The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions involve running shell commands to build projects (e.g., mvn verify, ./gradlew build installDist) and execute the resulting Java artifacts. Additionally, the provided test templates (McpServerAppTest.java) programmatically spawn the server as a subprocess using the java binary from the current java.home environment.
[DYNAMIC_EXECUTION]: The skill generates Java source code, project metadata files (pom.xml, build.gradle.kts), and client configurations from predefined templates. These components are designed to be compiled and executed on the host system as part of a standard development workflow.
[INDIRECT_PROMPT_INJECTION]: The generated MCP server creates a runtime interface where an AI agent can supply arguments to Java tool handlers, representing a surface for indirect prompt injection.
Ingestion points: External data enters the server through the arguments map in the CallToolRequest within the callHandler method in McpServerApp.java.
Boundary markers: The skill utilizes JSON Schema (inputSchema) to define and validate the structure of incoming data, which helps delimit inputs.
Capability inventory: The generated skill can execute arbitrary Java logic defined in tool handlers and can perform filesystem or network operations depending on the tools the user instructs the agent to create. Integration tests also demonstrate the capability to launch subprocesses.
Sanitization: The instructions highlight that the MCP SDK automatically validates tool arguments against the provided inputSchema, ensuring the handler only receives data that matches the expected format.

java-mcp-server-generator