generate_retry_message

Instructions

Implementation Reference

• src/server.ts:218-229 (handler)

  The handler function that executes the 'generate_retry_message' tool logic. It creates a ToolArgError with the tool name, validation error, and attempted args, then calls toLLMFeedback() to build the LLM-facing retry message.

  function generateRetryMessageTool(input: { tool_name: string; validation_error: string; attempted_args: any }) {
    const err = new ToolArgError(input.tool_name, input.validation_error, input.attempted_args);
    const feedback = err.toLLMFeedback?.() ?? err.message;
    return {
      content: [
        {
          type: 'text',
          text: JSON.stringify({ retry_message: feedback }, null, 2),
        },
      ],
    };
  }

• src/server.ts:95-108 (schema)

  The input schema definition for 'generate_retry_message' tool. It defines three required properties: tool_name (string), validation_error (string), and attempted_args (object).

  {
    name: 'generate_retry_message',
    description:
      'Given a tool name, validation error, and attempted args, build the canonical LLM-facing retry feedback message. Uses agentvet\'s ToolArgError.toLLMFeedback() formatting so the wording matches what runtime callers see.',
    inputSchema: {
      type: 'object',
      properties: {
        tool_name: { type: 'string' },
        validation_error: { type: 'string' },
        attempted_args: { type: 'object' },
      },
      required: ['tool_name', 'validation_error', 'attempted_args'],
    },
  },

• src/server.ts:53-109 (registration)

  The tool is registered in the TOOLS array (line 95-108) which is served via ListToolsRequestSchema handler. The tool dispatch in the CallToolRequestSchema handler (line 125-126) routes the tool name to the handler function.

  const TOOLS = [
    {
      name: 'validate_tool_args',
      description:
        'Validate a tool-call args object against a small shape spec. Returns { valid, error?, retry_hint? } where retry_hint is a ready-to-send LLM feedback message describing exactly what was wrong.',
      inputSchema: {
        type: 'object',
        properties: {
          tool_name: {
            type: 'string',
            description: 'Name of the tool being called (surfaces in retry_hint).',
          },
          args: {
            type: 'object',
            description: 'The args object the LLM wants to pass.',
          },
          shape: SHAPE_SCHEMA,
        },
        required: ['tool_name', 'args', 'shape'],
      },
    },
    {
      name: 'lint_tool_definition',
      description:
        'Sanity-check a tool definition for common mistakes that hurt LLM tool-use accuracy: missing description, vague description, no required fields, schema fields without descriptions, non-snake_case names.',
      inputSchema: {
        type: 'object',
        properties: {
          tool: {
            type: 'object',
            description: 'A tool definition: { name, description, inputSchema }.',
            properties: {
              name: { type: 'string' },
              description: { type: 'string' },
              inputSchema: { type: 'object' },
            },
            required: ['name'],
          },
        },
        required: ['tool'],
      },
    },
    {
      name: 'generate_retry_message',
      description:
        'Given a tool name, validation error, and attempted args, build the canonical LLM-facing retry feedback message. Uses agentvet\'s ToolArgError.toLLMFeedback() formatting so the wording matches what runtime callers see.',
      inputSchema: {
        type: 'object',
        properties: {
          tool_name: { type: 'string' },
          validation_error: { type: 'string' },
          attempted_args: { type: 'object' },
        },
        required: ['tool_name', 'validation_error', 'attempted_args'],
      },
    },
  ] as const;

• src/server.ts:125-126 (registration)

  Dispatch case in CallToolRequestSchema handler that routes 'generate_retry_message' to the generateRetryMessageTool function.

  case 'generate_retry_message':
    return generateRetryMessageTool(args as { tool_name: string; validation_error: string; attempted_args: any });