Ramón Medrano Llamas opened Gemini CLI PR #28063 with two public repairs in one patch: failed workspace publishing and a macOS hang during waves of parallel tool calls.
The public trail
The merged commit be7ba2c22 changes packages/core/src/scheduler/scheduler.ts so waiting tool calls yield through setTimeout(resolve, 10) instead of queueMicrotask. The PR says the old loop could starve timers, I/O, child-process completion, and rendering callbacks.
The patch also adds --ignore-scripts to the release publishing action for core, CLI, and A2A packages. That keeps workspace package publishing from re-running package lifecycle scripts in the wrong package context.
Why this contribution matters
This is not a launch feature. It is the kind of contribution that makes an agent CLI usable when the model asks for several tool calls at once. The user-facing effect is simple: a tool wave should keep moving, the release path should keep shipping, and the terminal should not sit hot at 100 percent CPU.
The integration test change is part of the story. PR #28063 updates the parallel-tool test fixture, waits longer for approval and final output, and records PTY output on failure. The repair came with a way to see the failure next time.
The conversation to open
The constructive follow-up for Medrano Llamas and Gemini CLI maintainers is whether scheduler starvation should become a visible diagnostic: when a tool wave waits on approval, child processes, or I/O, can the CLI show what it is yielding for before users assume the agent froze?