Aglet

Triage an Agent Handoff Evaluation

A handoff can fail even when the agent recognizes its limit. Triage the transfer contract before reviewing tone: identify why control changed, what information moved, and whether the recipient can act without reconstructing the task. Include deliberate non-handoff cases with no receiving owner.

Establish what is happening

  1. Define the transfer trigger

    Record conditions for escalation, refusal, clarification, or continued execution. State whether the trigger depends on uncertainty, missing authority, task scope, user preference, or tool failure. Separate a useful handoff from a generic apology.

  2. Map recipient needs

    List the request, completed work, evidence, unresolved question, constraints, risk, and proposed next action the recipient receives. Include who owns the next decision and what the user is told about the transfer.

  3. Bound the score

    Specify whether the evaluation measures trigger choice, context completeness, ownership clarity, user expectation, or all four. Mark missing context and unavailable recipient behavior as evidence gaps rather than automatic agent failures.

What to carry forward

The triage output is a handoff contract with triggers, recipient context, ownership, user notice, and score boundaries. Stop when a reviewer can tell why control changed and what happens next. Keep the result provisional when the recipient path is not observable in this test.

Technical background: Google DeepMind evaluation research.

Keep the decision with the work.

Use a Work Item in Aglet to record the problem, the evidence you have, and the next decision. Add an owner and priority, then keep updates in the discussion so the next person can follow the reasoning.

Create an account See the product workflow