The Agent Skills Directory

[DATA_EXFILTRATION]: Optional cloud-based features in scripts/cloud/ai-review-video.py and scripts/cloud/tts-doubao.mjs transmit user-supplied text or rendered video content to ByteDance/Volcengine APIs for processing. These features require personal API keys and include an explicit consent mechanism via the --yes flag.
[COMMAND_EXECUTION]: The skill uses subprocesses to execute local media utilities such as ffmpeg, ffprobe, and playwright. It also contains instructions for the agent to run git commands locally to facilitate update checks and version tracking within its directory.
[DYNAMIC_EXECUTION]: Scripts like render-video.js and render-video-seek.js use Playwright to dynamically inject CSS and JavaScript into browser instances to manage UI state during rendering. Additionally, shell scripts use Python's eval() to calculate values derived from media file metadata.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input and images fetched from Wikimedia to generate HTML-based designs and media files, presenting a vulnerability surface for indirect prompt injection.
Ingestion points: User input text used for design generation and external images retrieved from Wikimedia Commons via scripts/fetch_images.py.
Boundary markers: Not clearly implemented for text interpolation in design templates.
Capability inventory: Subprocess execution (ffmpeg, playwright), file system writing, and network operations to cloud APIs.
Sanitization: No explicit sanitization or validation of user-provided strings before interpolation into HTML/JS templates was identified.
[EXTERNAL_DOWNLOADS]: The skill programmatically fetches images from Wikimedia Commons for use in content generation. It also recommends using npx for certain initialization tasks, which triggers the download of external packages from the npm registry.

huashu-design