wtf is harness-for-codex?

This repository is a minimal starting template for projects where you plan to do AI-assisted coding work with Codex. The idea is that any new task done in such a repo begins from a predictable layout, so the next agent or person can find the same set of commands and the same conventions every time. The core of the template is a set of shell scripts in the scripts/ folder. scripts/bootstrap installs or prepares dependencies when a known stack is detected. scripts/check runs formatting, linting, type checks, and tests if those tools are available. scripts/test runs the test suite on its own. scripts/eval runs doctor, bootstrap, and check together as a handoff gate. scripts/doctor prints readiness information about the repo and its tooling. scripts/hooks installs optional local pre-commit hooks. If you have the just command installed, the same operations are exposed in a justfile, so you can type just check or just eval and get the same behaviour. There is also a small YAML metadata file at the root that records the canonical command names, the expected documentation files, and the stages of the task loop. The author suggests keeping that file aligned with the AGENTS.md doc and the scripts. The suggested workflow is to write a task brief in the tasks/ folder when context is needed, make the code changes, record any lasting decisions in docs/decisions.md, and run scripts/check before wrapping up. The repo also ships a minimal devcontainer config that runs the bootstrap script after the container is created. Using Docker is optional, local work is fine without it. The scripts are described as safe defaults for an empty or early-stage repository, meant to be extended as the project grows.