The Agent Skills Directory

Dynamic Component Execution: The gr.HTML component documentation and examples in SKILL.md describe the use of js_on_load, html_template, and css_template.
Pattern: This allows for the injection of arbitrary JavaScript and CSS into the rendered UI.
Context: While a standard extensibility feature of Gradio for creating custom components, it allows for the execution of code in the user's browser context.
External Resource Reference: The StarRating example in SKILL.md fetches a decorative image from a remote source.
Evidence: https://upload.wikimedia.org/wikipedia/commons/d/df/Award-star-gold-3d.svg (Wikimedia Commons).
Context: This is a reference to a well-known, trusted media repository for UI assets.
Network Interaction via CLI: The SKILL.md file documents the gradio info and gradio predict CLI commands.
Pattern: These commands are designed to interact with remote Gradio Spaces and endpoints to discover parameters and send prediction data.
Context: This is standard functionality for programmatically interacting with Gradio applications.
Indirect Prompt Injection Surface: As a UI-building toolset, the skill provides numerous components for ingesting untrusted data.
Ingestion points: Untrusted data enters the context through gr.Textbox, gr.File, gr.Audio, gr.Image, and gr.Chatbot components across both SKILL.md and examples.md.
Boundary markers: Generic examples do not typically include explicit boundary markers or "ignore" instructions for the data they process.
Capability inventory: The skill facilitates wiring these inputs to arbitrary Python functions, and the gradio CLI provides network interaction capabilities.
Sanitization: Components like gr.Chatbot and gr.Markdown include sanitize_html=True by default to mitigate typical web-based injection attacks.

huggingface-gradio