The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several external command-line utilities to process video files.
scripts/vd_probe.py uses subprocess.run to call yt-dlp and ffprobe for metadata extraction.
scripts/vd_ingest.sh executes yt-dlp for video downloads, ffmpeg for audio conversion, and uvx for transcription.
scripts/vd_single.py uses ffprobe to determine video duration.
[EXTERNAL_DOWNLOADS]: The skill fetches external content and software components during operation.
scripts/vd_ingest.sh uses yt-dlp to download video content from provided URLs.
scripts/vd_ingest.sh uses uvx --from mlx-whisper mlx_whisper to download and execute the Whisper transcription tool from the Python Package Index (PyPI).
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources, creating a vulnerability surface.
Ingestion points: scripts/vd_probe.py reads YouTube titles and metadata; scripts/vd_ingest.sh generates transcripts from video audio.
Boundary markers: The generated output files (transcript.tsv, timeline_30s.txt) do not include explicit boundary markers or instructions for downstream LLMs to ignore embedded commands.
Capability inventory: The skill has the ability to execute shell commands, perform network downloads, and write to the local file system.
Sanitization: Basic shell variable quoting is used in the bash scripts, and Python subprocess calls use list-based arguments to mitigate direct command injection.

video-digest-ingest