The Agent Skills Directory

[DATA_EXFILTRATION]: The skill provides functional payloads designed to steal sensitive user information, including document cookies and session data. Evidence includes scripts that redirect the browser to an external domain with cookie data appended to the URL and the use of the fetch API to POST localStorage and cookie contents to an external server (attacker.com).- [OBFUSCATION]: The instructions detail how to bypass security controls using various obfuscation and encoding techniques. This includes the use of eval(atob(...)) for executing Base64-encoded JavaScript and the application of HTML entity, Hex, and Unicode encoding to evade detection by Web Application Firewalls (WAFs) and Content Security Policies (CSPs).- [COMMAND_EXECUTION]: The skill demonstrates the use of dangerous JavaScript sinks, such as eval(), setTimeout(), and the Function() constructor, to execute dynamically generated or encoded strings as code. These are identified as vulnerable sinks for DOM-based XSS exploitation.- [INDIRECT_PROMPT_INJECTION]: The skill processes data from untrusted external web applications, creating a vulnerability surface for indirect prompt injection.
Ingestion points: Targeted web application URLs and user input fields (documented in SKILL.md).
Boundary markers: The instructions lack the use of delimiters or specific warnings to ignore embedded instructions within the data being analyzed.
Capability inventory: The skill facilitates network requests and browser DOM manipulation, which could be misused if the target data contains malicious agent instructions.
Sanitization: There is no requirement or guidance for sanitizing or escaping content retrieved from target applications before processing.

Cross-Site Scripting and HTML Injection Testing