Priority: Medium Complexity: High
Add support for Google's Agent-to-Agent (A2A) protocol to enable:
- Spawning remote agents (Vertex AI, Azure AI, LangGraph, etc.)
- Exposing Synapse as an A2A-compatible agent
- Agent discovery via Agent Cards
- Task status streaming/push notifications
Current state: spawn_agent only spawns local Synapse processes.
With A2A:
{"tool": "spawn_agent", "params": {
"task": "Research auth patterns",
"agent": "vertex://research-agent"
}}Resources:
Priority: Low Complexity: Medium
Replace lib/llm.sh with LiteLLM proxy for:
- 100+ providers out-of-box
- Built-in tool-use translation
- Cost tracking, rate limiting
- Fallback/retry logic
Options:
- LiteLLM as proxy server (keep bash)
- Rewrite Synapse in Python with LiteLLM SDK