The Agent Skills Directory

[COMMAND_EXECUTION]: The skill builds shell commands by directly inserting external URLs (e.g., VIDEO_URL, EPISODE_URL) into tool executions for yt-dlp and curl. Without rigorous sanitization, this allows an attacker to execute arbitrary commands by providing a specially crafted URL containing shell metacharacters.
[DATA_EXFILTRATION]: The instructions include the use of the --cookies-from-browser chrome flag with yt-dlp. This action enables the tool to read and potentially expose the user's local browser cookies, which are sensitive credentials, for the purpose of bypassing anti-bot measures.
[EXTERNAL_DOWNLOADS]: The skill performs multiple outbound network connections to fetch media files from various hosting platforms and to communicate with the Groq API for audio transcription services.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its ingestion of untrusted external media content. Ingestion points: External media URLs and extracted transcripts in SKILL.md. Boundary markers: No delimiters or instructions to ignore embedded commands are present in the final summary prompt. Capability inventory: yt-dlp, ffmpeg, curl, and shell execution. Sanitization: No evidence of input validation or sanitization for the data processed from external sources is present.

video