The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill ingests untrusted data from external sources (YouTube transcripts and web pages) and processes it to generate summaries and quiz questions. There is a risk that malicious instructions embedded in a video's subtitles or a fetched web page could influence the agent's behavior.
Ingestion points: scripts/extract_transcript.sh (YouTube subtitles) and references/deep-research.md (WebFetch tool for web pages).
Boundary markers: None. The skill does not use delimiters or instructions to ignore embedded commands in the fetched data.
Capability inventory: The skill can execute shell scripts, perform web searches, and fetch web content.
Sanitization: None. The workflow includes correcting proper nouns but does not sanitize the input for potential prompt injection patterns.
[External Downloads / Binary Dependency] (LOW): The skill depends on yt-dlp, an external command-line utility. While yt-dlp is a standard tool for video metadata extraction, it must be pre-installed on the host system, and its behavior depends on the version installed.
[Command Execution] (LOW): The skill executes local shell scripts (scripts/extract_metadata.sh and scripts/extract_transcript.sh) using arguments provided by the user (YouTube URLs). While the scripts use proper quoting ("$URL") to mitigate simple command injection, the execution of shell scripts with external parameters remains a notable attack surface.

youtube-digest