An agent handoff packages the information needed for another actor to continue work without rebuilding the task from scratch. It can include the goal, current state, completed actions, evidence, unresolved issues, constraints, permissions, artifacts, and recommended next step.
Reliable handoffs separate facts from assumptions, preserve provenance, minimize private data, and make responsibility explicit. The receiving actor should verify current state and authority because context may be incomplete or stale by the time work resumes.
ELI5
An agent handoff passes work from one person, agent, or session to another. It includes enough information for the next participant to continue safely and understand what already happened.
For example, a research agent can hand a verified source list and open questions to a writer. The writer checks the evidence instead of repeating the complete search or trusting an unexplained summary.
What should an agent handoff include?
It should include the goal, current state, completed actions, evidence, decisions, artifacts, constraints, permissions, failures, unresolved issues, and next step.
Why should the receiver verify a handoff?
The context may be incomplete, stale, mistaken, or created under different permissions, so the receiver must confirm state and authority before acting.
