About this project, https://github.com/Barrixar/copilot-instructions.md/blob/main/copilot-instructions.md

This is a drop-in file to be your "copilot-instructions.md" in [repository root]/.github (The folder called .github) which your IDE will load for AI chat interactions. It aims to improve engineering quality and guarantee deep thinking. Responses will take longer, but will be consistently better, with less bugs and regressions and more thought into the bigger picture and broader design of your program and its stack.

It punishes AI labs that cut costs and with that, corners. It makes AI work and 'vibe coding' more suitable for real-world application with less risk for stability, security issues. Of course, this doesn't eliminate your responsibility to audit AI-written code, but it will transform expectations of poor work to quite good.

This copilot-instructions file will be constantly updated to close loopholes and ways that an AI agent, cutting costs and just trying to get the job done, would seek to work around it and bypass important provisions (Due to ambiguity and such). The initial revision of the .MD file is already a result of many months of hardening against many classes of these scenarios.

What I am doing is clearly a form of prompt engineering. This particular instructions file began written from scratch by me, and was for personal use ever since, even as it matured due to constant revisions and practical results. Now, I believe it's time to share this goodness with the OSS community, as it cacn shift your outcomes when coding with AI agents, and we are currently at risk of "AI eating itself" due to rigid pre-training data and patterns (or shall I say, tendencies to introduce bugs), so projects like mine could help to alleviate it.

Specifically, to avoid AI eating itself and introducing real bugs, this instructions file enforces that the Agent truly understands every aspect of its planned changes, the surrounding context, and the bigger picture of your codebase & intricate workings. It pushes the agent to produce work of a super high engineering quality, everything considered, as if a senior dev on your team that does a lot of investigation before acting, and doesn't act on a whim. This should be the more organic footprint of code changes produced by an agent using this instructions file, which is hugely different in nature than pre-trained output, therefore boosting correctness, staability, and working against the "AI eating itself eventually" concept. We only have this global concern because AI labs try to squeeze cost-efficiency, because of course they would know how to bake in the project that i have going on here (The specifics of this custom copilot-instructions.md file).