The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The download script (scripts/download-video.sh) utilizes the --cookies-from-browser chrome flag when retrying failed TikTok downloads. This action causes the tool to read from the user's local browser cookie database to authenticate requests, which is a sensitive operation involving the access of local session data.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied URLs, creating an attack surface for indirect prompt injection if the fetched metadata or the sites visited via browser automation contain malicious instructions.
Ingestion points: The scripts download-video.sh and download-fallback.sh take a URL as their primary input from the user/agent.
Boundary markers: The scripts wrap the URL variable in quotes when passing it to tools, but there are no explicit boundary markers or instructions to the LLM to ignore content within the processed URLs.
Capability inventory: The skill possesses network access (curl, yt-dlp), file system write access, and browser automation capabilities (agent-browser) including JavaScript evaluation.
Sanitization: The skill relies on the input handling of the underlying command-line utilities and does not perform independent sanitization of the URL beyond basic shell quoting.
[EXTERNAL_DOWNLOADS]: The fallback mechanism (scripts/download-fallback.sh) automates interaction with the third-party service https://greenvideo.cc/en/. It extracts video URLs from the site's runtime environment and downloads them using curl.
[COMMAND_EXECUTION]: The skill executes various shell scripts and command-line tools, including yt-dlp, agent-browser, and curl, to perform video extraction and downloading.

video-download