Hermes A2a Setup

hermes-a2a Plugin Setup

Source repo: hermes-a2a (iamagenius00 on GitHub)

Hermes plugin for peer-to-peer agent communication via Google's A2A protocol. Zero external dependencies, pure Python stdlib.

Prerequisites

• Hermes Agent v2026.4.23+ (plugin API with register_command)
• CLI mode only (gateway does NOT load plugins)

Install Steps

1. Clone repo and run install.sh — copies 7 files to plugins directory
2. CRITICAL: Run hermes plugins enable a2a — plugins are opt-in
3. Set A2A_ENABLED=true and A2A_PORT=8081 in environment
4. Restart CLI session (plugin loads at session start)

Pitfall: Plugin Must Be Explicitly Enabled

The plugin system requires explicit enablement. After install, run:

hermes plugins enable a2a

Without this, the plugin is discovered but NOT loaded. User plugins are opt-in — only bundled backends auto-load.

Pitfall: Gateway Does NOT Load Plugins

Only CLI mode supports A2A. Gateway (Telegram/Discord/Feishu) won't start the A2A server. Plugin hooks only fire in CLI conversation loop.

Tools Exposed

Configuring Remote Agents

Add entries under a2a.agents in config yaml:

a2a:
  agents:
    - name: "friend"
      url: "https://friend-a2a-endpoint.example.com"
      description: "My friend's agent"
      auth_token: "their-bearer-token"

Security

• Bearer token auth (localhost-only without token)
• 20 req/min rate limit per IP
• 9 prompt injection patterns filtered
• Outbound secrets redacted
• Audit log at a2a_audit.jsonl

Architecture

Remote Agent sends A2A JSON-RPC to localhost:8081 → task_queue.enqueue() → webhook trigger fires agent turn → pre_llm_call hook injects message into current session → agent replies with full context → post_llm_call captures response → response returned synchronously (120s timeout) → exchange persisted to a2a_conversations/

Files

Routing Keywords (in skill-router.md)

A2A agent 对话、agent 间通信、给别的 agent 发消息 → A2A 插件