xss-html-injection
Warn
Audited by Gen Agent Trust Hub on Jun 13, 2026
Risk Level: MEDIUMDATA_EXFILTRATIONREMOTE_CODE_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [DATA_EXFILTRATION]: The skill provides multiple functional payloads designed to exfiltrate sensitive user data to an external domain (attacker.com). Evidence includes a cookie theft payload using document.location redirect, a session hijacking template using the fetch API to POST cookies and localStorage data, and a keylogger script that transmits keystrokes via image source requests.
- [REMOTE_CODE_EXECUTION]: The skill documents numerous methods for achieving code execution within a browser context, which is the primary mechanism for XSS. It identifies dangerous DOM sinks such as eval(), document.write(), and innerHTML, and provides examples of triggering execution via script tags and event handlers like onload and onerror.
- [COMMAND_EXECUTION]: The instructions include techniques for executing obfuscated JavaScript to bypass detection. This includes the use of eval(atob('YWxlcnQoMSk=')) for Base64-encoded execution, as well as string concatenation and constructor-based execution to hide malicious patterns.
- [PROMPT_INJECTION]: The metadata description explicitly directs the AI agent to perform potentially malicious tasks such as 'steal cookies' and 'bypass content security policies', which serves as an instruction to override typical safety constraints.
Audit Metadata