Excessive weekly usage limit consumption (71% in 2 prompts)

Resolved 💬 3 comments Opened Nov 22, 2025 by aronstoppacher Closed Nov 26, 2025

Preflight Checklist

  • [x] I have searched existing issues and this hasn't been reported yet
  • [x] This is a single bug report (please file separate reports for different bugs)
  • [x] I am using the latest version of Claude Code

What's Wrong?

My weekly usage limit jumped from 13% to 84% after just 2 prompts discussing the usage limit itself. The conversation was minimal (no code generation, no file operations, just talking about quota usage), yet it consumed 71% of my weekly limit in ~30 minutes.

What Should Happen?

What Actually Happened

  1. I noticed my usage limit has been filling up faster recently
  2. I asked Claude Code about why my limit went from 13% to 84%
  3. We only discussed the usage limit problem itself - no coding, no file reads, minimal conversation
  4. After this short discussion, the weekly limit was completely exhausted

This is ironic: Talking about high usage caused the highest usage spike yet.

## Steps to Reproduce

  1. Start new session on 2025-11-22 (daily limit just reset)
  2. Send 2 short prompts asking about usage limit consumption
  3. Check usage: Weekly limit increased by 71% in ~30 minutes
  4. No code generation, no file operations, no agents launched intentionally

## Expected Behavior

  • A brief conversation about usage limits should consume minimal tokens
  • Maybe 2-5% of daily limit for 2 simple prompts
  • Weekly limit should reflect accumulated daily usage

## Actual Behavior

  • Weekly limit: 13% → 84% (71% increase)
  • Daily limit: Still has capacity (not full) ⚠ This is the red flag
  • Conversation content: Just discussing the usage limit issue
  • Session duration: ~30 minutes
  • Number of prompts: 2

Mathematical impossibility: How can weekly limit be 84% full when daily limit is not exhausted?

## Environment

  • Platform: Linux (Nobara FC42, Kernel 6.17.4)
  • Claude Code Version: [Latest as of 2025-11-22]
  • Model: Claude Sonnet 4.5
  • Working Directory: /run/media/sxc/Volume/Projekte/Server/WerkstattDashboard/
  • Project Context: ~15KB CLAUDE.md file loaded
  • Subscription: Pro

## Conversation Context
The prompts were:

  1. Asking why usage limit jumped from 13% to 84% in one session
  2. Expressing frustration about weekly limit being full after 30 minutes of work

No coding tasks, no file operations, no complex queries - just a discussion about usage.

Error Messages/Logs

Steps to Reproduce

  1. Start new session on 2025-11-22 (daily and week limit just reset)
  2. Send 2 short prompts asking about usage limit consumption
  3. Check usage: Weekly limit increased by 71% in ~30 minutes
  4. No code generation, no file operations, no agents launched intentionally

Claude Model

Sonnet (default)

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

Claude Code v2.0.50

Platform

Anthropic API

Operating System

Other Linux

Terminal/Shell

Other

Additional Information

## Additional Context

  • CLAUDE.md file is ~12,000 words (loaded as project context)
  • No agents were intentionally launched (didn't check /tasks yet)
  • Interleaved thinking mode is enabled
  • This was my first session after daily quota reset on 2025-11-22

## Possible Causes

  1. Token counting bug: Project context being counted multiple times per prompt?
  2. Background agents: Agents spawned without user knowledge consuming quota?
  3. Thinking tokens: Internal thinking tokens being miscalculated?
  4. Weekly vs daily limit logic: Quota buckets are desynchronized?
  5. Context reloading bug: CLAUDE.md being reloaded repeatedly?

## Impact

  • Severity: Critical (tool becomes completely unusable)
  • User Impact: Weekly quota exhausted in single 30-minute session with minimal conversation
  • Workaround: None found - even removing project context doesn't explain the discrepancy
  • Business Impact: Forces users to consider switching to alternative tools

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗