What they did
The poster kept losing Claude usage at the weekly reset when they were away or had slow days, while their backlog held plenty of unattended work: flaky tests, stale TODOs, doc updates, research and bug hunting. They built Redline, a free open-source scheduler, mostly with Claude and some Codex. You list jobs; it watches your usage and starts one when spare capacity looks likely, keeping a reserve for your own sessions. Each job runs in its own git worktree. It ships with 16 maintenance job templates, works with Claude, Codex and Pi, and runs as a macOS menu bar app, CLI and dashboard. On plans without a five-hour limit it paces against the weekly reset.
How it went
The poster says it lets them run more than 50 extra jobs a week on usage that would have expired. Small, bounded jobs worked best: "find and stabilise one flaky test" usually lands in one shot, while "clean up the tests" needs several rounds. A model-specific reset once looked like a weekly reset to the scheduler until a fix. The thread had no replies with other results.
Worth knowing
It is built to fail closed: if it cannot read usage or is unsure, it does nothing. Linux and Windows support was still in testing at the time of the post. The post is the author's own launch.