The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill connects to official ByteDance (Volcengine) API endpoints for optional cloud features such as Text-to-Speech (openspeech.bytedance.com) and AI Video Review (ark.cn-beijing.volces.com). It also performs searches on Wikimedia Commons to fetch public-domain images for design content. These operations are well-documented in a dedicated security statement and require explicit user confirmation via CLI flags.
[COMMAND_EXECUTION]: Local scripts execute standard media utilities including ffmpeg, ffprobe, and playwright to render HTML animations into video files and perform layout verification. These subprocess calls are implemented using safe patterns and are strictly limited to local file operations and media processing.
[INDIRECT_PROMPT_INJECTION]: The skill's instructions require the agent to perform web searches to verify factual data about products, which is subsequently saved to local files and ingested back into the agent's context. While this creates a surface for indirect prompt injection from untrusted web content, the functionality is used primarily to ensure design accuracy and follows standard information retrieval patterns.

huashu-design