anthropics / anthropics/claude-code

[Bug] Fable safeguards causing excessive false positives and forced model switching

オープン
#94,279 コメント 0 件 リアクション 0 件 担当者 0 名 GitHub で見る
area:model bug platform:windows
主要言語
Python
スター
145k
フォーク
23.1k
PR マージ指標
PR 指標を取得中

説明

**Bug Description**
Your Fable safeguards are buggy as hell. I can not use Fable for any security/cryptography related work, and even in normal work messages are being flagged for nonsense. Those safeguards force a model switch, which does then result in lower quality output as well. That's not useable in any professional environment. That issue has been opened many times by others as well, and then closed because it was ignored. Seriously not good advertising.

**Environment Info**
- Platform: win32
- Terminal: null
- Version: 2.1.270
- Feedback ID: 2712c54b-997f-4780-9f74-5aac13d50e08

**Errors**
```txt
● Bash(sed -n 96,245p "Project Name/TenantExport_Tests.DeliveryState.cs")
⎿   ///
/// A tenant folder carrying only part of a capability's sources switches the whole capability off:
/// every one of its tables goes out as "not delivered" and the run completes. That replaces the earlier
… +142 lines (ctrl+o to expand)

Searched for 2 patterns (ctrl+o to expand)

● Fable 5.1's safeguards flagged this message. Our intentionally broad safeguards allow us to deliver more capabilities faster, but can sometimes flag
legitimate coding, cybersecurity, and biology tasks. Switched to Opus 5. Send feedback with /feedback or learn more:
https://support.claude.com/en/articles/15363606

Details: `[bio]`
```

コントリビューションガイド

このリポジトリのコントリビューションガイドは索引されていません

評価

この issue はまだ評価されていません。

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。