The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes transcripts and video metadata from external platforms like YouTube, creating a potential surface for indirect prompt injection where malicious instructions could be embedded in the source material to influence the AI's summarization or synthesis stages.
Ingestion points: The skill fetches data from YouTube URLs, playlists, and local video folders as defined in SKILL.md and digest.json.
Boundary markers: The instructions do not define specific delimiters or instructions to ignore embedded prompts within the ingested transcripts.
Capability inventory: The workflow involves writing markdown and JSON files to the local project structure and triggering publishing tools for the final site output.
Sanitization: A transcript cleaning step is mentioned, although specific prompt injection filtering is not described.
[COMMAND_EXECUTION]: The skill functions as a high-level orchestrator that invokes multiple sub-skills (e.g., video-digest-ingest, video-digest-summarize) and executes build scripts to generate the static site. This behavior is standard and expected for a project orchestration workflow.
[EXTERNAL_DOWNLOADS]: The skill automates the download of video and transcript content from YouTube and other provided sources. YouTube is a well-known service, and these downloads are essential to the primary functionality of generating digests.

video-digest