The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's scripts, specifically video-to-atlas.mjs and frame-atlas-core.mjs, execute external binaries (ffmpeg and ffprobe) using node:child_process.spawn. While this implementation avoids shell interpolation, the paths for these executables are configurable via CLI arguments (e.g., --ffmpeg, --ffprobe), which represents a minor attack surface if an agent is influenced to use unintended binaries.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided media files (videos, images) and their metadata, which are subsequently used by a sub-agent (gpt-6.1-sol) to generate animation prompts. This workflow creates a vulnerability surface for indirect prompt injection if asset metadata or filenames contain malicious instructions meant to influence the sub-agent's output.
[DYNAMIC_EXECUTION]: The utility script frame-atlas-core.mjs generates an HTML preview file containing a client-side JavaScript player using a template. Additionally, the test suite video-to-atlas.test.mjs uses node:vm's runInNewContext to execute scripts in a sandboxed environment for testing purposes. These are localized to specific utility and testing functions and do not facilitate arbitrary code execution during primary runtime operations.

phaser4-frame-animation