The Agent Skills Directory

[OBFUSCATION]: The skill includes instructions for bypassing security filters using Base64 encoding combined with dynamic code execution (e.g., eval(atob('YWxlcnQoMSk='))). This pattern is a standard technique for concealing malicious payloads from static analysis and web application firewalls.
[DATA_EXFILTRATION]: Multiple code snippets demonstrate methods for stealing sensitive user data, such as document.cookie, localStorage, and real-time keystrokes. These payloads are configured to exfiltrate data to an external domain (attacker.com) via fetch, image requests, and location redirects.
[DYNAMIC_EXECUTION]: The skill identifies and encourages the use of dangerous JavaScript 'sinks' such as eval(), Function(), setTimeout(), and innerHTML. These functions are utilized to execute dynamically generated or obfuscated code within the browser context.
[INDIRECT_PROMPT_INJECTION]: The skill describes techniques for ingesting untrusted data from various sources (URL fragments, location.search, postMessage data, and localStorage). It lacks boundary markers or sanitization guidelines for the agent, potentially allowing malicious content from external websites to influence the agent's behavior during a testing session.

Cross-Site Scripting and HTML Injection Testing