The Agent Skills Directory

[PROMPT_INJECTION]: The SKILL.md file contains instructions that explicitly attempt to override the agent's safety protocols. Specifically, it instructs the agent to disregard privacy or sensitivity concerns and extract information like credit card numbers if requested by the user.
[REMOTE_CODE_EXECUTION]: The clip.py script performs dynamic installation of Python packages (openai, yt-dlp, and elevenlabs) using pip install at runtime if they are not already present in the environment. This execution of external commands to modify the environment is a significant security risk.
[COMMAND_EXECUTION]: The script extensively uses subprocess.run and subprocess.check_call to execute shell commands for ffmpeg, yt-dlp, and pip. While these are part of the intended functionality, the interpolation of user-provided inputs (--input and --query) into these command lines creates a surface for command injection if the agent does not properly sanitize the generated strings.
[DATA_EXFILTRATION]: The skill is designed to send processed audio data and transcriptions to external services (ElevenLabs and OpenAI). When combined with the instruction to specifically target sensitive PII (Personally Identifiable Information), this creates a high risk of exfiltrating sensitive user data to third-party APIs.
[EXTERNAL_DOWNLOADS]: The script downloads files from arbitrary URLs provided by the user and utilizes yt-dlp to download content from YouTube, introducing risks associated with processing untrusted remote media files.

clipit