Agents can run shell commands, with an allowlist

Agents can now execute shell commands from chat under a configurable allowlist, plus a sidebar model selector and a Help menu with feedback.

Highlights

  • Run command tool: agents can now execute shell commands directly from chat, under a configurable allowlist so you stay in control. Running tasks show a read-only summary card.
  • New sidebar dropdown switches Claude models without leaving your session.
  • New Help menu with a direct Send Feedback path.

Fixes

  • MCP tools no longer get truncated by overly aggressive timeout caps on long-running tools.
  • The selected task isolation mode is now reflected correctly in the UI, and MCP import no longer drops authentication settings.

Improvements

  • Theme colors updated for better contrast across light and dark modes; clearer MCP import validation messages.

Related links

Share to