The Agent Skills Directory

[SAFE]: The skill provides templates for scaffolding Quarkus applications with LangChain4j. It correctly includes security best practices, such as a dedicated template for input and output guardrails (templates/Guardrails.java.template) to detect and mitigate prompt injection attacks.\n- [COMMAND_EXECUTION]: The application.properties.template includes a commented-out example for configuring an MCP client using the stdio transport, which executes a command (e.g., npm exec @modelcontextprotocol/server-everything). This is a standard and documented feature of the Model Context Protocol (MCP) for local tool integration and references a well-known community package.\n- [EXTERNAL_DOWNLOADS]: The skill references standard library dependencies from the dev.langchain4j and io.quarkiverse.langchain4j groups. These are well-known and official repositories for the Quarkus and LangChain4j ecosystem.\n- [PROMPT_INJECTION]: The skill templates facilitate the creation of AI services that ingest untrusted user data into prompts. \n
Ingestion points: templates/AiService.java.template (method parameters) and templates/Agent.java.template (WebSocket input in TriageSocket.java).\n
Boundary markers: Prompts in templates (e.g., Problem: {problem}) lack explicit delimiters or 'ignore' instructions by default.\n
Capability inventory: Generated applications can execute tools defined in templates/Tools.java.template and perform network operations via MCP clients in templates/McpClient.java.template.\n
Sanitization: The skill proactively mitigates this risk by providing templates/Guardrails.java.template, which demonstrates how to implement InputGuardrail and OutputGuardrail beans to filter malicious content like 'ignore previous instructions'.

quarkus-langchain4j-scaffolding