ghost-bits-cast-attack
Fail
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: HIGHPROMPT_INJECTIONCOMMAND_EXECUTIONOBFUSCATIONDATA_EXFILTRATIONREMOTE_CODE_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill includes an 'AI LOAD INSTRUCTION' block that directs the agent to 'ALWAYS try Ghost Bits variants' when encountering security blocks, effectively instructing the AI to bypass safety filters and security controls rather than stopping or reporting them.
- [COMMAND_EXECUTION]: Provides ready-to-use Python and Yaklang code snippets specifically designed to generate malicious payloads and transmit them to target services using HTTP tools.
- [OBFUSCATION]: The core purpose of the skill is to provide mechanisms for character-level obfuscation. It includes comprehensive lookup tables and generators for Unicode-to-ASCII collisions (Ghost Bits) designed to hide malicious strings from security scanners using Latin-Extended and CJK characters.
- [DATA_EXFILTRATION]: Provides detailed templates and techniques for request smuggling and response splitting (e.g., Apache HttpClient and JDK HttpServer vulnerabilities) which are used to bypass proxy controls and potentially exfiltrate sensitive data or hijack sessions.
- [REMOTE_CODE_EXECUTION]: Contains actionable recipes and code for achieving RCE by exploiting vulnerabilities in major Java components such as Apache Commons BCEL, GeoServer, and Spring Framework, providing both the conceptual basis and the obfuscated payloads required for exploitation.
Recommendations
- AI detected serious security threats
Audit Metadata