The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes external model card data from Hugging Face repositories to infer inference parameters and UI design.
Ingestion points: The agent fetches model card text using huggingface_hub.ModelCard.load(repo_id) in Phase 1 of SKILL.md.
Boundary markers: No explicit delimiters or instructions are provided to the agent to treat model card content as untrusted data.
Capability inventory: The skill can write code (app.py), upload files to remote repositories (HfApi().upload_file), and execute CLI commands for testing (gradio predict).
Sanitization: The skill does not perform specific sanitization or escaping of the fetched text before using it to generate the application files.
[DYNAMIC_EXECUTION]: The skill generates and writes executable Python code (app.py), configuration files (README.md), and dependency lists (requirements.txt) at runtime based on user input and external repo data. This is an intended function of the skill but represents a dynamic execution surface.
[COMMAND_EXECUTION]: The skill uses the gradio CLI tool to perform smoke tests on the generated application. It specifically uses gradio info and gradio predict commands (Phase 6 of SKILL.md) to verify the functionality of the published Space.
[EXTERNAL_DOWNLOADS]: The skill fetches libraries and configuration from well-known and trusted external sources.
It fetches the official diffusers library from Hugging Face's GitHub repository.
It downloads dependencies from Lightricks' official GitHub repositories for LTX video models.
It uses standard content delivery networks (CDNs) such as Cloudflare (cdnjs), jsdelivr, and unpkg for Gradio custom components.
[PROMPT_INJECTION]: A static detector flagged potential concealment instructions in references/adapting-to-the-lora.md. Analysis confirms this is a false positive; the instructions to "Hide the steps slider" refer to optimizing the user interface of the generated Gradio application for specific model types, rather than an attempt to conceal the agent's actions from the user.

huggingface-lora-space-builder