The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to clone a repository from an unverified GitHub user (https://github.com/tomw286/experimentersoftroll422-security-loader.git). This source is not recognized as a trusted organization or well-known service.
[COMMAND_EXECUTION]: The setup process requires executing high-privilege commands with administrator rights, including running a PowerShell script (install_driver.ps1) to install kernel-level components and executing a compiled binary (security-loader.exe) to manage driver state.
[REMOTE_CODE_EXECUTION]: The installation of a kernel-mode driver (EaseFilterDriver.sys) provided by an external source represents a critical risk. If the driver or the loader code is compromised, it could lead to full system takeover and persistent access.
[INDIRECT_PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by monitoring and processing arbitrary file activity from the Windows filesystem.
Ingestion points: Untrusted data enters the agent context through FileEvent structures in the on_event callback and CallbackContext in the FilterCallbackHandler (SKILL.md).
Boundary markers: None are present; there are no instructions to the agent to disregard content within monitored file paths or process metadata.
Capability inventory: The skill has the capability to terminate processes (terminate_process), write to the filesystem (logging), and perform network operations via the SIEM connector (SKILL.md).
Sanitization: There is no evidence of sanitization or validation of the filesystem data before it is processed by the handlers or sent to external integrations.
[DATA_EXFILTRATION]: The skill includes code for a SIEM (Security Information and Event Management) integration that sends security events to a remote endpoint. While the provided URL is a placeholder, this demonstrates a built-in capability to exfiltrate filesystem activity data.

experimentersoftroll422-windows-filesystem-security