Skip to content

Enabling Agents

Agents mode must be enabled before you start a task. Once a task begins, the setting is locked for that session.

Two ways to toggle agents mode:

In the action bar at the bottom of the window, click the Agents toggle to switch between enabled and disabled.

  1. Press / to open the command menu
  2. Select Enable Agents or Disable Agents

The current state is always visible in the action bar before you submit your prompt.

Once you submit your first message and the task begins, the agents setting cannot be changed. This ensures consistent behavior throughout the task lifecycle-the supervisor and any spawned subagents operate under the same configuration from start to finish.

If you realize mid-task that you need agents (or don’t), you’ll need to start a new task with the correct setting.

When you fork a task, the new task inherits the agents setting from the original:

  • Fork from an agents-enabled task → new task has agents enabled
  • Fork from an agents-disabled task → new task has agents disabled

This preserves consistency when branching off to explore alternatives. The forked task continues with the same architecture that was handling the original work.

Before starting, consider your task:

Task TypeRecommended
”Fix this typo”Agents off
”Refactor authentication across the app”Agents on
”Add a console.log here”Agents off
”Implement a new feature with tests”Agents on
”Explain this function”Agents off
”Migrate database schema and update all queries”Agents on

When in doubt: if your task touches many files or requires holding lots of context, enable agents. If it’s quick and focused, keep them off.