The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes system binaries such as ffmpeg and ffprobe for video processing. These executions are securely implemented using list-based arguments in subprocess.run(), which avoids shell interpretation and protects against command injection vulnerabilities.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill handles sensitive API keys via a local .env file and includes explicit instructions for the AI agent to manage these keys safely, including warnings about conversation log exposure. Video data is uploaded to the author's official service (upload-post.com) and Google Gemini, which are necessary for the skill's core functionality.
[INDIRECT_PROMPT_INJECTION]: The pipeline processes external video content and AI-generated hooks. Risks are mitigated by using distinct prompt sections for transcript data and by rendering text overlays to static images (Pillow) before video composition, ensuring that AI-generated content cannot be misinterpreted as executable commands by downstream tools.
[EXTERNAL_DOWNLOADS]: The skill relies on standard, well-known Python packages and model weights for Whisper and Gemini. All dependencies and API endpoints are consistent with the stated purpose of the tool and the author's own infrastructure.

autoshorts