The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts use subprocess.run to invoke ffmpeg, whisper-cli, and npx. All calls use lists for arguments rather than shell strings, which effectively mitigates command injection vulnerabilities. User-provided input like video titles or hook text is sanitized via slugification or character escaping before being passed to filters.
[EXTERNAL_DOWNLOADS]: The skill facilitates downloading video content from external URLs via a helper script. It also utilizes npx to execute the Remotion rendering engine, which may involve downloading Node.js packages from the official npm registry at runtime if they are not already cached.
[PROMPT_INJECTION]: The skill processes untrusted audio content into transcripts. It includes a robust safety warning in SKILL.md (Category 8) instructing the agent to treat all generated captions and transcripts as data for analysis rather than authoritative instructions.
Ingestion points: Video audio files and derived transcripts generated by Whisper.
Boundary markers: The skill includes a block-quoted warning in the frontmatter and body of SKILL.md.
Capability inventory: Multiple scripts utilize subprocess calls for video manipulation and transcription.
Sanitization: Employs slugification for directory names and character escaping for FFmpeg drawtext filters.

video-transcriber