The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill instructs the agent to execute commands with elevated privileges using sudo to install system packages.
Evidence: The command sudo apt update && sudo apt install -y yt-dlp is provided for Linux environments.
[COMMAND_EXECUTION]: The skill executes multiple command-line utilities (yt-dlp, whisper, python3) using user-provided URLs as input, which can be risky if not properly handled by the agent platform.
Evidence: The skill defines a complete workflow involving shell scripts and command substitutions like VIDEO_TITLE=$(yt-dlp --print "%(title)s" "$VIDEO_URL" | tr '/' '_' ...). There is a risk that maliciously crafted video titles containing shell metacharacters not covered by the tr filter (e.g., backticks, dollar signs) could cause issues during filename expansion.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from an external source (YouTube) and processes it within the agent's context.
Ingestion points: Video titles and subtitle/transcript content fetched via yt-dlp and whisper (found in SKILL.md).
Boundary markers: No delimiters or specific "ignore instructions" warnings are used when processing the external transcript text.
Capability inventory: The skill has access to Bash, Read, and Write tools, allowing it to execute shell commands and modify local files.
Sanitization: The skill uses basic character replacement (tr) for filenames and regex-based cleaning for subtitle formatting, but does not sanitize the transcript content itself before it is presented to the user or agent.
[DYNAMIC_EXECUTION]: The skill uses Python one-liners to dynamically process data at runtime.
Evidence: A python3 -c script is used to deduplicate and clean VTT subtitle files during the post-processing phase.

youtube-transcript