The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the subprocess module and command-line recipes to interact with Sandboxie-Plus components (Start.exe, SbieIni.exe, sc.exe). These commands are used to launch sandboxed processes, terminate instances, and reload configuration files, which is necessary for the skill's operational purpose.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates a workflow for executing untrusted payloads and collecting resulting artifacts from the sandbox file silo (C:\Sandbox\). This creates a theoretical attack surface where malicious data generated by a sandboxed application could influence an agent if it later processes those files.
Ingestion points: Artifact collection routine in run_sandboxed_payload.py that reads from the virtualized sandbox root.
Boundary markers: None explicitly defined in the provided automation scripts.
Capability inventory: Subprocess execution, file system management (shutil.copy2), and native API interaction via ctypes.
Sanitization: The skill recommends defensive INI settings (BlockInternetAccess=y, DropAdminRights=y) but does not perform content-level sanitization on harvested files.
[PRIVILEGE_ESCALATION]: Troubleshooting and installation instructions involve starting the Sandboxie kernel driver service (sc start SbieDrv) and modifying global configuration files (C:\Windows\Sandboxie.ini), both of which require administrative privileges to execute.

sandboxie-plus