hermes - 💡(How to fix) Fix Multi-agent system: inter-agent communication and persistent agent identities

Problem

Hermes currently supports spawning additional Hermes instances via tmux/profiles, but lacks:

1. Inter-agent communication — agents cannot send messages or delegate tasks to each other without user mediation
2. Persistent agent identities — each agent has no stable identity, memory, or persona that survives across sessions
3. Coordination protocols — no way to elect a leader, distribute tasks, or share context between agents

Use Case

A user wants a "personal agent matrix":

• Agent A: manages WeChat public account (content, publishing, data)
• Agent B: manages job search (resume, applications, interview prep)
• Agent C: handles family/life admin (schedule, reminders)

All three run as independent Hermes processes. Users want:

• Agents to communicate with each other ("Agent B, what's Agent A's latest post status?")
• Each agent to have its own memory/persona that persists
• A shared "inbox" or bus for task distribution

Proposed Solution

First-class multi-agent support including:

• Agent registry — name, persona, capabilities, profile path
• Inter-agent messaging — agent.send(to="career-ops", message="check today's jobs")
• Shared memory bus — optional shared knowledge layer (e.g. "shared inbox" for tasks)
• Task delegation — one agent can spawn a task for another with result callback
• Identity persistence — each agent's memory/personality survives restarts

Alternatives Considered

• Using Feishu/Discord group chat as message bus — works but requires user as router
• Using tmux + profiles — current workaround, agents are siloed
• delegate_task — only within same process, no persistence

Priority

Mid — enables more autonomous workflows; workaround exists via manual coordination.