The Agent Skills Directory

External Resource Dependencies: The skill contains instructions for downloading and installing code from various external repositories (e.g., GitHub and community-managed hubs) to support specialized tasks like 3D asset generation and GPU-accelerated computation. These dependencies are common in the machine learning research community but represent external code sources.
Runtime Environment Configuration: To address hardware-specific requirements in restricted build environments, the skill describes methods for installing and compiling software packages at startup using shell commands (e.g., subprocess.run, os.system). These patterns are intended to optimize the application environment for high-performance models.
Indirect Prompt Injection Surface: The skill guides the creation of applications that ingest untrusted user data, such as text prompts or images, for model inference.
Ingestion points: Interactive input components like gr.Image and gr.Textbox described in references/gradio.md.
Boundary markers: The instructions do not explicitly detail the implementation of delimiters or markers to isolate user-supplied content from application instructions.
Capability inventory: Resulting applications typically possess capabilities for file system operations, network downloads for model weights, and command execution for environment setup (references/3d-cuda-extensions.md).
Sanitization: The instructions assume the use of standard development kits which provide baseline sanitization for external content.

huggingface-spaces