The Agent Skills Directory

[OBFUSCATION]: The skill provides multiple examples of techniques used to hide malicious code and bypass security filters.
Base64 encoding: The skill demonstrates executing obfuscated JavaScript using eval(atob('YWxlcnQoMSk=')), which decodes to alert(1).
Character encoding: Examples include HTML entity encoding (alert(1)), Hex encoding (alert(1)), and Unicode escapes (\u0061lert(1)).
Code masking: Demonstrates evading detection through string concatenation ('al'+'ert(1)') and using the Function() constructor or array constructors to execute code without using typical keywords.
[DATA_EXFILTRATION]: The skill documents procedures and provides payloads for exfiltrating sensitive information to a third-party domain (attacker.com).
Cookie theft: Templates are provided to send document.cookie to external servers via document.location redirection and Image object requests.
Keylogging: Payloads demonstrate capturing e.key events and exfiltrating them in real-time to an external URL.
Session exfiltration: Includes fetch templates to POST localStorage and cookie data to external endpoints.
[DYNAMIC_EXECUTION]: The skill relies on and promotes the use of dangerous sinks for executing dynamically generated or obfuscated JavaScript.
It identifies and encourages the use of eval(), Function(), setTimeout(), and innerHTML as primary methods for achieving code execution during assessments.
[INDIRECT_PROMPT_INJECTION]: The methodology described involves the agent ingesting and processing potentially malicious untrusted data from web applications.
Ingestion points: Search boxes, query parameters, user profile fields, URL fragments, and HTTP headers as identified in Phase 1.
Boundary markers: None; the instructions lack delimiters or warnings to treat reflected data as untrusted text rather than instructions.
Capability inventory: The skill instructs the agent to use browser developer tools and manual console execution to interact with targets.
Sanitization: The skill focuses on exploiting the absence of sanitization rather than implementing it.

Cross-Site Scripting and HTML Injection Testing