Claude Code Prompt Engineer

Instructions

Implementation Reference

• index.ts:583-623 (handler)

  Main handler for the engineer_prompt tool in the CallToolRequestSchema. Validates arguments, handles interactive and direct modes, manages sessions, and returns optimized prompts.

  case "engineer_prompt": {
    if (!isEngineerPromptArgs(args)) {
      throw new Error("Invalid arguments for engineer_prompt");
    }
    
    const { prompt, language, context, interactive } = args;
    
    if (interactive) {
      // Start interactive session
      const sessionId = generateSessionId();
      const detectedInfo = await detectLanguageAndTaskType(prompt);
      const questions = await generateClarifyingQuestions(prompt, language || detectedInfo.language, context);
      
      activeSessions.set(sessionId, {
        originalPrompt: prompt,
        context: context ? [context] : [],
        refinements: [],
        language: language || detectedInfo.language,
        complexity: detectedInfo.complexity as any,
        taskType: detectedInfo.taskType as any
      });
      
      return {
        content: [{ 
          type: "text", 
          text: `Interactive prompt engineering session started (ID: ${sessionId}).\n\nTo help create the best prompt for Claude Code, please answer these questions:\n\n${questions.map((q, i) => `${i + 1}. ${q}`).join('\n')}\n\nUse the answer_questions tool with session ID "${sessionId}" to provide your answers.`
        }],
        isError: false,
      };
    } else {
      // Direct prompt engineering
      const engineeredPrompt = await engineerPrompt(prompt, language, context);
      return {
        content: [{ 
          type: "text", 
          text: `**Optimized Prompt for Claude Code:**\n\n${engineeredPrompt}\n\n**Are you ready to proceed with this task?**`
        }],
        isError: false,
      };
    }
  }

• index.ts:27-53 (schema)

  Tool definition including name, description, and input schema for engineer_prompt.

  const ENGINEER_PROMPT_TOOL: Tool = {
    name: "engineer_prompt",
    description: "Intelligently engineers and optimizes prompts for Claude Code, with interactive refinement and automatic optimization for maximum effectiveness.",
    inputSchema: {
      type: "object",
      properties: {
        prompt: {
          type: "string",
          description: "The raw user prompt that needs engineering"
        },
        language: {
          type: "string",
          description: "The programming language (optional, will be detected if not provided)"
        },
        context: {
          type: "string", 
          description: "Additional context about the codebase or project (optional)"
        },
        interactive: {
          type: "boolean",
          description: "Whether to enable interactive Q&A refinement (default: false)"
        }
      },
      required: ["prompt"],
      title: "engineer_promptArguments"  
    }
  };

• index.ts:570-572 (registration)

  Registers the engineer_prompt tool by including ENGINEER_PROMPT_TOOL in the ListTools response.

  server.setRequestHandler(ListToolsRequestSchema, async () => ({
    tools: [ENGINEER_PROMPT_TOOL, ASK_CLARIFICATION_TOOL, ANSWER_QUESTIONS_TOOL, AUTO_OPTIMIZE_TOOL],  
  }));

• index.ts:462-470 (helper)

  Internal helper function that performs the core prompt engineering by delegating to generateOptimizedPrompt.

  async function engineerPrompt(
    prompt: string, 
    language?: string, 
    context?: string, 
    refinements: string[] = []
  ): Promise<string> {
    // Always use built-in optimization - no external API calls
    return generateOptimizedPrompt(prompt, language, context, refinements);
  }

• index.ts:137-149 (schema)

  Type guard function for validating input arguments to the engineer_prompt tool.

  function isEngineerPromptArgs(args: unknown): args is { 
    prompt: string; 
    language?: string;
    context?: string;
    interactive?: boolean;
  } {
    return (
      typeof args === "object" &&
      args !== null &&
      "prompt" in args &&
      typeof (args as { prompt: string }).prompt === "string"
    );
  }