# green goblin

- **Event:** [Built with Opus 4.6: a Claude Code hackathon](https://cerebralvalley.ai/e/claude-code-hackathon)
- **When:** Feb 10 at 12:00 PM – Feb 17 at 10:00 AM (EST)
- **Where:** Location TBA
- **Team:** [Nishant Joshi](https://cerebralvalley.ai/u/nishantjosh)
- **GitHub:** https://github.com/NishantJoshi00/notch
- **Demo video:** https://youtu.be/BKEBYKLFEYw
- **Gallery:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery/113

Every AI agent optimizes for what happens during a conversation. Notch treats the silence between conversations as the product.

It's a macOS menu bar app. Double-tap Option, say something, close it, go back to work. It stays running. Every 45 to 90 minutes, a timer fires - the system takes a screenshot of your desktop, loads conversation history, pulls journal entries from the filesystem, and sends it all to Claude. The model returns one of two tool calls: `send_message` or `stay_silent`.

Most calls return `stay_silent`. Not a failure to respond - an explicit output. The model processed your screen, its own accumulated journal, and produced: nothing. Sixteen times a day, it looks and says nothing.

Its identity is a hardcoded enum it cannot modify, even through its own tools. Its behavior prompts live on disk, and the model can rewrite those. Over months, behavior shifts. Identity doesn't.

4,300 lines of Swift. No dependencies. No database - the filesystem is the memory.

---

Markdown version of https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery/113. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt