| name | handoff |
|---|---|
| description | Write a handoff document so a fresh agent can pick up the current work. Use when the user asks for a handoff, to hand off work, or to compact the session into instructions for another agent. |
| disable-model-invocation | true |
Compact the current session into a handoff document that a fresh agent (with none of this context) can pick up and continue from. The agent may be on an entirely different harness. Do not reference task IDs, sub-agents, or other artifacts of this session.
- Write the document from current context only. Once handoff is requested, make no further tool calls except the single write of the document itself. Do not re-read files, re-run commands, or gather new information.
- Write the document to the temporary directory, e.g.
/tmp/handoff-<topic>-<YYYYMMDD-HHMM>.md. - After writing, report to the user a concise summary of the handoff document and its full path. Nothing else.
Write for an agent that knows the repo conventions (AGENTS.md) but nothing about this session. Include:
- Goal: what the user ultimately wants, in their words where possible.
- State: what has been done and verified so far, including files changed and commits made.
- Remaining work: concrete next steps, in order, with enough detail to act without rediscovery.
- Key knowledge: load-bearing facts learned this session — file paths, function names, data findings, failed approaches and why they failed, user decisions and preferences stated during the session.
- Verification: how to check the work (commands, queries, expected output).
Omit sections that are empty. Prefer specific paths, names, and numbers over prose summaries; the reader cannot see this conversation.