[BUG] Usage display shows 28% weekly but session is blocked — numbers don't reflect actual limit

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?

I was blocked from continuing work mid-session and told to purchase usage credits, despite the /usage command showing:

• Current session: 4% used
• Current week (all models): 28% used
• No additional features shown as consumed

The session cost shown was $24.77, heavily driven by cache reads (46M tokens cache read, 2M cache write) rather than expensive output tokens.

The displayed percentages do not reflect what limit was actually exhausted. If I'm 28% through my weekly limit, I should not be blocked. If I'm actually at 100% of something, the display should say so.

What Should Happen?

Show which specific limit (daily/weekly/model-specific/token-rate) was hit and what percentage of that limit is consumed. The current display is misleading and led me to spend $100 in usage credits I did not intend to use.

Error Messages/Logs

Total cost:            $63.44                         
  Total duration (API):  1h 10m 2s
  Total duration (wall): 4h 23m 9s
  Total code changes:    557 lines added, 109 lines removed
  Usage by model:                        
     claude-sonnet-4-6:  8.1k input, 247.7k output, 110.2m cache read, 7.1m cache write ($63.44)
    
  Current session
  ██                                                 4% used
  Resets 12:40am (America/Chicago)             

  Current week (all models)
  ██████████████                                     28% used
  Resets Jun 10 at 2pm (America/Chicago)

Steps to Reproduce

Upgrade from the $20 plan to the $100 plan
Do a /plan viewing every now and then to make sure things are okay
Continue working
You get a prompt to use your usage credits, saying something about context you don't understand.
Comply with the instructions.
Bang-whammo, a couple of hours later, you discover you've burned through $100 of additional credits.
Buy $20 more credits because Claude won't talk to your about your problem unless you add money.
Claude explains that somehow, in ways you still don't understand, that little graph has nothing to do with your token burn.
You're pissed off because you spent $200 this month and you can't afford any more, and you could have had the bigger plan for the same money, but the tools misguided you.

Claude Model

Sonnet (default)

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.153 (Claude Code)

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

Session

Total cost: $63.44
Total duration (API): 1h 10m 2s
Total duration (wall): 4h 23m 9s
Total code changes: 557 lines added, 109 lines removed
Usage by model:
claude-sonnet-4-6: 8.1k input, 247.7k output, 110.2m cache read, 7.1m cache write ($63.44)

Current session
██ 4% used
Resets 12:40am (America/Chicago)

Current week (all models)
██████████████ 28% used
Resets Jun 10 at 2pm (America/Chicago)

What's contributing to your limits usage?
Approximate, based on local sessions on this machine — does not include other devices or claude.ai

Last 24h · these are independent characteristics of your usage, not a breakdown

98% of your usage came from subagent-heavy sessions
Each subagent runs its own requests. Be deliberate about spawning them — and
consider configuring a cheaper model for simpler subagents.

98% of your usage came from sessions active for 8+ hours
These are often background/loop sessions. Continuous usage can add up quickly
so make sure it is intentional.

90% of your usage was at >150k context
Longer sessions are more expensive even when cached. /compact mid-task, /clear
when switching to new tasks.

44% of your usage came from MCP server
"plugin:iso20022-staging:iso20022-staging"
MCP tool results stay in context for the rest of the session. /compact to flush
them, or disable servers you don't need.

Subagents % of usage
general-purpose 3%

MCP servers % of usage
plugin:iso20022-staging:iso… 44%

d to day · w to week