The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use sudo apt install -y yt-dlp in SKILL.md, which is an attempt to acquire elevated privileges on the host system.
[DATA_EXFILTRATION]: In references/error-handling.md, the skill recommends using the --no-check-certificate flag. This disables SSL/TLS certificate verification, exposing the agent to Man-in-the-Middle (MitM) attacks and potentially compromising data integrity or leading to credential exposure in other contexts.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to YouTube to fetch video metadata and subtitle files. It also initiates package installations via Homebrew, apt, and pip.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests and processes untrusted external content (YouTube transcripts).
Ingestion points: Video transcripts are downloaded from YouTube and saved locally as .vtt then converted to .txt files.
Boundary markers: Absent; the skill does not wrap the processed transcript in delimiters or instruct the agent to ignore instructions within the text.
Capability inventory: The skill executes subprocesses for yt-dlp and runs a local Python script (scripts/vtt-to-txt.py).
Sanitization: The scripts/vtt-to-txt.py file performs basic regex cleaning to remove VTT tags and HTML-like sequences, but it does not sanitize the text content against prompt injection instructions that may be interpreted by the LLM when the user reviews the file.

youtube-transcript