The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/native_shell_executor.gd provides a wrapper around OS.execute(), allowing for arbitrary shell command execution with captured output. This is a high-privilege capability that could be abused if an agent passes unsanitized input to the command or arguments.
[COMMAND_EXECUTION]: The platform_desktop_patterns.gd script uses OS.create_process() to fork background processes and OS.shell_open() to open URLs or files in the system's default handler, which are powerful system-level interactions.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection through several data ingestion points:
Ingestion points: scripts/native_dialog_helper.gd (file dialog results) and scripts/platform_desktop_patterns.gd (_on_files_dropped callback) ingest external file paths and user-selected data into the application state.
Boundary markers: No delimiters or instructions are present to prevent the agent from being influenced by data found in these external file paths.
Capability inventory: The skill includes OS.execute() in scripts/native_shell_executor.gd and OS.create_process() in scripts/platform_desktop_patterns.gd.
Sanitization: There is no evidence of sanitization or validation of the ingested file paths or command arguments before they are passed to execution APIs.

godot-platform-desktop