The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes external CLI tools including yt-dlp, ffmpeg, and agy using the Python subprocess module. These calls are used to download video data, extract frames, and perform vision-based analysis, which are core components of the skill's functionality.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted transcript data fetched from YouTube videos.
Ingestion points: Transcripts are downloaded into the workspace as raw_transcript.json by scripts/anibon/ytdlp.py and then loaded for keyword scanning in scripts/_vision.py.
Boundary markers: The skill lacks explicit boundary markers or instructions to ignore embedded commands within the processed transcript text during its scanning and annotation phases.
Capability inventory: The skill possesses file-write capabilities (extracted frames and updated metadata files) and command execution capabilities (ffmpeg, yt-dlp, agy).
Sanitization: While the skill includes utilities for cleaning garbled ASR text in scripts/anibon/cleaner.py, it does not perform sanitization for potential prompt injection instructions embedded in transcripts.
[EXTERNAL_DOWNLOADS]: The skill fetches media content and metadata from YouTube. As YouTube is a well-known service, these downloads are documented as intended functionality.
[DYNAMIC_EXECUTION]: The skill uses a dynamic import of the sys module in scripts/anibon/markdown.py to facilitate writing to the standard error stream. This is a common and benign implementation for localized logging purposes.

antigravity-vision-proxy