The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script modules/external-source/scripts/read-segments.py by interpolating user-provided video URLs or local file paths. The Python script itself uses list-based subprocess.run calls to execute yt-dlp, ffmpeg, and ffprobe, which is a secure practice to prevent shell injection within the script's own operations.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources, including YouTube caption tracks and local subtitle files. These transcripts are used to score content and generate social media posts, creating a surface for indirect prompt injection.
Ingestion points: modules/external-source/scripts/read-segments.py reads YouTube captions and local subtitle files.
Boundary markers: The skill uses markdown blockquotes in its output templates (assets/clip-card.md) to delimit ingested transcript content.
Capability inventory: The skill uses subprocess.run for media processing and the Fountain API to create draft social posts.
Sanitization: A 'safety pass' is explicitly defined in modules/copy/MODULE.md to flag risks such as unverified claims or misleading content.
[EXTERNAL_DOWNLOADS]: The skill uses yt-dlp to fetch caption data from YouTube and utilizes web search tools to resolve media for podcast episodes when they are not available on the primary platform. These actions are directed by user input and are fundamental to the skill's purpose.

fountain-clip-finder