We made a source-guided episode for engineers already using Claude Code or Codex, focused on what RTK changes at the command-output boundary and how the Rust repo is structured: https://youtu.be/5p0f2p9Gn0o
The design point we found most useful while reading the repo was the distinction between prompt-level guidance and a deterministic output proxy. Asking an agent to ignore noisy output still feeds it the noise. RTK tries to move that boundary into local code, with escape hatches for raw output.