The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill includes instructions to download and execute the 'uv' installation script from 'https://astral.sh/uv/install.sh'. This is the official installation method for a well-known Python toolchain component and is considered a trusted source.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted image data using vision models, which could potentially contain malicious instructions embedded within the image text.
Ingestion points: Image files processed by image_probe.py and visual_runtime.py as specified in the SKILL.md file.
Boundary markers: The skill uses a structured system prompt (assets/prompts/ui-semantic.md) to guide the model, although it lacks explicit negative constraints against instructions embedded in user data.
Capability inventory: The skill can execute local Python scripts via uv run and write layout specifications to the local file system (e.g., in the artifacts/ui-layout/ directory).
Sanitization: The visual_runtime.py script includes a normalize_result function and JSON repair logic to ensure the output from the vision model adheres to the expected schema before it is processed by the agent.

visual-ui-layout-spec