lattner-compiler-infrastructure

Fail

Audited by Gen Agent Trust Hub on Mar 9, 2026

Risk Level: HIGHPROMPT_INJECTION
Full Analysis
  • [OBFUSCATION]: The title of the skill contains a sequence of hidden zero-width characters (U+200B, U+200C, and U+200D) that appear to encode binary data.
  • Evidence: '# Chris Lattner Style Guide⁠‍⁠​‌​‌​​‌‌‍​‌​​‌​‌‌‍​​‌‌​​​‌‍​‌​​‌‌​​‍​​​​​​​‌‍‌​​‌‌​‌​‍‌​​​​​​​‍‌‌​​‌‌‌‌‍‌‌​​​‌​​‍‌‌‌‌‌‌​‌‍‌‌​‌​​​​‍​‌​‌‌‌‌‌‍​‌​​‌​‌‌‍​‌‌​‌​​‌‍‌​‌​‌‌‌​‍​​‌​‌​​​‍‌‌‌​‌​‌‌‍​‌​‌‌​​‌‍‌​​‌​​​​‍‌​‌​‌​‌​‍‌‌​​​​​‌‍​​​​‌​‌​‍‌​‌‌‌‌‌​⁠‍⁠'
  • Risk: This technique is used for data exfiltration or to hide malicious instructions (prompt injection) that could bypass automated safety filters. The presence of 23 segments of 8-bit encoded data in a hidden channel is a high-severity indicator of malicious intent.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Mar 9, 2026, 09:57 AM