[Bug]: Exhausted Fable weekly limit marks all Claude models as out of usage #32015

Before submitting

• I searched existing issues and did not find a duplicate.
• I included enough detail to reproduce or investigate the problem.

Area

packages/contracts or packages/shared

Steps to reproduce

1. Use a Claude subscription where the model-scoped weekly bucket for Fable is used up. The Limits tab shows Weekly · Fable at 100%, while Session and Weekly are below 100%.
2. Make a Claude model that is not Fable, such as claude-opus-5-5, the default for new threads, with a fallback provider configured.
3. Start a new thread.

Expected behavior

Only Fable is treated as out of usage. Opus 5.5 is still within its limits, so it stays usable and new threads keep using it.

Actual behavior

T3 Code treats the whole Claude provider as out of usage. New-thread selection falls back to the other provider, and the server's text-generation model selection skips Claude too, even though Opus 5.5 still works.

Cause: exhaustedUsageWindow in packages/shared/src/usageLimits.ts has model-aware cases for cursor, antigravity and opencode. claudeAgent falls through to default: return exhausted[0], so any Claude window at 100% blocks every Claude model. That includes the model-scoped seven_day_<model> windows from claudeUsageResponseToLimits (apps/server/src/provider/claudeUsageLimits.ts), such as seven_day_fable.

Callers affected: resolveNewThreadModelSelection (ChatView, DraftHeroHeadline, useHandleNewThread, MCP project handlers) and TextGeneration.ts.

Impact

Major degradation or frequent failure

Version or commit

main @ 6577c6b

Environment

macOS, T3 Code with the Claude Agent provider, Claude subscription with a model-scoped Fable weekly limit

Workaround

Pick the Claude model explicitly for each thread. An explicit selection skips the fallback.

Related

#30722 confirmed that the Weekly · Fable row belongs under Claude. This issue is about that row blocking the other Claude models.