The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses the Roslyn compiler to dynamically compile and execute C# code provided in the csharpCode parameter. This allows for the execution of arbitrary logic that is generated or provided at runtime.
[REMOTE_CODE_EXECUTION]: By design, the skill accepts raw source code as input and executes it. This is a high-risk capability that can be exploited for remote code execution (RCE) if an agent receives malicious code from an untrusted source and passes it to this skill.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The csharpCode input field in SKILL.md serves as the primary entry point for untrusted code.
Boundary markers: The skill documentation lacks any instructions or mechanisms to delimit user data from code or to prevent the agent from executing instructions embedded in external data.
Capability inventory: The execution environment provides full C# capabilities, including file system access, network communication, and process management, as demonstrated by the use of unity-mcp-cli in SKILL.md.
Sanitization: No sanitization, code validation, or sandboxing measures are described to restrict the actions of the executed C# code.
[COMMAND_EXECUTION]: The skill instructions require the execution of the unity-mcp-cli tool, which acts as the runner for the dynamically compiled payloads.
[EXTERNAL_DOWNLOADS]: Troubleshooting instructions suggest the installation of the unity-mcp-cli package from the npm registry, which is an external dependency required for the skill to function.

script-execute