🦋 New version release

[brentrager]

This PR was opened by the Changesets release GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated.

Releases

@smooai/smooth-operator@1.52.1

Patch Changes

• 888cbef: feat(go): client-initiated turn cancellation — Client.Cancel() + MessageTurn.Cancel()

  The wire protocol and all five servers already honor the cancel frame / cancelled
  event, but the Go client had no way to send it. Add the missing client surface, mirroring
  the TypeScript reference (sdk: client-initiated turn cancellation — SmoothAgentClient.cancel() + MessageTurn.cancel() #459):

  • Client.Cancel(CancelParams{RequestID, SessionID?}) sends the cancel frame.
  • MessageTurn.Cancel() is the ergonomic "stop THIS turn" convenience — it cancels using
    the turn's own requestId + originating sessionId. Fire-and-forget, idempotent.
  • A terminal cancelled event settles the matching turn as a user-stop: the event is
    delivered on Events(), the channel closes cleanly, Wait resolves WITHOUT an error,
    and MessageTurn.Cancelled() reports true so a UI tells a user-stop apart from a
    failure. Errors still go the error path.
  • cancel is now first-class in the ActionType set and cancelled in the EventType
    set (with ServerEvent.AsCancelled()), not stringly-typed.
  • Idempotent: a cancel with no active turn, or a cancelled with no matching turn, is a
    harmless no-op.

@smooai/smooth-operator-web-chat-example@0.0.89

Patch Changes

• Updated dependencies [888cbef]
  • @smooai/smooth-operator@1.52.1

@smooai/smooth-operator-server@1.12.1

Patch Changes

• 89c5ca7: fix(ts): cancel discards a HITL-parked confirmation so the parked turn drops cleanly

  Cancelling a turn parked at a write-confirmation (HITL) freed the slot and emitted cancelled
  correctly, but the park in turnRunner.ts awaits a bare deferred from ConfirmationRegistry.register
  (const approved = await verdict) that the turn's cancelSignal abort does NOT itself complete. The
  cancel path already discarded it via a connection-wide confirmations.rejectAll(), but that is a
  broader sweep than the cancel needs.

  FrameDispatcher.cancelActiveTurn now discards precisely the cancelled turn's pending confirmation
  (confirmations.resolve(turn.sessionId, false)) after aborting the controller, so the parked await
  unblocks immediately (resolves denied; the result is dropped because the sink is gagged and the slot
  is already cleared). activeTurn now carries a sessionId, stamped where the turn is created in the
  send_message handler. The disconnect path still rejects every outstanding confirmation separately
  via rejectPendingConfirmations, so nothing dangles there. Mirrors the Rust reference dropping the
  confirmation future on abort and the .NET fix (dotnet: cancel discards a HITL-parked confirmation so the parked turn drops cleanly #460). No behavior change for a non-parked cancel or
  the no-active-turn no-op.

  Adds a parity test (mirroring the .NET CancelUnparkTests) that drives a turn to
  write_confirmation_required, cancels it, and asserts cancelled is emitted, a later
  confirm_tool_action returns NO_PENDING_CONFIRMATION, the slot is freed (a new send_message is
  accepted), and no stray events leak from the abandoned turn.