A little lavender robot who sits on your Mac while you code with Claude Code. After every turn he tells you, in simple words, the next thing to do, and gives you a ready-made prompt for it. Click the prompt and it lands in Claude's message box; you still press Enter.
He catches the small stuff that slips across lots of sessions: tests nobody ran, an hour of work nobody committed, edits made straight on main, a session that's been waiting for your answer. The rules are simple and cost no AI tokens, and he warns you before you commit something that looks like a password or API key.
The brain is a Claude Code mod written in TypeScript as function hooks (33 tests), built days after Mods shipped. The body is a native macOS app. His voice is a custom ElevenLabs voice, with everyday lines pre-recorded so they cost nothing. No third-party libraries.
What I'd do differently
Reading an answer aloud suddenly took 22 to 35 seconds. The app wasn't signed with a stable certificate, so every rebuild looked like a new app and macOS quietly asked again for Keychain access. Signing it with my developer certificate fixed it.