Cubicle shows your AI agents as people in a pixel-art office: they walk to a desk when they start working, raise a hand when they need you, and go for coffee when they’re done. Last week’s post was about answering them from Telegram. Here’s what changed since.
The office grows sideways. A big company on a wide screen used to stack more and more rows. Now the office opens extra desk wings so its shape follows the window; phones and tall windows still get rows.
Approvals you can read. When a Paperclip agent asks the board to approve something, clicking its raised hand now shows the approval’s title, summary and risks, and the button opens Paperclip’s approval page directly. Telegram messages carry the same summary. Cubicle stays read-only: you decide in Paperclip.
Fewer false alarms. A failed run that was dealt with (agent paused and resumed, or a later run finished) no longer leaves “Last run failed” on the card. Telegram uses the same rule, and a “needs you” message is marked done only after it has been answered.
More agents in the office, thanks to contributions from Serkan Özcan: a read-only Codex Desktop/CLI observer, feed examples for GitHub Copilot, Antigravity and Grok CLI, a macOS LaunchAgent setup, and Telegram access controls (choose exactly which Telegram users can pair and get alerts).
Try it: npx @caglarutkuguler/cubicle, or install it as a Paperclip plugin with npx paperclipai plugin install @caglarutkuguler/cubicle.
Repo: https://github.com/caglarutkuguler/cubicle
The open theme ideas (hospital, kitchen, classroom, newsroom) are still up for grabs. A theme is one file; docs/THEMES.md explains how to write one.
Top comments (0)