[Bug] Agent fails on basic reasoning tasks that should be trivial for Claude 3.5 Sonnet

**Bug Description**
agent chasing secnarios that ANY basic AI harnest should be able to answer, let it alone fable 5.1

**Environment Info**
- Platform: darwin
- Terminal: Apple_Terminal
- Version: 2.1.266
- Feedback ID: c2471bde-63de-41a8-9c09-e0eb36aca867

**Errors**
```json
[]
```

The issue gives only environment details: darwin, Apple_Terminal, Claude Code 2.1.266, and Feedback ID c2471bde-63de-41a8-9c09-e0eb36aca867. Start by finding or requesting a minimal reasoning prompt and expected answer; as written, there are no files, tests, or reproduction steps. Done would need a reproducible failing case and evidence the agent behavior is fixed.

索引モデルが issue の本文から書いたものです。