Agent skill · software engineering · aradotso
meta-harness-optimization
Framework for automated search over task-specific model harnesses — the code around a fixed base model that decides what to store, retrieve, and show while the model works.
Why this skill is useful
Adds executable scripts and a framework for optimizing model harnesses that enhance AI capabilities in task-specific contexts.
What it needs
Requires uv installed locally. About 6k tokens when loaded. Last updated 2026-07-09. 70 stars on the source repository.
What this skill does
Meta-Harness Optimization Skill by ara.so — Daily 2026 Skills collection. Meta-Harness is a framework for automated end-to-end search over model harnesses — the scaffolding code around a fixed base model that controls what the model stores, retrieves, and sees while working on a task. Rather than hand-crafting prompts and memory systems, Meta-Harness proposes, evaluates, and evolves harness implementations automatically. Paper: Meta-Harness: End-to-End Optimization of Model Harnesses Homepage: https://yoonholee.com/meta-harness/ Core Concepts Term Meaning ------ --------- Harness All code around the base model: memory, retrieval, prompt construction, tool use Proposer Agent LLM (e.g. Claude Code) that proposes new harness variants Evaluator Runs proposed harnesses on a benchmark, returns a score Meta-Loop Iterative propose → evaluate → feedback cycle Installation Meta-Harness uses uv for dependency management. Each reference experiment is self-contained: No global pip install is needed. All dependencies are managed per-experiment via pyproject.toml. Quick Start Text Classification (Memory System Search) Terminal-Bench 2 (Scaffold Evolution) Applying Meta-Harness to a New Domain The recommended workflow uses the onboarding document with your AI coding assistant: Minimum Required Components for a New Domain Implementing a Harness A harness wraps a base model and manages context/memory/tools: Implementing the Evaluator The Meta-Harness Loop Proposer Agent Wrapper (Claude Code) The shipped examples use Claude Code as the proposer. …
How to use it
Reference it in AdaL, Claude Code, Cursor or any coding agent — nothing to install:
@skills aradotso/meta-harness-optimization