The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes uvx yt-dlp to fetch transcripts from video URLs. This is a standard and appropriate use of a well-known utility for the skill's primary purpose. It also suggests brew install deno as a troubleshooting step for the user environment, which is a common developer practice.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content from URLs and transcripts, which creates a potential surface for indirect prompt injection. However, the risk is significantly mitigated by the skill's internal logic:
Ingestion points: Content is retrieved from external URLs using WebFetch and video transcripts via yt-dlp in SKILL.md.
Boundary markers: While formal delimiters are not specified, the skill mandates a classification step and human review.
Capability inventory: The skill has file system access to ~/Obsidian/ and uses subprocesses for transcript extraction.
Sanitization: The skill explicitly defines a 'Content-class gate' to prevent writing sensitive data (Class A) and a 'Propose, then judge' step that requires user approval for conceptual updates, effectively acting as a human-in-the-loop security control.

ingest