Usage: Adaptive AI Usage-Limit Monitoring & Reset Alerts

I built Usage to keep people informed about ChatGPT/Codex capacity without a stream of repetitive alerts. It reads rate-limit state from the Codex app-server and checks less often when capacity is healthy, then more frequently as it becomes scarce.

The skill reports newly crossed thresholds below 20%, below 5%, and at 1% or exhaustion. It treats 1% and exhaustion as one critical episode, tracks multiple windows independently, and combines simultaneous changes. Fresh data must confirm recovery before a reset notification is sent.

The workflow reuses one persistent local task, a same-task heartbeat, one ChatGPT Work task, and one server Scheduled Task. State-based deduplication and recovery thresholds prevent notification spam.

No separate monitoring program, external backend, or new conversation per check is required. Local checks depend on the host being awake and online, and consume some usage.

Source: https://github.com/maddiedreese/usage