The Agent Skills Directory

[COMMAND_EXECUTION]: The skill installs and starts Windows kernel-mode drivers using the diagnostics::install_driver() and diagnostics::start_driver() functions.
[COMMAND_EXECUTION]: The skill explicitly requests and enables high-privilege Windows security tokens, including SeDebugPrivilege, SeBackupPrivilege, and SeRestorePrivilege, which are typically used for administrative or system-level tasks.
[EXTERNAL_DOWNLOADS]: The installation instructions require cloning a repository from an external GitHub account (tomw286/experimentersoftroll422-security-loader.git) and building the binary locally, which introduces external code from an unknown source into the environment.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface by ingesting untrusted filesystem metadata (such as process names, paths, and user SIDs) which are then processed by the agent to make security decisions or generate logs.
Ingestion points: loader.on_file_event and loader.on_access_request in SKILL.md.
Boundary markers: None identified in the prompt logic.
Capability inventory: Process termination (block_process), driver installation, and filesystem access control policies.
Sanitization: No explicit validation or escaping of external filesystem strings before they are used in logging or decision-making logic.

experimentersoftroll422-windows-filesystem-security