Human work, knowledge and external systems
Agent Rooms model human work as durable contribution requests rather than authorization. Requests support assignment, dependencies, deadlines, structured results and Room artifact references. Completion can resume waiting coordination, but only through the Room's persisted lifecycle.
Work-management delivery
The work-management port projects contribution requests into an external system with idempotent enqueue, leased dispatch, stable external identity, provider lookup recovery, revision updates, bounded exponential backoff, attempt limits and retry operations.
@agentplat/work-management-asana is an optional reference adapter using OAuth
and stable external identity. AgentPlat remains authoritative: an external task
cannot automatically complete or override a Room contribution. Live Asana
validation is optional and is not a condition for Agent Room conformance.
Knowledge bundles
Knowledge bundles are immutable, content-addressed revisions scoped to a Room and authorized agent revision. Standard tool bridges support:
- revision-authorized knowledge reads;
- scoped memory access;
- versioned artifact emission with provenance;
- typed Handoff proposals.
Tool bridges preserve policy and provenance boundaries. A downstream tool must still implement the required idempotency and effect fencing; AgentPlat cannot make a non-idempotent external system safe by declaration.