feat: Implement A2A (Agent-to-Agent) Protocol for Syner OS

[rbadillap]

Summary

Implement the Agent-to-Agent (A2A) Protocol to enable Syner OS to communicate with other AI agents (Claude, Codex, custom agents) as peers.

Background

What is A2A?

A2A is an open protocol (Google/Linux Foundation, 150+ organizations) that enables communication between AI agents. It complements MCP:

Protocol	Direction	Purpose
MCP	Vertical (agent ↔ tools)	Agent accesses external tools/data
A2A	Horizontal (agent ↔ agent)	Agents collaborate as peers

Why A2A for Syner?

1. Industry standard - Backed by Google, Anthropic, OpenAI, and Linux Foundation
2. GitHub Agent HQ - The path to becoming a GitHub coding agent starts here
3. Interoperability - Any A2A-compliant agent can communicate with Syner
4. Syner OS vision - Positions Syner as a platform that orchestrates and creates agents

Technical Specification

Required Endpoints

```
https://api.syner.dev/
├── /.well-known/agent-card.json # Discovery (who is Syner)
└── /a2a/
├── POST /messages # Receive messages
├── GET /tasks/:id # Task status
├── GET /tasks # List tasks
└── POST /tasks/:id/cancel # Cancel task
```

Agent Card Structure

```json
{
"id": "syner",
"name": "Syner OS",
"version": "1.0.0",
"description": "Agentic Operating System - orchestrates and creates AI agents",
"provider": {
"name": "Syner",
"url": "https://syner.dev"
},
"serviceEndpoint": "https://api.syner.dev/a2a",
"capabilities": {
"streaming": false,
"pushNotifications": false
},
"skills": [],
"interfaces": [{ "type": "http", "version": "1.0" }]
}
```

Messages vs Tasks

Concept	What it is	Duration
Message	Single turn of communication	Immediate
Task	Stateful unit of work with lifecycle	Can be long-running

Task states: `submitted` → `working` → `completed`/`failed`/`canceled`

Use Cases

1. Agent-to-Agent Communication (Basic)

Other agents (Claude, Codex, custom) can send tasks to Syner:

```
Claude ──A2A──► Syner: "Help me understand this codebase"
Syner ──A2A──► Claude: { task: "completed", artifacts: [...] }
```

2. Syner OS as Meta-Agent Platform

Syner OS can create and configure other agents (IA creating IA):

Capability	Description
Generate Agent Cards	User describes intent → Syner generates Agent Card JSON
Create AGENTS.md	Syner writes instructions for coding agents
Orchestrate agents	Syner decides which agent handles each subtask
Generate skills	Syner creates tools/skills dynamically for agents
Spawn specialized agents	Syner creates agent instances with specific configurations

Example flow:
```
User: "I need an agent that reviews security PRs"

Syner OS:

1. Generates Agent Card with security skills
2. Generates AGENTS.md with security rules
3. Registers agent in system/registry
4. New agent can now receive tasks via A2A
   ```

3. GitHub Integration (Future)

Once A2A is implemented, Syner can:

• Respond to `@syner` mentions in GitHub via webhooks
• Participate in Agent HQ ecosystem
• Collaborate with Claude/Codex on GitHub issues and PRs

Implementation Plan

Phase 1: A2A Server (this issue)

1. Agent Card endpoint - `/.well-known/agent-card.json`
2. Messages endpoint - `POST /a2a/messages`
3. Tasks endpoints - CRUD for task management
4. Connect to Syner agent - Process messages with SDK

Proposed location: `apps/dev/app/api/a2a/`

Phase 2: Skills Definition

Define Syner's A2A skills in the Agent Card:

• Code assistance
• Agent generation (meta)
• Codebase analysis

Phase 3: GitHub App Extension

Add webhooks to `extensions/github` to:

• Receive `@syner` mentions
• Bridge GitHub events to A2A tasks

References

• A2A Protocol Specification
• A2A GitHub Repository
• MCP + A2A Webinar (Anthropic)
• A2A-MCP Bridge
• Linux Foundation A2A Announcement

Labels

• enhancement
• agent
• protocol