Hermes integration: add session/thread sync path and move toward Claude Code plugin parity

[HamsteRider-m]

Summary

The current nowledge-mem-hermes integration is useful as a knowledge/tool bridge, but it does not yet provide a session/thread persistence path comparable to the Claude Code integration.

Based on the current community README and local behavior inspection:

• Working Memory is injected automatically
• relevant recall/prefetch is available
• native nmem_* tools are exposed
• Hermes user-profile writes can be mirrored to Mem
• but there is no transcript-backed auto-capture / auto thread sync for Hermes sessions
• and nmem t save currently has no --from hermes source

The README already states this explicitly:

The provider teaches Hermes when to use them, but it does not pretend to offer transcript-backed auto-capture or silent background distillation.

That boundary is clean, but it leaves Hermes materially behind the Claude Code integration on the thread-sync side.

Why this matters

If Hermes is meant to be a first-class Mem-integrated agent, users will reasonably expect something closer to the Claude Code plugin experience:

• conversation-period memory tools during the session
• plus a supported path for persisting the session/thread itself into Mem
• plus a clear boundary between provider/plugin responsibilities and host lifecycle responsibilities

Right now Hermes has the first part, but not the second.

Requested direction

Please consider adding a doc-aligned Hermes session/thread sync path, ideally converging toward Claude Code plugin parity where Hermes architecture allows.

Concretely, a good direction would be:

1. Keep the current provider clean

   • working memory injection
   • proactive recall/prefetch
   • native nmem_* tools
   • lightweight mirror of Hermes profile memory if desired

2. Add a host/session-lifecycle sync path separately

   • do not overload the provider with transcript persistence
   • instead add an explicit Hermes-side session export / transcript capture + Mem import path

3. Support a Hermes thread importer in nmem

   • e.g. nmem t save --from hermes or equivalent importer workflow
   • so session persistence is a supported contract, not a one-off script

4. Document the layering clearly

   • provider layer = recall/tools/working memory
   • host lifecycle layer = session/thread sync
   • optional future distill layer = explicit and inspectable, not silent magic

Proposed acceptance target

A practical target would be "rough parity with the Claude Code integration" on the following points:

• session-period memory use works automatically
• Hermes sessions can be saved/searchable as Mem threads through an official path
• docs explain what is automatic vs manual
• implementation follows the community nowledge-mem-hermes integration docs rather than ad-hoc local patching

Not claiming this is a bug

This may be better treated as a feature / integration-completeness issue rather than a bug report, since the current README is honest about the limitation.

But the current limitation is important enough that it probably deserves an explicit roadmap item.