[Bug] Anthropic API Error: Safeguard Blocked Message During Multi-Session Transcript Investigation

**Bug Description**
after Fable 5 failed to read other session transcript, it asked me to do another experiment between two claude sessions. I did and safeguard blocked its investigation in session transcript.

❯ [Image #7] and after /exit and --resume [Image #8]
[Image #7]
[Image #8]

That's the complete picture — and your two screenshots caught the subtle failure mode exactly. Terminal 1's answer ("you asked me to remember the codeword") proves its in-memory context never saw the injection; the resumed session showing only terminal 1's turns suggests the injected exchange got orphaned once terminal 1 continued. Let me confirm the mechanism on disk — structure only, parent links, no content:

Ran 1 shell command

⏺ API Error: Fable 5's safeguards flagged this message (https://www.anthropic.com/legal/aup). This sometimes happens with safe, normal conversations. Claude Code can't respond to this message with Fable 5.

Double press esc to edit your last message, or try a different model with /model.

Send feedback with /feedback or learn more: https://support.claude.com/en/articles/15363606

Details: `[reasoning_extraction]`

Request ID: req_011CeDUika6To5pfhU5nqjp9

**Environment Info**
- Platform: darwin
- Terminal: ghostty
- Version: 2.1.237
- Feedback ID: 2cf8eb05-8408-4d4e-bf58-9a080734397e

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

Reproduce the two-session transcript flow in Claude Code, then retry after /exit and --resume while observing the reported reasoning_extraction safeguard error. Compare the resumed transcript behavior with the screenshots and use request ID req_011CeDUika6To5pfhU5nqjp9 and feedback ID 2cf8eb05-8408-4d4e-bf58-9a080734397e when checking findings; done means the failure mode is isolated.

Rédigé par le modèle d'indexation à partir du texte de l'issue.