[BUG] "Out of usage credits" blocks chat while Plan usage limits panel shows headroom (Pro plan)

### 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?

### Preflight Checklist
- [x] I have searched existing issues and this hasn't been reported yet (related but distinct: #61125, #67412)
- [x] This is a single bug report
- [x] I am using the latest version of claude.ai (web)

### What's Wrong?

**Summary**
On claude.ai (Pro plan), sending a message is blocked with "You're out of usage credits. Buy more to keep going now, or wait until [time] when your plan usage resets" — even though the Plan usage limits panel (opened at the same time) shows both Current session and Weekly limits with significant headroom (well under 100%).

**Steps to reproduce**
1. Be on a Pro plan, mid-session, with normal usage.
2. Attempt to send a message in claude.ai chat.
3. Get blocked with "You're out of usage credits" message, citing a reset time matching the *session* limit reset (not weekly).
4. Open the Plan usage limits panel (account menu → usage) in parallel.
5. Panel shows: Current session ~47-49% used, resets in ~2h53m-2h56m; Weekly (all models) 54% used, resets Sunday 10:00 AM.
6. Click the refresh icon in the panel — values update slightly (47%→49%) but the blocking message persists unchanged, and no way to retry/dismiss without waiting or buying credits.

**Expected behavior**
If the usage limits panel shows headroom on both session and weekly limits, the chat should not be blocked. The two surfaces (blocking gate vs. usage panel) should read from the same source of truth.

**Actual behavior**
The blocking gate appears to fire independently of the displayed usage percentages, showing a hard lockout with "Buy more" as the only visible action, despite the usage panel confirming the limits are not exhausted.

**Additional notes**
- No way to file feedback on this specific event via the normal 👎 mechanism, since the block happens *before* Claude generates any response to react to.
- This looks like the same underlying issue as #61125 ("Out of usage credits" banner shown while session and weekly limits both have headroom), but observed on claude.ai web chat rather than Claude Code — flagging in case it's useful confirmation that this isn't CLI-specific.

**Environment**
- Surface: claude.ai (web)
- Plan: Pro
- Screenshots: [attach the two usage-panel screenshots — 47% used / 49% used, taken minutes apart, refresh clicked between them]

### What Should Happen?

The chat should remain usable when the Plan usage limits panel shows headroom on both Current session and Weekly limits. The blocking gate and the usage panel should read from the same underlying usage data, so a user is never shown "you have used less than 50% of your limit" and "you're out of usage credits" at the same time. If a hard lockout is triggered, the panel should reflect that (e.g. show 100% used) instead of contradicting the block.

### Error Messages/Logs

```shell

```

### Steps to Reproduce

Be logged into claude.ai on a Pro plan, with normal usage during the day (not right after a reset).
Send a message in an ongoing chat.
Receive the block: "You're out of usage credits. Buy more to keep going now, or wait until [time] when your plan usage resets."
Open the account menu → Plan usage limits panel in the same browser session, immediately after.
Observe: Current session shows ~47-49% used (well under limit), resets in ~2h53m-2h56m; Weekly "All models" shows 54% used, resets Sunday 10:00 AM.
Click the refresh icon in the panel — the percentage updates slightly (e.g. 47% → 49%), confirming the panel is pulling live data, but the chat is still blocked with the same "out of usage credits" message.
No way to retry or dismiss the block — only options shown are "Buy more" or wait for reset.

### Claude Model

Opus

### Is this a regression?

I don't know

### Last Working Version

_No response_

### Claude Code Version

2.1.238 (Claude Code)

### Platform

Anthropic API

### Operating System

Windows

### Terminal/Shell

Cursor

### Additional Information

_No response_

No repository files, tests, or entry points are identified in the report. Start by reproducing the Pro-plan block on claude.ai while comparing the chat gate with the Plan usage limits panel; done means both surfaces agree and chat remains available when the displayed session and weekly limits have headroom.

由索引模型根据 Issue 内容生成。