I'm evaluating Hermes Agent (v0.11) for a multi-agent workflow and ran into a fundamental limitation I can't find documentation for.
**What I'm trying to build:**
A commander agent that dispatches tasks to sub-agents (data analyst, content creator, publisher) and can have back-and-forth discussions with them — not just fire-and-forget task execution. Think of it like a team chat where the commander can ask follow-up questions, request clarification, and iterate on results with sub-agents.
**What OpenClaw apparently supports:**
From reading OpenClaw's session traces, it has a proper subagent mechanism where a parent agent spawns a sub-agent that runs in the same conversation context:
- Session keys like `agent:main:subagent:<uuid>` show the hierarchy
- Sub-agent results come back into the parent's conversation history
- The parent can continue the dialogue, ask clarifying questions, give feedback — it's a real multi-turn conversation within the same thread
**What Hermes Agent appears to support instead:**
- `delegate_task` — synchronous task dispatch, but creates a completely isolated session. When it finishes, the conversation ends. No persistence, no ability to continue.
- `hermes chat -q --profile <name>` — same problem, each invocation is a new session with no memory of previous exchanges
- Cron jobs — for scheduled tasks, not real-time discussion
- The `--continue` / `--resume` flags only work for the CLI user's own session history, not for sub-agent sessions
**The core question:**
Is there any mechanism in Hermes Agent — or any known workaround — to have two agents engage in a persistent, multi-turn conversation where context carries across exchanges? I'm not talking about task queuing or cron-based scheduling. I mean: Agent A asks Agent B something, Agent B answers, Agent A follows up, Agent B refines — the kind of iterative collaboration that requires shared session context.
**What I've tried:**
- `delegate_task` with continuation prompts — doesn't work because each delegate_task is a fresh session
- Passing conversation history in the prompt — token limits make this unsustainable
- Using cron jobs to maintain "always-on" sessions — the scheduler itself has bugs that prevent execution
- The `script -q /dev/null hermes chat` workaround on macOS — works for firing off tasks but not for ongoing dialogue
For context, my use case is a content operations team: a commander agent that needs to discuss data analysis approaches with a data analyst agent, refine queries, iterate on results — all in a persistent conversational thread.
Would appreciate any pointers — even "this is planned for v0.12" or "here's how to work around it." Thanks.