agentclash-multi-turn-operator

To provide human input during a multi-turn AI agent evaluation, you submit operator messages directly into the agent's transcript. This unblocks the agent and allows the evaluation run to progress through its specific challenge pack phases.

Human-in-the-loop interaction for paused agent evaluation runs works by monitoring the turn status to detect when an agent is blocked. Once blocked, you inject a user response into the transcript to resume the evaluation process.

You need a valid workspace ID, run ID, and run agent ID to manage multi-turn agent evaluation turns. These identifiers are required to interface with the AgentClash API and monitor or submit messages.

Yes, you can monitor when an AI agent is blocked awaiting human feedback in a live evaluation. The skill detects the exact turn status to identify when manual verification or input is required.

You need to inject operator messages into an agent's transcript when a complex evaluation scenario requires manual verification. This typically occurs when a customer support agent reaches a phase where it cannot progress without human input.

The best way to resolve bottlenecks in multi-turn AI agent evaluations is to enable seamless human intervention. This allows you to provide necessary human responses directly when an agent pauses to await operator input.