lopopolo/harness-engineering

This repository is a collection of writing, not a piece of software you install and run. It is Ryan Lopopolo's anthology and field guide on a practice he calls shaping the working environment around an AI coding agent to get better results from it, while treating the underlying AI model and coding tool itself as fixed. The core idea is that you improve outcomes by working on two things around the agent: the context it is given and the tools it can use, plus how that surrounding setup is organized. The writing argues that a well built working environment should carry an organization's real requirements, things like reliability, security, how maintainable something is, performance, and how much risk is acceptable, and should also carry local decisions about how to balance those requirements against each other. It also argues that because doing this kind of work happens repeatedly over time, a good environment can turn past lessons, corrections, and failures into context, boundaries, and checks that make later work more consistent, rather than starting from scratch each time. Practically, the repository is meant to be pointed at by a coding agent alongside whatever system that agent is meant to improve. An included AGENTS.md file routes the agent to the relevant essays, examples, and supporting proof. Readers who want the core argument can start with a thesis index, and those looking for a concrete application can look at a set of included playbooks. The repository also links out to related essays and sources that influenced this thinking. This is intended for people building or maintaining AI coding agents at an organizational level, especially those thinking about how to make agent behavior consistent and improvable over time rather than a one off experiment. Repository authored material is licensed under CC BY 4.0, which requires giving credit to the original author when you reuse it.