The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill is configured to automatically download and install a pinned FreeCAD 1.1.x AppImage (approximately 1 GB) on Linux-x86_64 systems upon the first tool invocation. This binary download occurs from a remote source that is not explicitly detailed in the provided instructions.- [DYNAMIC_EXECUTION]: The skill includes execute_code and execute_code_async tools. These allow the agent to run arbitrary Python code within the FreeCAD environment, which can include operations that interact with the host system's resources.- [REMOTE_CODE_EXECUTION]: The combination of automatic binary downloading and the capability to execute arbitrary scripts constitutes a remote code execution surface. Malicious code or compromised binaries could lead to host-level command execution.- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from CAD documents (e.g., via get_objects and get_object) and uses this information to generate or modify Python scripts for execution. This creates a vulnerability where untrusted CAD files could influence the generated code.
Ingestion points: Data enters the context through get_objects, get_object, and list_documents in SKILL.md.
Boundary markers: The instructions do not specify delimiters or security warnings to prevent the agent from obeying instructions embedded in object properties or metadata.
Capability inventory: Sensitive tools include execute_code, execute_code_async, create_object, and edit_object.
Sanitization: There is no evidence of sanitization or character escaping for data retrieved from the FreeCAD model before it is used in prompt logic or script generation.

qwen-mm-plugins-freecad