An input-required workflow pauses an A2A task when an agent needs clarification, then lets the client reply so the same task can continue. section 10.1.3
The exchange works as follows:
- The agent marks the task input_required and sends a Message with context or a clarifying question. section 10.1.3
- The server stores the task; the client replies with a new Message associated with the existing Task ID. section 10.1.3
Clarifying a Hailstone request
A coordinator sends “start at the number four” to a CrewAI Hailstone agent. The remote agent asks which positive integer to use; the coordinator resubmits “4” with the existing task ID, and the task completes. section 10.2.3
Figure 10.7 The sequence of interactions for an input-required workflow

Sequence diagram of the client and remote agent exchanging messages when input is required. section 10.1.3
The demonstrated client-side integration supports this flow, but its server-side LLMAgent execution does not: each run starts a fresh rollout. Continuing server-side would require passing an ongoing rollout to LLMAgent.run(). section 10.4.2
