anthropics / anthropics/claude-code

[BUG] Continuing from a session limit hit causes 45% of immediate quota usage (Fable 5.1)

未關閉
#92,903 0 則留言 0 個 reaction 已指派 0 人 在 GitHub 檢視
area:cost bug platform:linux
主要語言
Python
星號
145k
分支
23.1k
PR 合併指標
PR 指標待擷取

描述

### Preflight Checklist

- [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) 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?

When I hit the limits with Fable 5.1 and I have the auto-resume feature on, even asking a simple question causes immediate burn of ~half of the quota.

I assume this has something to do with cache being automatically invalidated after 5 hours of wait and this causing high usage; still it's very frustrating to wait 5 hours just to lose most of the quota immediately.

Image

### What Should Happen?

If this is indeed a cache issue Claude code should probably warn the user and ask if they want an auto-compacting before execution, this could save a lot of usage.

### Error Messages/Logs

```shell

```

### Steps to Reproduce

1) Have a long session that uses Fable 5.1 on a Max plan ~500-600K tokens
2) Hit the quota
3) Prepare & send a simple message like "stop whatever you were doing and give me a short wrap up message, make sure not to burn any token"
4) Wait and observe most of the quota disappear

### Claude Model

Other

### Is this a regression?

No, this never worked

### Last Working Version

_No response_

### Claude Code Version

2.1.263

### Platform

Other

### Operating System

Ubuntu/Debian Linux

### Terminal/Shell

Other

### Additional Information

_No response_

貢獻指南

這個儲存庫沒有索引到貢獻指南

研究方向

No source file, test, or entry point is identified. Start by reproducing the quota jump with auto-resume after a Fable 5.1 session limit, then trace the continuation and cache behavior; done would require a defined, verified outcome for quota usage or user warning.

由索引模型根據 Issue 內容生成。

評估

領域
cli
Issue 類型
缺陷
難度
5/5
預估耗時
一週以上
活躍度
活躍
描述清晰度
需要釐清
新手友好度
25/100

把新 issue 寄到你的電子郵件信箱

精選適合新手參與的 GitHub issue 摘要。