Work safely in a WorkspaceCopy Anchor Link

Workspace lets you work with an AI Agent in a selected local project and review the work as it proceeds. Begin with a small, reversible task so you can learn how your setup behaves.

Open a project and choose a modelCopy Anchor Link

1. Open Workspaces in Lumoswitch.
2. Select or open the local project you want to work on. Check the project name and folder before sending instructions.
3. In the composer, open the Configuration menu and choose which model configuration to use. This selects a connection, not a specific model. If no suitable configuration appears, select Add configuration and choose where to create one, or set it up under Models and return to the Workspace.
4. In the separate Model selector, choose the model available from that configuration. Confirm both selections before starting. If you switch the configuration, check the Model selector again.
5. If you want a reusable capability, review and select an installed Skill. Only select Skills you trust for the task.
6. Before a first task, check Default permissions in the access-permissions menu and leave auto approval off. Review each approval request and decide whether to approve it. Auto approval, if enabled, can allow file changes, commands, network access, and access outside the project without asking each time; a prompt asking the Agent to avoid tools does not enforce that restriction.
7. Enter a narrow request with a clear result, such as asking the Agent to explain one file or suggest a small change.

Review actions and approvalsCopy Anchor Link

Read the plan and task updates as they appear. Inspect proposed file changes and tool actions before approving them. Some operations wait for your decision; a waiting state means the action has not been approved yet. Approve only actions you understand and want performed in the selected project.

Do not use a first run for deleting files, changing credentials, publishing, or other hard-to-reverse work. Keep secrets out of prompts and project files unless you have verified that the selected model and tools are appropriate for them. Default permissions ask when an action needs approval; this is not an operating-system sandbox, and the Agent may still have access to project files and tools needed for the task.

Stop or continueCopy Anchor Link

• Use Stop when you want the current work to end. The app may show that a stop was requested while it confirms the run has actually stopped; wait for the status to update.
• If work is waiting for approval, review the request and choose whether to approve or reject it. Do not assume the Agent will continue without that decision.
• After a stop or interruption, review the Workspace status and recent activity before continuing. Resume only when the displayed state offers a continuation and the next action is clear.
• If Lumoswitch or the Agent reports a recovery or confirmation state, follow the on-screen status. Do not start a duplicate task while the earlier run's outcome is unknown.

Check the resultCopy Anchor Link

Compare the changed files or answer with your request. Confirm the result meets the task's goal before using it elsewhere. Workspace history and activity help you inspect what happened; they do not replace reviewing the actual project changes.

If work is stuck or unexpectedCopy Anchor Link

• If no model is available, return to Models and confirm that a compatible connection is ready.
• If an operation is waiting, inspect its approval request and the project path before deciding.
• If Stop remains pending, wait for confirmation and check the run status before retrying or creating another task.
• If the Agent changes an unexpected file, stop the run if possible, inspect the project diff, and restore the file using your normal project backup or version control process.
• If the project or history cannot be opened, use the Workspace's displayed error and recovery guidance before repeating work.

For a first run, see Quick start. Learn about Agents and Skills.