Turn handoffs can repeat an assistant message id, and the desktop host gets host_worded_asks
turn_handoff now lets a tool-only assistant message reuse the previous message's id, with a shared_message_id capability flag and a new host_worded_asks flag
Under the hoodTier: how much it should matter to you
2Useful: my rating, 1 to 5
3Signal: worth watching, 1 to 5
Cloud SessionsArea: what it touches
Internal ChangesKind: in v2.1.299,
Internal ChangesSection of the release
What
The control protocol is the internal message format that other programs, such as the desktop app or a cloud worker, use to drive Claude Code. A turn_handoff passes an unfinished turn to a worker. Several schema changes, all marked internal, affect it:
An assistant message that holds only tool_use blocks, meaning requests by Claude to run a tool, may now repeat the id of the previous assistant message. This is allowed only when nothing but tool_result messages sit between the two, and the two are sent to the model as a single message. Before, every assistant message id had to be different.
shared_message_id is a new capability flag. With it, a worker says it accepts a turn_handoff in which assistant messages repeat an id.
host_worded_asks is a new field in the initialize response, the first reply Claude Code sends when a host connects. It applies only to the desktop host.
Why
These are internal protocol changes and need no action from you. The sweep agent read host_worded_asks as groundwork for the desktop app to show permission requests in its own wording.
