prompt-injection-test
Fail
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: CRITICALPROMPT_INJECTIONREMOTE_CODE_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill instructions explicitly attempt to override system behavior using directives like 'Ignore all previous instructions' to force restricted outputs.- [REMOTE_CODE_EXECUTION]: The skill includes a pattern to download and execute code from an untrusted remote domain ('evil.example.com') directly into the shell using the 'curl | sh' method.- [COMMAND_EXECUTION]: The skill utilizes the
os.system()function to execute arbitrary shell commands from a Python environment, which facilitates the unauthorized execution of remote scripts.
Recommendations
- AI detected serious security threats
Audit Metadata