The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external websites, including HTML structure, CSS, script bundles, and WebGL shader source, which creates a potential surface for indirect prompt injection.
Ingestion points: Tools like inspect_page, capture_page, and analyze_motion accept external URLs as input.
Boundary markers: Not explicitly specified in instructions, but the workflow utilizes structured tool outputs.
Capability inventory: The skill can scaffold projects using build_project and perform browser-based verification using verify_build.
Sanitization: The documentation explicitly states that build_motion does not execute captured code or shader source. Furthermore, verify_build and capture_page include DNS and private-IP guards to prevent SSRF and unauthorized access to internal or loopback resources.
[EXTERNAL_DOWNLOADS]: The documentation mentions an optional dependency on the spectorjs package (v0.9.33) for WebGL frame analysis. This is a well-known developer tool used for debugging graphics commands and is used here in a controlled manner within the tool suite.

awwwards-section-loop