The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection through its core workflows.
Ingestion points: The skill is instructed to perform WebSearch to verify product facts (SKILL.md) and to read user-provided codebases, screenshots, and design systems (references/design-context.md) to extract design tokens.
Boundary markers: There are no explicit instructions or delimiters used to separate untrusted external content from the agent's instructions, nor are there warnings to ignore embedded commands within ingested files.
Capability inventory: The skill possesses significant execution capabilities, including running shell commands through various scripts in the scripts/ directory that invoke ffmpeg, ffprobe, node, and python. It also uses npx for rendering and managing the hyperframes animation backend.
Sanitization: No explicit sanitization, escaping, or validation of ingested external content is mentioned before it is interpolated into the agent's context.
[COMMAND_EXECUTION]: The skill relies extensively on executing local system commands to perform its primary functions. It uses ffmpeg and ffprobe for audio/video processing and validation (scripts/add-music.sh, scripts/verify-video.sh) and playwright via Chromium for rendering HTML designs and taking screenshots (scripts/render-video.js, scripts/verify.py).
[EXTERNAL_DOWNLOADS]: The skill interacts with external services to fetch assets and process data.
It downloads public domain images from Wikimedia Commons (commons.wikimedia.org) using scripts/fetch_images.py.
It optionally uses ByteDance's Volcengine cloud APIs (ark.cn-beijing.volces.com, openspeech.bytedance.com) for video review and text-to-speech services (scripts/cloud/). These are recognized well-known services and use the user's own API keys.

huashu-design