SuperagenticAI

SuperagenticAI / metaharness

Public

Meta Harness Implementation

superagenticai.github.iometaharness agentic-coding harness harness-engineering
19
3
100% credibility
Found Apr 09, 2026 at 19 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

An open-source Python library that iteratively optimizes the instructions, scripts, and workflows surrounding AI coding agents by generating proposals, validating changes, and selecting improvements based on scores.

How It Works

1
📰 Discover MetaHarness

You hear about a helpful tool that improves the instructions and scripts for AI coding assistants by automatically testing better versions.

2
📦 Install easily

You add the tool to your computer in moments so it's ready to use.

3
🏗️ Set up your workspace

You create a simple starting folder with basic instructions and check scripts for your AI helper.

4
🚀 Launch improvements

You start the magic: the tool asks an AI to suggest upgrades, checks them against your goals, and picks the winners.

5
🔍 Review the results

You browse friendly reports showing scores, changes, and side-by-side comparisons of the best ideas.

🎉 Enjoy better AI coding

Your AI assistant now has smarter instructions and reliable checks, making it safer and more effective every time.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 19 to 19 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is metaharness?

metaharness is a Python CLI tool for optimizing harnesses around agentic coding systems, like instruction files, setup scripts, validation logic, and tests. It runs an outer loop: a coding agent proposes workspace improvements, you validate and score them automatically, then keep the best candidate with full artifacts stored on disk. Inspired by the Meta Harness paper, it targets failures from poor repo instructions or broken tests, with Codex CLI as the validated backend and experimental Gemini, Pi, OpenCode support.

Why is it gaining traction?

Unlike ad-hoc prompt tuning, metaharness treats the entire harness as an optimization target, capturing environment snapshots and enforcing write scopes to avoid off-target edits. The CLI shines with commands like `metaharness run`, `inspect`, `ledger`, and experiment matrices for benchmarking backends—hosted Codex solves real Python benchmarks in one iteration per docs. Filesystem storage makes runs debuggable, standing out from black-box agent evals.

Who should use this?

Agent builders tuning workflows for tools like Codex CLI, or teams iterating on AGENTS.md and scripts in coding-tool projects. Ideal for devs harnessing meta learning in video stabilization pipelines or ecology analyses, akin to meta layers in GitHub repos like meta-raspberrypi, meta-freescale, or meta-openembedded—without the Yocto overhead.

Verdict

Alpha status (19 stars, 1.0% credibility) means expect rough edges, but excellent docs site, PyPI CLI (`uv tool install superagentic-metaharness`), and benchmark results make it a low-risk experiment for agent harness optimization. Skip unless you're deep in meta GitHub repos or quest harness tweaks.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.