Harness Engine: a Codex skill for agent-first repository harnesses

[tomatopunk]

Hey everyone 👋

I’d like to share a project I’ve been working on: Harness Engine.

GitHub: https://github.com/hallucination-studio/harness-engine

Harness Engine is an open-source Codex skill that bootstraps an agent-first repository harness for real projects.

The idea is inspired by OpenAI’s “Harness engineering” article: if we want Codex and other AI coding agents to work reliably in complex codebases, the repository itself needs to become the system of record.

Harness Engine helps by:

• analyzing the target repository
• generating a short routing-style AGENTS.md
• creating durable docs like ARCHITECTURE.md, docs/RELIABILITY.md, docs/SECURITY.md, and docs/QUALITY_SCORE.md
• setting up execution-plan folders for active/completed work
• adding SOPs for architecture setup, knowledge capture, observability, and UI validation
• enforcing a local closed-loop workflow around acceptance criteria, quality scoring, continuation decisions, and plan closure

The goal is not to blindly copy a template, but to help Codex create and maintain a repo-specific harness so future work is more recoverable, reviewable, and less likely to drift.

Install locally:

npx @hallucination-studio/harness-engine install --local