Fleet Mapping

How the mock patterns in this repo map to LangSmith Fleet and the Fleet Engineering stack.

This repo	LangSmith Fleet	Fleet Engineering concern
TokenStore.get(user_id)	Per-user OAuth + channel user mapping	Identity & credentials
load_agent_credentials(...)	Claw provisioning at agent creation	Identity & credentials
AssistantMemory	Per-user threads	Identity & credentials
ClawMemory + sender	Shared resource; sender is context only	Identity & credentials
AssistantInbox	Private per-user inbox for sensitive work	Inbox / HITL
ClawInbox	Editor-only review before sensitive actions ship	Inbox / HITL
AuditEvent.to_dict()	Cross-agent audit trail	Accountability
notion_search(creds, query)	Tool call scoped to principal	Permissions

Assistant flow (pseudocode)

slack_user = event["user"]
user_id = fleet.map_channel_user(slack_user)  # U_ALICE → alice@co
creds = token_store.get(user_id)                # OAuth per human
pages = notion_search(creds, query)             # only that human's scope
audit.emit(principal=user_id, model="assistant", action="notion.search", ...)

Claw flow (pseudocode)

creds = agent.credentials                 # fixed at provisioning
pages = notion_search(creds, query)     # same scope for every asker
inbox.submit_if_sensitive(action, sender=event["user"])
audit.emit(principal=creds.id, model="claw", triggered_by=event["user"], ...)

Related fleet docs

• Five Concerns — identity is concern #3
• Shared Inbox HITL — Claw editor review
• Failure Modes — mixed claw/assistant semantics
• Cross-Agent Audit — structured accountability