The Agent Skills Directory

[DYNAMIC_EXECUTION]: The execute_code tool (documented in references/tools-reference.md) allows for the execution of arbitrary C# method bodies directly within the Unity Editor. The documentation notes that while safety checks are enabled by default, the tool is not a sandbox and imports sensitive namespaces like System.Reflection.
[REMOTE_CODE_EXECUTION]: The manage_packages tool supports the installation of Unity packages using arbitrary Git URLs. This allows for the introduction of unverified third-party code into the development environment, a high-risk vector for supply chain attacks.
[DATA_EXFILTRATION]: Multiple resource URIs, such as mcpforunity://project/info and mcpforunity://editor/state, expose absolute local file system paths (e.g., projectRoot, assetsPath) to the AI agent. This metadata provides reconnaissance information that could be used in targeted attacks.
[INDIRECT_PROMPT_INJECTION]: The skill represents a significant attack surface because it ingests untrusted data from the Unity project environment—including GameObject names, component properties via mcpforunity://scene/gameobject, and console logs via read_console—and provides the agent with high-privilege tools to act on that data.
Ingestion points: read_console (console logs), mcpforunity://scene/gameobject (scene data), and manage_asset (search results).
Boundary markers: Absent. The instructions do not mandate the use of delimiters or specific safety instructions for processing data from these sources.
Capability inventory: create_script, execute_code, manage_packages, and manage_editor(action="deploy_package").
Sanitization: No evidence of sanitization or validation of data retrieved from the Unity Editor before it is used in subsequent tool calls.
[SAFE]: The unity_docs tool fetches documentation from the official docs.unity3d.com domain, which is an established and trusted service.

unity-mcp-orchestrator