Summary
The Honcho integration spec also identifies several structural backports Hermes should still adopt from openclaw-honcho: peerPerspective on context retrieval, workspace-level agentPeerMap, and a multi-agent observer hierarchy. Hermes already supports strong single-agent Honcho behavior, but these cross-agent/workspace semantics still appear missing.
Evidence
Docs
docs/honcho-integration-spec.md:92-98
peerPerspective on context()
- workspace
agentPeerMap
- multi-agent observer hierarchy
docs/honcho-integration-spec.md:349
- these are called out as already done in openclaw-honcho and not yet ported into Hermes
Current implementation
- repo search did not find
peerPerspective usage in the Python Honcho integration
- repo search did not find
agentPeerMap in the Python implementation
- current Honcho manager/session/plugin code is centered on a single user peer + single assistant peer session model
plugins/memory/honcho/session.py
plugins/memory/honcho/__init__.py
Problem
Hermes' Honcho integration is strong for single-agent memory, but it still lacks the workspace/session semantics needed for richer multi-agent orchestration and cleaner cross-peer context retrieval.
Missing these features makes it harder to:
- preserve parent/child observer relationships for delegated agents
- reason about which agent peer belongs to which workspace/session lineage
- request context from the correct peer perspective during recall
Scope
- Add
peerPerspective-aware context retrieval where supported by the Honcho SDK/API
- Introduce workspace metadata mapping from agent/runtime identity to Honcho peer ID (
agentPeerMap or equivalent)
- Design and implement a parent/child observer hierarchy for delegated or spawned agents
- Ensure the implementation composes with existing
delegate_task / child-agent execution model
- Add tests for peer mapping, perspective-aware context, and child-agent lineage behavior
Acceptance criteria
- Hermes can resolve the correct agent peer for a workspace/runtime identity
- Honcho context retrieval can be requested from an explicit peer perspective when available
- Child/subagent runs can be represented in Honcho without collapsing into a single flat assistant identity
- Tests cover workspace peer mapping and multi-agent lineage behavior
Related
Summary
The Honcho integration spec also identifies several structural backports Hermes should still adopt from openclaw-honcho:
peerPerspectiveon context retrieval, workspace-levelagentPeerMap, and a multi-agent observer hierarchy. Hermes already supports strong single-agent Honcho behavior, but these cross-agent/workspace semantics still appear missing.Evidence
Docs
docs/honcho-integration-spec.md:92-98peerPerspectiveoncontext()agentPeerMapdocs/honcho-integration-spec.md:349Current implementation
peerPerspectiveusage in the Python Honcho integrationagentPeerMapin the Python implementationplugins/memory/honcho/session.pyplugins/memory/honcho/__init__.pyProblem
Hermes' Honcho integration is strong for single-agent memory, but it still lacks the workspace/session semantics needed for richer multi-agent orchestration and cleaner cross-peer context retrieval.
Missing these features makes it harder to:
Scope
peerPerspective-aware context retrieval where supported by the Honcho SDK/APIagentPeerMapor equivalent)delegate_task/ child-agent execution modelAcceptance criteria
Related
docs/honcho-integration-spec.md