@ruiconti yes – shared filesystem between the root and declared subagents. atm only the built-in agent copies share the root’s sandbox. declared subagents get their own, so the moment if want a custom system-prompts or tools you lose the shared files.
first post 👋
built a little launchpad for agent work.
T-0 = a real terminal in a browser tab + a pad to pick a repo and fire Claude / Codex / Pi / Cursor / Grok Build / etc. few clicks, off you go 🚀
under the hood: local PTY, xterm.js, and a Ratatui launcher (`t0`). plays nice in your usual browser (Dia, Helium, etc.) and with terminal multiplexers like https://t.co/qMW4cReWSG
made it for myself. local-only. MIT. open to contributions if you live in agent CLIs too.
https://t.co/5Mv92BbUvC
@marvinvonhagen@steipete more customizability and extended memory would be nice (for example edit and add memories + longer context) also "skills" would be really cool: customizable workflows that you could control in more detail.