Know when to stop an AI task and inspect its work

Unexpected file changes, repeated tool errors or a request outside the brief deserve inspection. Stopping to read the events can prevent repeated attempts from hiding the original issue.

Start with a small handoff

Select a role by its working guide and deliverables, then verify the connected AI tool and workspace. Keep the first run small enough that you can inspect every output. The desktop needs to stay running while work executes. Marketplace sign-in, model access and file-generation tools are separate setup requirements; a success in one does not prove the others are ready.

Inputs for this workflow: The current task, latest events and expected output scope.

Work through the task

1. Read the newest relevant event.
2. Inspect the workspace and any pending prompt.

A brief you can adapt

After stopping, document what was written and what remains incomplete. Start a narrower follow-up only after understanding the partial state.

The filenames above are examples. Replace them with your actual inputs and destination before submitting the task.

Review the deliverable

Compare files with the pre-task copy and identify any action that needs manual follow-up.

Where this approach can fail

Cancelling execution does not undo files already written or external actions already accepted.

For the related desktop setup, see first task.