The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes external shell commands via the unity-mcp-cli tool to interact with a Unity project runtime. The instructions guide the agent to run commands such as unity-mcp-cli run-tool reflection-method-call --input ... which directly interfaces with the operating system's command line.
[REMOTE_CODE_EXECUTION]: The skill provides a mechanism for arbitrary code execution within the host application by leveraging C# reflection. It explicitly supports calling private methods and instantiating types based on string inputs (typeName, methodName), which bypasses standard language access controls and allows execution of any logic present in the loaded assemblies.
[INDIRECT_PROMPT_INJECTION]: The skill defines a large attack surface where untrusted data could influence code execution.
Ingestion points: The filter, targetObject, and inputParameters fields in SKILL.md are populated by the agent based on its current context.
Boundary markers: The skill does not implement or recommend any boundary markers or delimiters to separate user-provided data from method call instructions.
Capability inventory: The skill can invoke any method (public or private), including those in standard libraries (e.g., System.IO, System.Diagnostics) or internal application logic, providing high-impact capabilities.
Sanitization: There is no evidence of sanitization, allowlisting, or validation of the methods being targeted, allowing for potentially dangerous calls if the agent is manipulated by external content.

reflection-method-call