Scaffold repo-specific agent harnesses with Metaharness
Use Metaharness to score a repository, generate a branded agent harness with CLI, MCP, memory, and safety defaults, then package it for repeatable team use.
npx skills add agentskillexchange/skills --skill scaffold-repo-specific-agent-harnesses-with-metaharness
# Scaffold repo-specific agent harnesses with Metaharness
What this skill actually does
Use this skill when an operator needs to turn a repository or workflow lane into a repeatable agent harness instead of hand-writing prompt files, MCP glue, memory folders, release gates, and npm packaging from scratch. Metaharness generates focused, branded harnesses for Claude Code, Codex, OpenClaw, OpenCode, Hermes, pi.dev, RVM, and related agent hosts.
## What the agent/operator does
1. Inspect the target repository and decide whether it needs a dedicated coding, support, research, devops, or other vertical harness.
2. Optionally run the Metaharness scoring flow against the repo to estimate harness fit, likely build success, tool safety, and rough run cost before scaffolding.
3. Generate a harness with the desired template and host target, such as Claude Code, Codex, OpenClaw, OpenCode, Hermes, pi.dev, RVM, or GitHub Actions.
4. Review the generated CLI, MCP server policy, memory namespace, release gates, and witness-signed provenance files before use.
5. Tune the generated harness configuration, tools, model routing, and safety boundaries to match the team’s repository and operating rules.
6. Package or publish the generated harness under the team’s own name so teammates can invoke the same workflow through a stable npx command.
## When to invoke
Invoke this instead of using an agent product normally when the goal is to create a reusable agent harness that other operators can run consistently across a team or repo family. It is useful when a one-off prompt has grown into a maintained workflow that needs installation, versioning, MCP policy, memory, model routing, and release provenance.
## Scope boundary
This is not a generic agent framework listing or a broad multi-agent architecture card. Keep the workflow scoped to using Metaharness to score, scaffold, inspect, tune, and package a concrete repo-specific agent harness. Do not frame it as a replacement for the underlying agent host or as a guarantee of autonomous production deployment.
## Requirements
– Node.js and npm/npx.
– A target repository or workflow lane to scaffold into a harness.
– The Metaharness CLI or web Studio.
– Optional access to the target agent host, such as Claude Code, Codex, OpenClaw, OpenCode, Hermes, pi.dev, RVM, Cursor, or GitHub Actions.
## Installation
Run the CLI directly with npx:
npx metaharness –wizard
npx metaharness my-bot –template vertical:coding –host claude-code
Metaharness also publishes host-specific wrappers such as @metaharness/claude-code, @metaharness/codex, @metaharness/openclaw, @metaharness/opencode, @metaharness/hermes, and @metaharness/github-actions.
## Verification notes
Source metadata was verified from the exact GitHub repository, npm registry, and upstream README. As of this intake run, ruvnet/metaharness had 529 GitHub stars, 64 forks, an MIT license, same-day push activity, and the metaharness npm package showed 29,085 downloads in the last week. The README documents repo scoring, wizard generation, host-specific wrappers, MCP policy output, release gates, memory, and npm-publishable generated harnesses.