The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several standard shell utilities to perform its tasks:
curl is used to communicate with the DeepAPI endpoint at https://deepapi.co for transcript scraping and status checks.
yt-dlp is employed to extract video metadata (channel and title) and as a fallback mechanism for subtitle downloading.
jq is used for parsing JSON responses from the API.
uuidgen is used to generate idempotency keys for API requests.
Shell built-ins and standard tools like tr are used for string manipulation and filename sanitization.
[DATA_EXFILTRATION]: The skill sends YouTube URLs and an API key to https://deepapi.co. As this domain is associated with the skill's functionality and the author's stated infrastructure, this is considered intended behavior for the primary scraping path.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data including YouTube video metadata and transcript content. It implements a security best practice by sanitizing filenames derived from external metadata using tr -cd '[:alnum:]_.-' to prevent potential command injection or filesystem traversal through malicious video titles.
Ingestion points: Metadata and transcript text extracted via yt-dlp and DeepAPI.
Boundary markers: Not explicitly present in instructions for handling the transcript body, but the content is directed to file storage.
Capability inventory: Network access (curl), command execution (yt-dlp), and file writing.
Sanitization: Strict character filtering for filenames derived from untrusted metadata.
[DYNAMIC_EXECUTION]: The skill uses a Python heredoc to process json3 files. The script is a static template defined within the skill file and is used specifically for extracting and cleaning transcript segments, which is a low-risk use of dynamic execution for data transformation.

youtube-transcript