[Bug] A2A Communication Fails When Context Message Volume Exceeds Threshold

[Clawiee]

Bug Description

A to A (Agent to Agent) communication in Clawith is always based on the same conversation context. When two agents communicate multiple times, the context accumulates a large number of dialogue messages. Once the message volume reaches a certain threshold, A to A communication crashes and becomes ineffective.

Steps to Reproduce

1. Set up two or more agents that need to communicate with each other
2. Trigger multiple rounds of A to A communication between the agents
3. Observe that as communication continues, the message count in the shared context grows
4. When the accumulated messages exceed a certain threshold (likely related to LLM context window limits), the A to A communication fails

Expected vs Actual Behavior

Expected: A to A communication should either:

• Be isolated per conversation with proper context management
• Have a mechanism to summarize or prune old messages to prevent context overflow
• Support cross-context communication when needed

Actual: A to A communication is limited to the same context window and conversation. All messages accumulate in one shared context, eventually causing the system to crash or become unresponsive when the message volume exceeds limits.

Additional Context

This appears to be a fundamental architectural issue with how agent-to-agent communication is implemented. The current design assumes all A to A messages exist within a single conversation context, which is not scalable for production use cases requiring frequent agent collaboration.

Suggested Fix Direction

• Implement conversation isolation for A to A communication
• Add context summarization/pruning for long-running agent dialogues
• Consider message threading or hierarchical context management for agent communications
• Set appropriate context limits with graceful degradation when approaching limits