Designing a routing graph

When designing how your sub-agents communicate, the shape of the graph significantly impacts the reliability and capability of the overall agent.

The golden rules of routing

1. The entry agent must route: If the graph's entry sub-agent ends its very first turn with a plain answer instead of routing anywhere, the runner will nudge it to actually route (or ask one clarifying question) rather than accepting a fabricated direct answer.
2. Ensure reachability: Design the handoff graph so every sub-agent is reachable from the default one. An unreachable specialist can never be routed to. Reachability may use either edge type, but prefer delegate edges unless ownership transfer is required.
3. Avoid the transfer-mesh: Do not wire every sub-agent to be able to transfer to every other sub-agent. This leads to thrashing and confusion.

Typical shapes

A well-designed routing graph usually looks like a tree with a clear hierarchy:

• Triage → Specialist: A triage/router default sub-agent often TRANSFERS to one owner specialist based on the user's initial request.
• Specialist → Helper: Specialists DELEGATE to helpers (like a researcher or a calculator) to gather information, before returning a final answer.

Dealing with cross-agent references

A sub-agent that is really a different, externally-linked agent (a cross-agent reference) can only ever be a DELEGATION target, never a transfer target.