The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from external URLs, YouTube transcripts, and PDF files. Malicious instructions embedded within these sources (e.g., in a YouTube auto-generated transcript or a web article) could potentially influence the agent's behavior during the classification and writing phases.
Ingestion points: Data enters the context via WebFetch for URLs, uvx yt-dlp for video transcripts, and Read for PDF files.
Boundary markers: The instructions do not specify any delimiters or safety markers to isolate the ingested content from the system instructions.
Capability inventory: The skill has the ability to execute shell commands (uvx, brew) and write files to the local filesystem (Obsidian vault).
Sanitization: While the skill mentions a 'Content-class gate' to filter out credentials and PII (Class A data), it does not include measures to sanitize the content against embedded malicious instructions.
[COMMAND_EXECUTION]: The skill uses uvx yt-dlp to retrieve video transcripts and suggests brew install deno if a JavaScript runtime is missing. These commands are limited to the stated purpose of the skill but involve direct shell interaction.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch content from external URLs and YouTube services as part of its primary function.

ingest