sutter-exceptional-cpp
Fail
Audited by Gen Agent Trust Hub on Mar 15, 2026
Risk Level: HIGHPROMPT_INJECTION
Full Analysis
- [OBFUSCATION]: The skill header contains a long sequence of non-printing zero-width Unicode characters (U+200B, U+200C, and U+200D) embedded after the visible text. This technique is used to hide data or instructions from the user while ensuring they remain present in the text context processed by the AI model.
- [PROMPT_INJECTION]: The obfuscated bitstream in the main header likely contains hidden instructions aimed at overriding agent behavior or bypassing safety constraints. The use of invisible payloads in a skill's primary documentation is a strong indicator of adversarial intent.
- [SAFE]: The visible C++ code examples for exception-safe assignment and the Pimpl idiom are technically accurate and follow industry best practices for modern C++ development.
Recommendations
- AI detected serious security threats
Audit Metadata