alchemiststudiosDOTai/harness-engineering

99 stars · Last commit 2026-03-17

harness-engineering discussion of shortcuts, automation, hacks and overall productivity with code agents like claude code, codex, and other harness.

README preview

# Harness Engineering

This repository is an **active attempt to formalize and implement Harness Engineering**—a mechanical, repo-enforced framework for making large (often agent-driven) code changes safe, reviewable, and incremental.

Inspired by [OpenAI's work on Harness Engineering](https://openai.com/index/harness-engineering/), this project provides the scaffolding, tooling, and conventions necessary for humans to steer while agents execute.

---

## What is Harness Engineering?

Harness Engineering is a set of constraints and feedback loops that enable high-throughput agent-driven development without architectural decay. It treats the repository as a **system of record** for everything agents need to know.

Key insight from OpenAI's experience:
> "What Codex can't see doesn't exist."

Knowledge in Slack threads, Google Docs, or engineers' heads is illegible to agents. The solution: encode everything—architecture, taste, quality standards—into mechanical, verifiable structures in the repo itself.

---

## Core Principles

View full repository on GitHub →