
Key features
- Generates a repo-aware CLI, coding agent and local MCP server
- Project-scoped memory and skills generated from the file layout
- Governance policy, release gates and witness-signed provenance
- Browser Studio and npx metaharness entry points
- Targets Claude Code, Codex, pi.dev, Hermes, OpenClaw and RVM
When to use it
- Turning a repository into its own branded agent CLI
- Shipping a Hermes-compatible harness with an explicit tool policy
- Publishing an agent package to npm with signed release provenance
Who it is for: Developers and teams who want a focused, repo-specific agent harness instead of a general agent framework.
How it fits with Hermes Agent
A general harness generator that lists Hermes Agent as one of several supported targets, alongside Claude Code, Codex, pi.dev, OpenClaw and RVM.
How to install MetaHarness
These commands are copied from the project's README. Check the repository for the latest steps before you run them.
npx metaharnessNote: The README describes its ARC-AGI-3 packages as private and experimental.
FAQ
What is MetaHarness?
MetaHarness is a factory for agent harnesses. It takes a GitHub URL or a blank slate and generates a repo-aware CLI, MCP server, memory scope, skills and governance policy packaged as an npm-publishable zip.
Does MetaHarness work with Hermes Agent?
Yes, as one of several targets. The README says the generated harness drops into Claude Code, OpenAI Codex, pi.dev, Hermes, OpenClaw, RVM or Prime Agent, and the repository is tagged hermes-agent.
Is MetaHarness free and open source?
Yes. The repository is published under the MIT license.
Similar dev tools for Hermes Agent
All dev toolsOpen-source voice AI SDK that gives an agent a phone number, with an installable skills bundle
wk42worldworld CyberCodeClaude Code style desktop and terminal agent with persistent memory and self-evolution
mvanhorn agentcookieSyncs your Mac's Chrome sessions to the Linux box your agent runs on, over Tailscale
es617 claude-replayTurns Hermes Agent, Claude Code, Codex and other coding agent sessions into self-contained HTML replays
sbhooley AI Native Lang (AINL)Compact graph-based language for deterministic AI agent workflows, installable into Hermes over MCP
firecrawl Firecrawl CLICommand-line web search, scraping, crawling and browser interaction with skills for coding agents
Related guides: How to install Hermes Agent · Run multiple Hermes agents with profiles