An agent handoff prompt records the objective, completed work, current state, important decisions, constraints, unresolved issues, evidence locations, and next useful steps. It transfers task knowledge while avoiding a full copy of every prior message or tool output.
A good handoff distinguishes durable facts from assumptions and does not silently expand authority. The receiving agent should verify current files and external state because either may have changed since the handoff was written.
ELI5
An agent handoff prompt is a short briefing for the next AI agent or session. It explains what the task is, what has already been done, which choices were made, what limits apply, and what should happen next.
For example, a handoff might say that a migration is written, two tests pass, one test still fails, and a protected file must not be changed. The next model can begin from that useful summary while checking the current project for itself.
