HTML Injection Testing
Fail
Audited by Gen Agent Trust Hub on Mar 16, 2026
Risk Level: HIGHDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill metadata and description contain explicit instructions to trigger the agent to perform malicious acts, such as 'perform HTML injection attacks' and 'deface web applications', which directly encourage behavior that bypasses standard AI safety guidelines.
- [DATA_EXFILTRATION]: The skill provides numerous payloads designed to harvest and exfiltrate sensitive user credentials to external, untrusted domains like 'attacker.com' and 'evil.com' via phishing form overlays and CSS tracking techniques.
- [COMMAND_EXECUTION]: The skill documentation includes ready-to-use 'curl' commands and a functional Python fuzzer script intended to automate the discovery and exploitation of injection vulnerabilities on target web servers.
Recommendations
- AI detected serious security threats
Audit Metadata