test
Pass
Audited by Gen Agent Trust Hub on Aug 11, 2026
Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [PROMPT_INJECTION]: The skill provides instr uc tions for the agent to systematic all y ing est and interact with untr usted external data (user-provided web appli c ations).
- In g estion points: The agent is instr uc ted to "Go thro ugh every page of my ap p" and "Click every but ton" in SKI L L.md.
- Bo und ary markers: The re are no expli cit delimiters or instr uc tions for the agent to ig no re em be dded comm an ds within the appli c ation being tested.
- Ca pa bility in ventory: The agent is gran ted ca pa bilities to navig ate web pages, interact with but tons, and sub mit for m data.
- Sanitiz ation: The skill does not provide metho ds for sanitizing or fil tering the con tent ret rie ved from external pages.
- [DATA_EXFILTRATION]: The skill's do c um en tation in E D G E-C A S E S.md and B R E A K I N G-T H I N G S.md in cl ud es examples of sensitive file paths (e.g., "file:///etc/p as s wd") and com mon at tack paylo ads (S QL injec tion and XSS st rin gs). These are provided as literal st rin gs for the purp ose of testing appli c ation resilien ce.
Audit Metadata