The Agent Skills Directory

[COMMAND_EXECUTION]: Utilizes command-line tools such as ffmpeg, yt-dlp, and you-get to download video streams, extract audio, and capture keyframes. These operations are performed via subprocess calls as part of the core video processing workflow.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external video sources, including transcriptions and subtitles, which are then used to generate summaries and slide content. This creates a potential surface for instructions embedded within the video content to influence the agent's behavior.
Ingestion points: External video URLs, audio-to-text transcriptions, and remote subtitle files.
Boundary markers: None explicitly defined in the summarization prompts.
Capability inventory: Local file system writing (output/ directory), subprocess command execution (ffmpeg, yt-dlp), and network access via browser automation (playwright).
Sanitization: No specific sanitization or filtering of the transcribed text is described before it is processed by the LLM.
[EXTERNAL_DOWNLOADS]: Fetches content from major video hosting platforms like YouTube, Bilibili, and Coursera. The use of established tools for these well-known services aligns with the skill's stated goal of video digestion.

video-digest