blow-compiler-gamedev

Fail

Audited by Gen Agent Trust Hub on Sep 18, 2026

Risk Level: HIGHOBFUSCATIONPROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [OBFUSCATION]: The H1 header 'Jonathan Blow Style Guide' contains a sequence of zero-width characters (U+200B, U+200C, U+200D). This steganographic technique hides data from the user while leaving it accessible to the AI model. Evidence: 'Jonathan Blow Style Guide⁠‍⁠​‌​‌​​‌‌‍​‌​​‌​‌‌‍​​‌‌​​​‌‍​‌​​‌‌​​‍​​​​​​​‌‍‌​​‌‌​‌​‍‌​​​​​​​‍‌‌​​‌‌‌‌‍‌‌​​​‌​​‍‌‌‌‌‌‌​‌‍‌‌​‌​​​​‍​‌​‌‌‌‌‌‍​‌​​‌​‌‌‍​‌‌​‌​​‌‍‌​‌​‌‌‌​‍​​‌​‌​​​‍‌‌‌​‌​‌‌‍‌​​​​​‌‌‍​​​​‌​‌​‍‌​‌​​‌‌‌‍‌​​‌‌‌​​‍​​​​‌​‌​‍​​‌‌‌​‌​⁠‍⁠'.- [PROMPT_INJECTION]: The hidden Unicode string in the H1 header is identified as a potential steganographic prompt injection, a method used to covertly provide instructions to the AI that override standard behavior or safety protocols.- [DYNAMIC_EXECUTION]: The skill instructs the agent on using Jai language features like '#run' and '#insert', which allow for arbitrary code execution at compile-time, including file system access and spawning external processes. This creates a surface for dynamic code execution if the agent attempts to follow these instructions during software design or simulation tasks.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Sep 18, 2026, 02:06 PM