feat: add Hermes Agent adapter

[Jordanm9936-FS]

Summary

Adds a built-in Hermes Agent adapter so OpenAgents can run Hermes profiles as workspace-connected agents.

What this adds:

• src/openagents/adapters/hermes.py — new adapter built on BaseAdapter
• src/openagents/registry/hermes.yaml — builtin runtime definition for Hermes
• README update to list Hermes Agent as supported

How it works:

• Uses Hermes CLI in non-interactive single-query mode (hermes chat -q ... -Q)
• Persists one Hermes session ID per workspace channel for continuity
• Injects workspace context, recent message history, and current agent roster into the prompt
• Infers a Hermes profile automatically from the OpenAgents agent name when a matching Hermes profile exists, otherwise falls back to default
• Keeps profile isolation intact by using Hermes subprocesses instead of importing Hermes runtime into the OpenAgents process

Why this approach

Hermes is closest to the OpenClaw/CLI family operationally, but unlike OpenClaw it already has strong profile isolation (HERMES_HOME, profile-specific memory/config/auth). Using subprocesses keeps that isolation clean and avoids cross-profile state bleed inside the OpenAgents daemon.

Test plan

Validated locally on macOS:

• openagents runtimes shows hermes as installed
• openagents create hermes --name builder ... succeeds
• Hermes agent joins the same hosted trial workspace as Claude Code + Codex CLI
• Hermes replies back into the workspace chat through the adapter

Notes

This is intentionally the simplest useful integration first:

• no MCP bridge yet
• no shared OpenAgents file/browser tool mapping yet
• no streaming token/tool events from Hermes yet

It is already enough to make Hermes profiles participate in OpenAgents workspaces and collaborate over shared threads.

Related: #328

[vercel]

@Jordanm9936-FS is attempting to deploy a commit to the Raphael's projects Team on Vercel.

A member of the Team first needs to authorize it.