Initially just a few sentences on what the project is about and the tech stack. Can be as little as 2 sentences + 5 bullet points.

After that it’s things I see the LLMs get wrong in practice. For almost all projects it’s when to use which build/test command + a line to use pnpm/cargo/… add instead of modifying dependency files directly + „don’t add any code comments unless specifically requested”. The rest varies a lot by project.

A letter from me addressed to the agent, describing me and the goals of the project. Not AI-generated, no code or commands.

Nothing specific to the technology, because that gets out of sync fast, or might steer the agent in the wrong direction.

Smart models can figure these things out themselves.

If for example the agent repeatedly wrongly guesses the command to run tests, I prefer to change the code so the wrong guess will actually work the next time. Or I will find what in the project lead to the wrong assumption and fix that.