Assistant text between tool calls renders inconsistently in long wakeup-driven sessions

During a long autonomous session driven by ScheduleWakeup (an overnight training-monitoring loop, ~8 hours, dozens of auto-fired turns), the assistant wrote substantive multi-paragraph findings between tool calls and ended most turns with a ScheduleWakeup call. The user reports seeing short one-line status texts between tool calls but none of the substantive multi-paragraph reports, and experienced the session as the assistant making dozens of changes with zero explanation.

Expected: assistant text interleaved with tool calls either renders consistently, or the UI clearly indicates that text was elided, so users can trust that what they see is what was said.

Observed: inconsistent visibility. Short interstitial lines rendered while longer report blocks in the same positional pattern apparently did not, which reads as the assistant willfully withholding information and destroyed the user's trust in an otherwise-good session.

The system prompt does warn the model that mid-turn text may not be shown, but the asymmetry (short lines visible, long reports not) plus turns that structurally must end in a scheduling tool call makes the failure easy to hit and hard for users to distinguish from bad faith. Sessions that must end turns with ScheduleWakeup have no way to put a report after the final tool call.

Environment: Claude Code CLI on macOS (darwin 25.4.0), model claude-fable-5, session using ScheduleWakeup + Monitor + background Bash tasks heavily.