The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/run_loop.py uses subprocess.run to invoke the ffmpeg command-line utility. This is used for extracting individual frames from video files as part of the animation loop creation process and is a standard operation for the skill's intended purpose.
[DYNAMIC_EXECUTION]: The SKILL.md file contains instructions for executing the build_scene.py script within an active Blender session using the exec(open(...).read()) method. This is a common technique for running external scripts inside specialized environments like Blender and does not pose a threat in this context.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it incorporates user-supplied prompts into AI model calls, and then processes the resulting assets with local scripts.
Ingestion points: User-provided prompts for the recommend and model_run tools in SKILL.md.
Boundary markers: Not present; the instructions do not include specific delimiters or safety warnings to separate user instructions from processed data.
Capability inventory: The skill can perform local file writes, image processing with Pillow, and subprocess execution of ffmpeg.
Sanitization: No specific validation or sanitization of the generated assets' content or metadata is performed before they are processed by the skill's Python scripts.

scenario-blender-2d-animation