Description

The specification needs to clarify the agent's responsibilities and flexibility regarding including state messages in the Task message history.

Questions to Address

1. Agent discretion: Are agents allowed to decide whether to include state messages (messages from TaskStatus objects) in the Task message history?

2. Requirements: Should state messages:

   • Always be included in the history?
   • Never be included in the history?
   • Be optional at the agent's discretion?

3. Client awareness: Do we need a mechanism to communicate to clients whether messages in the history originated from a TaskStatus object?

   • Should there be metadata or a field distinguishing state messages from regular messages?
   • How can clients determine the source of historical messages?

4. Consistency: Should this be left completely up to the agent implementation, or should the specification provide guidance for consistency across implementations?

Context

Currently, there is ambiguity about:

• Whether agents should include task status messages in the message history
• How clients can distinguish between different types of messages in the history
• Whether this behavior should be standardized or left to agent implementation

Impact

Clarifying this will:

• Help agent developers understand their responsibilities for maintaining message history
• Enable clients to properly interpret and display conversation history
• Improve interoperability between different A2A implementations
• Provide guidance for consistent behavior across agents

Related Issues

• [Bug]: Evaluate removing state_transition_history capability from AgentCapabilities #1228 - State transition history capability
• [Bug]: Agent capabilities has a state transition history but there is no defined way of exposing this history #1158 - Agent capabilities and state transition history
• Is "history[]" really a "state transition history"? #194 - Is "history[]" really a "state transition history"?