The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/qc.py utilizes the eval() function to parse the r_frame_rate field obtained from ffprobe output. This is a dangerous practice as it could allow for arbitrary code execution if a maliciously crafted video file causes ffprobe to return a non-numeric string.
[EXTERNAL_DOWNLOADS]: The scripts/vendor.py utility enables the AI agent to download and extract any package from the NPM registry using npm pack. This allows for the introduction of unverified third-party code into the project environment during the build process.
[INDIRECT_PROMPT_INJECTION]: The skill implements a pipeline that takes a natural language brief and generates executable JavaScript code in composition.html. This code is subsequently executed in a Chromium browser instance via Playwright. An adversarial brief could trick the agent into writing malicious scripts that might attempt to exfiltrate information from the local system or the local server used during rendering.
[COMMAND_EXECUTION]: Numerous scripts throughout the skill (including render.py, audio.py, qc.py, and vendor.py) make extensive use of the subprocess module to execute external binaries like ffmpeg and ffprobe. While these calls do not use shell=True, the complexity of the parameters increases the risk of command injection if project configurations or file paths are manipulated.

motion-studio