Skip to main content

WorkBoard delegation and child conversations

Read this if: you need the handoff model between interactive work capture and background progress.

Skip this if: you only need the high-level WorkBoard boundary; use Work board and delegated execution.

Go deeper: WorkBoard durable work state, Approvals, Conversations and Turns.

This page describes how WorkBoard captures long-lived work, chooses when to continue in the same conversation versus a child conversation, and routes progress back to the operator. It does not define a second top-level execution model.

Delegation flow

Standard intake flow

  1. Classify the request as inline work, action work, or initiative work.
  2. Write minimal acceptance criteria, budgets, and authoritative current-truth state.
  3. Seed the WorkItem with initial artifacts, risks, or reminders.
  4. Decide whether follow-up work stays in the same conversation or moves into a child conversation.

Child conversation model

A child conversation is a delegated context that shares the parent agent boundary but has its own conversation state and transcript.

Starting semantics:

  • same agent_id
  • same workspace and policy envelope
  • shared memory scope unless a narrower memory policy is explicitly chosen
  • different conversation_id, so it does not serialize behind the parent conversation
  • explicit parent/child linkage so operators can inspect why isolation exists

The WorkBoard is updated from durable conversation outcomes plus explicitly written WorkBoard records, not from transcript narrative alone.

Fan-out and synthesis

Figure out what to do can still be expressed as explicit fan-out tasks followed by a synthesis task that proposes next steps.

To keep planning inspectable and resilient under interruption:

  • fan-out tasks produce WorkArtifacts such as hypotheses, candidate plans, and verification reports
  • synthesis writes a DecisionRecord, updates the WorkItem task graph and state KV, and may create WorkSignals
  • when inputs conflict, the planner inserts an explicit read-only child conversation before side effects proceed

Status and notification routing

The interactive agent loop should answer progress questions using WorkBoard state:

  • work.status(last_active_work_id) or work.list_active() for current status
  • blockers, approvals, recent DecisionRecords, and next-step summaries from durable work state

Completion or blocked-state notifications should route to the originating conversation, with child-conversation linkage preserved for drill-down.

Backlog and WIP control

Multiple long-lived WorkItems are expected. The WorkBoard prevents overload and thrash through:

  • WIP limits on active work
  • overlap detection on target resources
  • explicit dependency links instead of implicit work merging
  • budgeted drill-down retention for artifacts, decisions, and signals

Safety integration

Delegation does not bypass Tyrum's enforcement model:

  • side effects still flow through the same turn-processing and approval rules
  • approvals still pause progress safely
  • verification and evidence remain required where feasible
  • drift checks pause and escalate instead of allowing helpful but unsafe improvisation