The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python CLI script (cs.py) using shell commands that interpolate the current working directory ($(pwd)) and C# code snippets. Examples are found in SKILL.md usage sections.
[REMOTE_CODE_EXECUTION]: The skill's primary function is to facilitate the execution of arbitrary C# code in a running Unity Editor process. This provides full access to the Unity API and underlying system resources accessible via C#.
[DYNAMIC_EXECUTION]: The tool utilizes a Roslyn-based REPL to compile and execute C# code at runtime. It supports top-level syntax, private member access bypass, and persistent state across submissions, which increases the potential impact of malicious snippets.
[INDIRECT_PROMPT_INJECTION]: There is a significant attack surface for indirect prompt injection. If the agent processes untrusted data (e.g., from a file or external API) and is prompted to 'run' or 'eval' that content as C#, it could lead to unauthorized code execution within the Unity environment.
[DATA_EXPOSURE]: The skill facilitates access to sensitive project data via the AssetDatabase and live scene queries (e.g., FindObjectsByType), which could be used to harvest information about the development environment.

unity-cli-exec-code