abramov-state-composition
Fail
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: HIGHOBFUSCATIONPROMPT_INJECTION
Full Analysis
- [OBFUSCATION]: The main title in
SKILL.mdcontains 207 hidden characters using steganographic zero-width Unicode characters (U+200B, U+200C, U+200D, and U+2060). This technique is specifically designed to evade static analysis and human inspection. - [PROMPT_INJECTION]: The hidden Unicode sequence in the header of
SKILL.mdacts as a steganographic prompt injection vector. By embedding instructions that are invisible to the user but part of the model's input stream, the skill attempts to silently influence the agent's behavior, override system instructions, or potentially exfiltrate session data.
Recommendations
- AI detected serious security threats
Audit Metadata