The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill provides several tools that allow an AI agent to generate and execute code on the host machine within the Godot Engine environment.
The execute-script tool permits the execution of arbitrary GDScript expressions in the editor context.
The evaluate-runtime-expression tool allows for the execution of code within the context of a running game instance.
The modify-script and create-script tools allow the agent to write or overwrite GDScript files (.gd), which can then be executed by the engine.
[REMOTE_CODE_EXECUTION]: The manual installation instructions direct users to clone a repository from an external, unverified source (github.com/yurineko73/Godot-MCP-Native.git), which constitutes the execution of unvetted third-party code.
[COMMAND_EXECUTION]: The installation and configuration guide requires the use of shell commands that download and run external software, including git clone, npm install mcp-remote, and npx mcp-remote.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a high-risk attack surface for indirect prompt injection due to the combination of data-reading and code-execution capabilities.
Ingestion points: The agent reads untrusted data via read-script, analyze-script, and get-editor-logs (SKILL.md).
Boundary markers: There are no instructions or delimiters defined to prevent the agent from obeying malicious commands embedded as comments or strings within the files it analyzes.
Capability inventory: The agent has access to sensitive tools including execute-script, modify-script, and run-project (SKILL.md).
Sanitization: No sanitization or validation logic is described for handling script content before it is processed or executed.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download of external assets and plugins from the Godot Asset Library and GitHub without built-in integrity verification.

godot-mcp-native