The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill transcribes audio from user-provided video and audio files into text. If the agent subsequently processes this transcript, it could be exposed to instructions embedded within the media content.
Ingestion points: Video and audio files processed by ./scripts/generate_srt.py.
Boundary markers: None identified in the skill instructions.
Capability inventory: The skill utilizes an external Python script and interactions with system tools like ffmpeg to generate and burn subtitles.
Sanitization: No sanitization or filtering of the transcribed text is mentioned.
[DYNAMIC_EXECUTION]: The skill uses the uv package manager to automatically install dependencies at runtime. Additionally, it downloads large transcription models (approximately 3GB each) from ivrit.ai and OpenAI during the first run. While typical for AI tasks, this involves dynamic execution of code and data downloads.
[EXTERNAL_DOWNLOADS]: The skill documentation explicitly states that models and dependencies are auto-downloaded on first use.
[NO_CODE]: The analysis is based solely on the documentation in SKILL.md as the referenced Python script ./scripts/generate_srt.py was not provided for auditing.

video-subtitles