metaharness
starsforks
How to install metaharness?
ruvnet/metaharness
$git clone https://github.com/ruvnet/metaharnessnpx metaharness · open the Studio →
README.md
| 1 | <div align="center"> |
| 2 | |
| 3 | # MetaHarness |
| 4 | |
| 5 | ### Mint a custom AI agent harness from any repo. |
| 6 | |
| 7 | `npx metaharness` · [open the Studio →](https://ruvnet.github.io/metaharness/) |
| 8 | |
| 9 | <sub>(Repo: `ruvnet/metaharness` · CLI: `metaharness` · Library: `@ruvnet/agent-harness-generator`)</sub> |
| 10 | |
| 11 | [](https://ruvnet.github.io/metaharness/) |
| 12 | [](docs/USERGUIDE.md) |
| 13 | [](docs/ARCHITECTURE.md) |
| 14 | [](https://github.com/ruvnet/metaharness/actions/workflows/published-smoke.yml) |
| 15 | [](LICENSE) |
| 16 | |
| 17 | [](https://ruvnet.github.io/metaharness/) |
| 18 | |
| 19 | </div> |
| 20 | |
| 21 | --- |
| 22 | |
| 23 | ## What this is |
| 24 | |
| 25 | **Every serious repo deserves its own agent.** A repo-aware CLI, a repo-aware coding agent, a local MCP server, memory scoped to the project, skills generated from the actual file layout, governance policy, release verification, witness-signed provenance. |
| 26 | |
| 27 | `metaharness` mints those, on demand, from a GitHub URL or a blank slate. **It is not another agent framework. It is a factory for agent frameworks.** |
| 28 | |
| 29 | The model is replaceable. The harness is the product. |
| 30 | |
| 31 | ## What it gives you |
| 32 | |
| 33 | In under 60 seconds, in your browser, with nothing leaving your machine: |
| 34 | |
| 35 | - A custom AI agent harness for your repo (or any repo) |
| 36 | - Recommended agents, skills, slash commands, MCP tools |
| 37 | - A scoped memory namespace + governance policy |
| 38 | - Witness-signed provenance + release gates |
| 39 | - Drops into Claude Code, OpenAI Codex, pi.dev, Hermes, OpenClaw, or RVM — pick one or all |
| 40 | |
| 41 | Output is an npm-publishable `.zip` with **your name on it, your branding, your `npx <your-name>` CLI**. |
| 42 | |
| 43 | ### New |
| 44 | |
| 45 | - **Score any repo before you build it.** `npx metaharness score <repo>` reads |
| 46 | the repo (never runs it) and prints a one-screen report card — how well a |
| 47 | harness fits, how likely it is to build, how safe the tools are, and the |
| 48 | rough **cost per run** — so you know what you'll get before scaffolding. |
| 49 | - **Pick the cheapest model that's good enough.** [`@metaharness/router`](https://www.npmjs.com/package/@metaharness/router) |
| 50 | routes each request to the right model from your own results — same quality, |
| 51 | far less spend. Works out of the box with zero native deps; **train it on your |
| 52 | data** for a sharper fit (`npm i @metaharness/router`). Add the optional |
| 53 | [`@ruvector/tiny-dancer`](https://www.npmjs.com/package/@ruvector/tiny-dancer) |
| 54 | to train a fast native model instead — same training data, no API change. |
| 55 | - **Let your harness improve itself.** Every scaffold now ships with **Darwin Mode** |
| 56 | ([`@metaharness/darwin`](https://www.npmjs.com/package/@metaharness/darwin)) wired in — |
| 57 | run `npm run evolve` and the harness mutates its own config, tests each change in a |
| 58 | sandbox, and keeps only what *measurably* improves. The model stays frozen; the harness |
| 59 | evolves. Safe by default (no network, no API key; pure refactor/tuning behind a safety |
| 60 | gate). Validated on real **SWE-bench Lite** bug-fixing. `--no-darwin` to skip. |
| 61 | - **Distil the cheap tier instead of escalating to a frontier model.** **Weight-EFT** |
| 62 | ([`@metaharness/weight-eft`](packages/weight-eft/), `metaharness weight-eft`) takes the |
| 63 | *complementary* lever to Darwin's gradient-free evolution: it exports the harness's gold-resolved |
| 64 | archive into standard SFT/DPO sets and LoRA-tunes the open cheap tier (GLM/Qwen), so the |
| 65 | cost-cascade escalates to Opus/GPT **less often**. It attacks **cost** (fewer $0.50 escalations), |
| 66 | not the frontier ceiling — and stays honest about it. Strict train/eval-disjointness + |
| 67 | reward-hacking filter |
People who liked this also viewed
6 picksType
- fastmcp🚀 The fast, Pythonic way to build MCP servers and clients.MCP ServersJul 2026
- playwright-mcpPlaywright MCP serverMCP ServersJul 2026
- ext-appsOfficial repo for spec & SDK of MCP Apps protocol - standard for UIs embedded AI chatbots, served by MCP serversMCP ServersJul 2026
- langchain-mcp-adaptersLangChain 🔌 MCPMCP ServersJul 2026
- mcporterCall MCPs via TypeScript, masquerading as simple TypeScript API. Or package them as cli.MCP ServersJul 2026
- mcp-proxyA bridge between Streamable HTTP and stdio MCP transportsMCP ServersJul 2026
