Claude Code's 7-Day Weekly Limit

The limit that ruins your week if you don't watch it. Here's how the rolling cap works and how to stay informed before it cuts you off.

The painful version: Hit the weekly cap on a Tuesday and you may be waiting until the following Tuesday to get capacity back. Unlike the 5-hour session limit, there's no quick drain — old requests fall off very slowly when the window is 7 days wide.

How the 7-day rolling window works

The weekly limit is not a fixed calendar week ("resets every Monday"). It's a 7-day rolling window: each request you make stays in the window for 7 days, then falls off.

If you used 80% of your weekly cap on Monday, that 80% doesn't clear until the following Monday — specifically, until exactly 7 days after each request was made, which trickles in slowly over the day.

Because the window is so long, the "drain" is much slower than the 5-hour session window. You can't wait it out in an afternoon.

Checking your weekly usage

Inside Claude Code: /usage shows the weekly % and approximate reset time.

From the terminal (requires Headroom):

jq '{weeklyUsagePct, weeklyResetSec}' ~/.claude/headroom-usage.json

Example output:

{
  "weeklyUsagePct": 83.7,
  "weeklyResetSec": 201600
}

weeklyResetSec is 201600 seconds — that's 2.3 days before the next significant drain. The weekly field is the one to watch over a multi-day horizon.

Convert weeklyResetSec to days

jq '.weeklyResetSec / 86400 | . * 10 | round / 10 | tostring + " days"' ~/.claude/headroom-usage.json

What Headroom shows

CC 23%·87%

The second number is the weekly %. When it turns red at 90%, you have roughly 10% of your weekly cap left. The dropdown shows the exact value and the time until the next significant drain.

This is why the weekly % is the most important number to keep ambient. The session % recovers in hours; the weekly % recovers in days.

The pace forecast

Headroom's dropdown also shows a pace forecast: "at current pace, session exhausted in ~Xh" or "weekly exhausted in ~Xd." This gives you advance warning before you hit the wall, not after.

Avoiding the weekly cap

Watch the weekly % from day 1. At 70% (amber), plan how many heavy sessions you have left in the week. At 85%, you're in conservation mode.
Front-load smaller tasks on high-% days. Context-heavy operations burn the weekly cap fast. Save those for the start of your weekly window.
Use /compact generously. Compacting context mid-session reduces tokens per request, stretching your weekly budget.
Plan across the rolling window, not the calendar week. If you burned 70% last Wednesday, your window resets next Wednesday — not "next Monday."

Weekly vs session: which is binding?

Limit	Window	How to recover	Which to watch
Session	5 hours	Wait a few minutes to a few hours	During active work sessions
Weekly	7 days	Wait days — the slow drain	Over multi-day project planning

The session limit hits you suddenly during a work burst. The weekly limit is a slow squeeze — you can see it coming if you're watching, but it's catastrophic if you're not. Both meters need to be visible.

Monitor the weekly cap always-on

brew install --cask patwalls/tap/headroom

Or download directly. Free, MIT, ~267 KB. The weekly % is the second number in the menu bar — always visible, color-coded, with a countdown in the dropdown.

→ The 5-hour session limit
→ When does the limit reset?
→ Full rate limits guide
→ Set up threshold alerts