The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to clone a repository from a GitHub account (henry-lewiskpp1107/acidrain-security-script-hub) that is not associated with the skill's author or a trusted organization.
[REMOTE_CODE_EXECUTION]: The skill provides instructions and examples for executing scripts downloaded from the external repository, including Node.js automated scanners and PHP injection tests.
[DATA_EXFILTRATION]: Includes code specifically designed to exfiltrate document.cookie data to a remote endpoint defined by an environment variable.
[OBFUSCATION]: Demonstrates the use of Base64 encoding (atob) to wrap JavaScript code which is then executed via eval().
[COMMAND_EXECUTION]: Includes documentation and shell scripts that execute commands such as git clone, curl, and python3 to facilitate security testing and remote resource interaction.
[DYNAMIC_EXECUTION]: Utilizes eval() in browser-based payloads, which executes encoded strings at runtime. This pattern is often used to hide malicious intent.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The testReflection function in scripts/javascript/reflection-detector.js and PHP scripts like scripts/php/input-validation-test.php ingest untrusted data from URL parameters.
Boundary markers: None present; the scripts are designed to reflect or process raw input to demonstrate vulnerabilities.
Capability inventory: The skill includes network exfiltration capabilities (fetch), dynamic code execution (eval), and database query execution via PDO.
Sanitization: Lacking by design in several examples, which could allow maliciously crafted input to trigger unintended behaviors in the agent's execution context.

acidrain-security-testing-scripts