The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the eval command in SKILL.md to run the manim rendering tool. The arguments for this command, such as the scene class name, are extracted from a JSON file generated by an AI agent based on the user's prompt. This allows for potential shell command injection if the AI generates a class name containing metacharacters like semicolons or backticks.
[COMMAND_EXECUTION]: The core logic of the skill involves generating Python scripts from user prompts and executing them via the manim engine. Because manim executes these scripts as full Python code, any malicious logic injected via the user prompt (such as file system operations or network requests) will be executed with the user's system permissions.
[COMMAND_EXECUTION]: When detecting the Claude CLI for sub-agent tasks, the skill defaults to using the --dangerously-skip-permissions flag. This disables security prompts for the sub-agent, allowing it to execute potentially harmful commands or modify files without user confirmation, which significantly increases the risk if the sub-agent is compromised via prompt injection.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. A malicious prompt could be crafted to bypass the sub-agent's internal rules and force the generation of Python code that performs unauthorized actions when rendered by Manim.
[EXTERNAL_DOWNLOADS]: The skill instructs users to install external dependencies like manim, ffmpeg, and latex using standard package managers. While these are well-known tools, they represent an external dependency chain required for the skill's operation.

animate