The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill retrieves design references and media assets from well-known services to inform its creative process. This includes fetching 3D models and textures from Poly Haven, fonts from Google Fonts, and images from Openverse and Are.na. These downloads are performed by the source.mjs and moodboard.mjs scripts to build a project's asset library.
[COMMAND_EXECUTION]: The framework executes several local utility scripts via the terminal tool to automate design tasks. This includes slopscan.mjs for CSS/JS linting, shoot.mjs for multi-breakpoint screenshots, and motionqa.mjs for performance profiling. It also utilizes system binaries like ffmpeg for media processing and npx for bundling dependencies with esbuild.
[DYNAMIC_EXECUTION]: The assets.md reference file describes a process for generating depth maps where the agent creates and executes a short Python script. This script utilizes the transformers library to perform depth estimation on hero images, which is a core component of the skill's 2.5D visual effect capabilities.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it ingests untrusted data from external websites, such as gallery descriptions from Awwwards or metadata from stock asset providers. However, the skill implements defensive instructions that direct the agent to treat this data exclusively as non-executable reference material and uses specific parsing logic to extract intended fields.

auteur