The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Python and Shell scripts to invoke various command-line utilities. Scripts such as make_cover.py and edge_tts_engine.py call FFmpeg for audio/video processing and headless Chrome for rendering still images from HTML templates. These operations are restricted to the project workspace.
[EXTERNAL_DOWNLOADS]: The setup_env.sh script downloads Noto SC fonts from Google's official GitHub repository and clones the HyperFrames framework from HeyGen's repository. These downloads target well-known and trusted technology organizations.
[DYNAMIC_EXECUTION]: At setup time, the skill creates a Python virtual environment and installs necessary libraries including edge-tts, fonttools, and mlx-whisper. It also uses npx to execute specific versions of the HyperFrames CLI tool.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from GitHub Pull Requests (including diffs, bodies, and comments) and external Markdown articles to generate narration scripts.
Ingestion points: Data is fetched from public GitHub repositories using fetch-pr.mjs and local file reads.
Boundary markers: The skill lacks explicit delimiters or instructions for the agent to ignore potentially malicious embedded commands within the ingested text.
Capability inventory: The skill environment can execute shell commands (ffmpeg, chrome), write to the local file system, and perform network requests.
Sanitization: External content is interpolated into internal scripts without visible sanitization, creating a surface for content to influence the agent's narration generation.

sglang-explainer-video