A weekly Claude Code bar at 86% with 73% of the week gone was heading for 117%

Why this label: A single post by the person who built the status line. The arithmetic checks out, but we have not run the script and the post gives no code. We read the 86% as used (not remaining) because 86 / 73 matches the 117% projection.

Tried

Judge how much room agents had left by reading the two percentages in the Claude Code status line (7-day and 5-hour windows).

Broke

The bare percentage did not show how long the budget would last. With 86% used and 73% of the week elapsed, the author projects 117% by the reset, a hard stop roughly 19 hours early (author's figures).

Cause

• confirmed A percentage without the pace hides the problem: usage was running ahead of the calendar. The arithmetic in the post is checkable (86 / 73 is about 1.18).

“Confirmed” means the source states it as established; “suspected” means the source treats it as a theory or it is unsettled. We did not verify either ourselves.

Fix

Show the pace in the status line. The author says Claude Code hands the status-line script the usage numbers with every reply, so nothing has to be polled. Their bar prints 7d:29%>61% 5h:45%: used so far, then (after the >) where the week lands at the current pace, then the 5-hour window. The pace leans on the last few hours, so slowing down shows up quickly. The post describes the script but we have not seen or run it.

Prevention

Compare percent used with percent of the week elapsed at least once a day, and slow or pause agents when the projection passes 100%, before the 5-hour limit becomes the problem.

Source

• Post on X by @daveremy (2026-10-27 UTC)

← All stories