I use a `justfile` to make it easy for the agent to run commands exactly how I want them.

Sometimes I add the GitHub Copilot agent as a reviewer to PRs to mix up the feedback.

I've been using LLMs for a while, but now I'm going all in.

I've been using the new Claude Code GitHub Actions to have Claude
- take a first pass at implementations directly from GitHub issues
- address PR comments and push changes directly to PR branch
- research and discuss tradeoffs of various approaches, and make recommendations

- when adding dependencies, use `cargo add` to get the latest version of a crate
- look up crate documentation on docs.rs when integrating newer crates
- break up large changes into smaller ones, going step by step
- explain every step at a high level to a professional software engineer

I use Claude Code (Sonnet 4) with the following directives in CLAUDE.md:
- ensure `cargo check` passes after completing every step
- run `cargo test`, `cargo clippy` and `cargo fmt` before committing anything, and explain any test failures (if expected)

I love running `bacon` in a separate terminal window so I can see how many errors/warnings I have in real-time: dystroy.org/bacon/