Author here. This started as a personal annoyance: when my Claude Max window closes mid-task, every subagent that was running dies, and after the reset the orchestrator re-dispatches the same tasks from scratch. I went through 6 of my own sessions: 449 subagents cut off by limits, 8 resumed by id (all 8 before the reset, so they died again), 438 redone.

The fix turned out to be mostly reading files Claude Code already writes. The exact reset time is in the transcript as `quotaLimits.resetsAt` on the `error: "rate_limit"` entry (the UI only shows "resets 8pm"). The status line gets `rate_limits.five_hour` / `seven_day` with `used_percentage` and `resets_at` as JSON on stdin. Each subagent's transcript under `<session>/subagents/` has its task, the files it touched, and its last steps; a stopped subagent resumes with `SendMessage <id>` with full context.

So the tool is four hooks and a status-line tap: save the percentages, ask the agent to write a handoff file at 90%, record the exact reset time on a hit, and for headless `claude -p` runs sleep until the reset and `--resume` the same session. No proxy, no account, no network. The subagent part (which of them were actually running at the cut, minus the ones Claude Code already reported since 2.1.270, and the instruction to resume them by id after the reset, for Claude Code / Codex / ZCode) is the paid part, $29 one-time with source.

Things it can't do, so nobody buys it for the wrong reason: it can't raise limits, doesn't rotate accounts, and Codex is verified on macOS only so far.