Agent-to-agent direct messaging (A2X protocol)

[Dhwanil25]

Overview

Currently agents run in parallel without direct communication — they only share context through the orchestrator and final synthesis. A2X (Agent-to-Agent Exchange) would let agents message each other mid-task to share intermediate findings, ask clarifying questions, or hand off work.

How it should work

1. An agent can emit a structured message: { to: 'agent-id', type: 'query' | 'handoff' | 'share', content: '...' }
2. The receiving agent gets the message prepended to its next prompt
3. The canvas shows a live animated arrow between the two nodes when a message is sent

Benefits

• Researcher can share a key finding with the Analyst before finishing its full output
• Coder can ask the Reviewer a clarifying question mid-implementation
• Planner can redirect a worker if the task scope changes

Architecture notes

• multiAgentEngine.ts — executeWorkers() runs workers concurrently via Promise.all; A2X would require a shared message bus (e.g. a Map<agentId, string[]>) polled between streaming chunks
• Canvas already renders messages array — extend with type field to show different arrow styles

Related

• Mentioned in README roadmap