The Agent Skills Directory

[COMMAND_EXECUTION]: Executes external tools like yt-dlp, uvx, or python3 via spawnSync to retrieve video information and subtitles.
Uses the --remote-components ejs:github flag, enabling yt-dlp to fetch and execute extractor updates from GitHub.
Supports the YOUTUBE_TRANSCRIPT_COOKIES_FROM_BROWSER environment variable, allowing the tool to access host browser cookies.
[COMMAND_EXECUTION]: Permits writing files to arbitrary system locations using the --output and --output-dir arguments.
[EXTERNAL_DOWNLOADS]: Retrieves data from YouTube and Google APIs and downloads images from i.ytimg.com.
[EXTERNAL_DOWNLOADS]: Uses uvx to dynamically download and execute the yt-dlp package if it is not locally available.
[PROMPT_INJECTION]: The skill is subject to indirect prompt injection (Category 8) as it feeds untrusted video metadata and transcript text into an AI sub-agent for speaker identification.
Ingestion points: Video title, description, and raw transcript content from meta.json and transcript-raw.json are processed by an AI sub-agent.
Boundary markers: The prompt template in prompts/speaker-transcript.md does not utilize delimiters or specific instructions to protect the sub-agent from malicious commands embedded in the data.
Capability inventory: The sub-agent is instructed to read the transcript file and overwrite it with the processed version.
Sanitization: There is no evidence of sanitization or filtering of the untrusted YouTube content before it is interpolated into the prompt.

baoyu-youtube-transcript