baskduf/harness-starter-kit

This repository is a starter kit for what the author calls "harness engineering," which is the practice of setting up a code repository so that AI coding agents work more reliably inside it. The core idea is that instead of spending time crafting better prompts for your AI assistant, you invest that effort into the repository itself: adding clear instructions, rules, memory files, and feedback mechanisms that the AI reads automatically whenever it works on your project. A harness, as defined here, is the combination of six things baked into the repo: instructions that tell the AI how the project works, constraints that limit risky behavior, feedback loops that let the AI verify its own changes, memory files that preserve decisions across sessions, evaluation checks that confirm quality, and governance rules about what the AI is and is not allowed to change. The starter kit gives you templates and a workflow for setting these up. You adopt it by giving your AI coding agent a single prompt that points it at this repository and asks it to analyze your project and add only the harness pieces it actually needs. The agent inspects your existing setup, avoids overwriting what is already there, and produces an adoption report listing what it changed and what remains for you to handle manually. Once adopted, four slash commands are available inside your project. "Harness doctor" scans the repository and gives it a score across the six harness categories, telling you where the gaps are. "Harness update" pulls in newer guidance from this starter kit and selectively applies it. "Harness refresh" looks for stale or duplicated guidance in your existing harness files. "Harness review" challenges any proposed changes from a critical angle before you commit them. The kit supports Python, TypeScript, Node, React, Vue, Django, Flask, FastAPI, Spring Boot, and Android projects. It is available in English, Korean, Japanese, and Simplified Chinese.