Tests of how AI coding agents handle API keys, with every transcript, fixture and script published.
202 runs of Claude Code, Codex and six other models against a .env of fake keys: who reads the file, what leaks through program output, and what masking stops.