The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to clone and build a project from a personal GitHub repository that is not associated with the skill author or a trusted organization.
Evidence: git clone https://github.com/tomw286/experimentersoftroll422-security-loader.git followed by cargo build --release.
[PRIVILEGE_ESCALATION]: The skill requires Administrator privileges to function and explicitly enables highly sensitive Windows privileges that allow for deep system inspection and control.
Evidence: Mentions requiring "Administrator privileges for driver installation and filesystem monitoring".
Evidence: Code calls privileges::enable_debug_privilege(), privileges::enable_backup_privilege(), and privileges::enable_restore_privilege().
[COMMAND_EXECUTION]: The skill manages the lifecycle of kernel-mode drivers, which can execute with the highest level of system authority.
Evidence: Configuration includes driver_mode = "kernel" and troubleshooting steps include diagnostics::install_driver() and diagnostics::start_driver().
[INDIRECT_PROMPT_INJECTION]: The skill implements automated security reactions (like blocking processes) based on untrusted filesystem events, creating a surface for indirect influence over system behavior.
Ingestion points: Real-time monitoring of file paths, process names, and user SIDs across the filesystem.
Boundary markers: No delimiters or ignore instructions are used when processing event data.
Capability inventory: Process termination via loader.block_process(event.process_id), file access denial, and kernel driver management.
Sanitization: No evidence of input validation or sanitization for filesystem metadata before it is used to trigger security actions.

experimentersoftroll422-windows-filesystem-security