Available outputs
Manual approved runs can prepare editable prompts, tasks, handoffs, notifications, skill requests, model requests, and approval checkpoints.
No model invocation
A prepared model request is a draft. The runtime does not send it to a model automatically.
No project or file writes
Draft outputs do not silently create tasks, write files, or mutate project state.
No external sends
The runtime does not post notifications, send email, call arbitrary networks, or launch shell commands.
Read the side-effect receipt
Each executed draft run records what was prepared and confirms that blocked side effects did not occur.
NEED A GUIDED ANSWER?
Ask LobeWork about this guide.
Ask LobeWork uses the same published article library and links back to its source guidance.
WAS THIS GUIDE HELPFUL?Help us improve the documentation.