Two clocks are running
Think of the short window as your near-term capacity and the weekly window as your overall budget. The short window can recover while the weekly allowance remains heavily used. The reverse can also happen: the week looks healthy, but an intense session fills the shorter window.
This is why one ring or one percentage is not enough. You need the headline limit for a quick glance and the full set of windows when you hover or open a detail view.
Why prompt counts are unreliable
Anthropic explains that usage depends on message length, conversation length, attachments, repository size, and the way Claude Code is allowed to act. Ten small prompts are not automatically equal to one large agent run.
A long conversation also carries more history. When a session becomes bloated, starting a focused session with a clean task description can be easier to reason about and can reduce repeated context.
A simple tracking routine
Before beginning a large change, note the short-window percentage and reset time. During the task, compare the movement with elapsed time. At the end, note the weekly change. You do not need a spreadsheet for every prompt. You need enough observations to recognize your expensive workflows.
If the short window is tight but the week is healthy, wait for the reset or switch to a lighter task. If the weekly number is the constraint, route routine work elsewhere and reserve Claude Code for the tasks where it gives you the most value.