The Agent Skills Directory

[COMMAND_EXECUTION]: The skill dynamically generates shell scripts (render.sh and merge.sh) and multiple Python files (chunk_*.py) based on user-provided topics.
[COMMAND_EXECUTION]: It utilizes chmod +x to grant execution permissions to these dynamically created shell scripts before executing them to run the rendering and merging process.
[COMMAND_EXECUTION]: The skill executes external command-line tools including manimgl, ffmpeg, and ffprobe to process content and generate video files.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted user input (the topic variable) which is interpolated into script filenames and generated Python code. While the skill includes a sanitization step for the TOPIC_SLUG (stripping special characters), it lacks explicit boundary markers or delimiters when interpolating the raw topic into the SCENE_PLAN.md and subsequent narrative planning phases. This ingestion point, combined with the capability to execute shell scripts and Python code via the manimgl and bash subprocess calls, creates a potential path for an attacker to influence the generated code's logic if the agent fails to maintain strict formatting constraints.

test-manim