The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the execution of several command-line utilities to function, specifically yt-dlp for fetching video metadata and streams, whisper for audio-to-text transcription, and magick (ImageMagick) for image processing and the creation of figure montages. These are legitimate uses within the scope of generating lecture notes.
[REMOTE_CODE_EXECUTION]: To provide accurate visualizations, the skill allows for the generation and execution of Python scripts using libraries such as matplotlib and seaborn. These scripts are used to create vector-based charts and diagrams from the data extracted from the video content, which is a primary feature for high-quality pedagogical output.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it processes external, untrusted content from Bilibili videos and incorporates it into a LaTeX document.
Ingestion points: External data enters the agent context through Bilibili video titles, metadata, and subtitle tracks (manual or AI-generated).
Boundary markers: No specific boundary markers or instructions to ignore embedded instructions are defined for the processed external text.
Capability inventory: The agent has the capability to write files (.tex, .pdf, .py, .srt), execute shell commands, and run generated Python scripts.
Sanitization: The skill lacks explicit instructions to sanitize or escape LaTeX-specific special characters (e.g., $, &, %, _) from the ingested content, which could lead to compilation failures or document manipulation if the source metadata is malicious.

bilibili-render-pdf