Mission execution and human intervention
Monitor Task execution progress in a Mission, respond to Human-in-the-Loop (HIL) intervention requests, and dynamically add new requirements to maintain full visibility and control over the AI's execution.
Task trigger methods
Tasks support the following trigger methods. Once triggered, a Task enters the running state and begins producing an output stream.
|
Trigger method |
Description |
|
Scheduled (cron) |
Runs automatically on a recurring schedule defined by a cron expression. Example: run a cluster inspection every day at 02:00. |
|
Manual |
You trigger the Task on demand by entering a conversational command on the Mission details page. |
View the execution output stream
When a Mission starts, the Digital Employee (Agent) produces a real-time stream of reasoning and actions for each Task. You can monitor the AI's reasoning process, tool calls, and execution results as they happen.
Steps
-
Log in to the STAROps console, navigate to the target Mission, and click the Execution tab.
-
In the Task execution record list, click Go to Session in the Actions column for the target record.
-
The output stream appears in a conversational view, showing each step of the AI's reasoning and actions.
The output stream includes the following types of information:
|
Information type |
Description |
|
Reasoning process |
The AI's analytical reasoning, showing how it interprets the situation and determines the next action. |
|
Tool invocations |
Names, input parameters, and results of tools called by the AI through MCP services or built-in integrations. |
|
Execution results |
The output and status of each step as it completes. |
|
HIL requests |
Requests generated when the AI reaches a decision point that requires human confirmation. Execution pauses until you respond. |
Human-in-the-Loop (HIL)
Human-in-the-Loop (HIL) is a core safety mechanism in STAROps. When the AI encounters a high-risk operation, an uncertain decision, or a scenario that requires business judgment, it pauses execution and sends an intervention request. Execution resumes only after you confirm.
When HIL is triggered
HIL requests are triggered in the following scenarios:
-
High-risk operations: The AI is about to perform a change that could affect a production environment, such as restarting a service or modifying a configuration.
-
Insufficient information: The AI cannot make a reliable decision based on available data alone and needs additional business context from you.
-
Multiple remediation options: The AI has identified several viable approaches and needs you to select the most appropriate one.
-
Anomaly confirmation: An unexpected situation has arisen during execution, and the AI needs confirmation on whether to continue.
Respond to an HIL request
When an HIL request is generated, follow these steps to respond:
-
Receive the AI assistance notification through your configured notification channel.
-
Review the AI's analysis and the details of the request to understand what decision is needed.
-
Enter your response in the conversational interface — for example, "Approve Plan A" or "Do not proceed; switch to Plan B."
-
The AI receives your response and continues execution according to your instructions.
If an HIL request is left unanswered for more than 3 hours, the Task automatically fails. Configure notification channels to ensure you receive HIL alerts promptly.
Add dynamic requirements
While a Mission is running, you can add new requirements or adjust existing Tasks through the conversational interface without stopping the Mission.
Steps
-
Navigate to the target Mission details page.
-
Enter your new requirement in the conversational interface — for example, "Add a resource cleanup task to run every Friday afternoon."
-
The AI analyzes the request and automatically updates the execution Blueprint. The new Task appears in the plan list.
Dynamically added Tasks have the same execution capabilities as Tasks defined at Mission creation time. Adding a new Task does not affect any Tasks that are currently in progress.