create-agentmark 0.9.0 → 0.10.1

package/dist/index.js

@@ -17,7 +17,6 @@ var createAdapterConfig = (provider) => {
       dependencies: ["ai@^5", `@ai-sdk/${provider}@^2`],
       classes: {
         modelRegistry: "VercelAIModelRegistry",
-        toolRegistry: "VercelAIToolRegistry",
         webhookHandler: "VercelAdapterWebhookHandler"
       }
     },
@@ -26,11 +25,11 @@ var createAdapterConfig = (provider) => {
       dependencies: [
         "@mastra/core@<0.20.0",
         "@mastra/mcp@<0.13.4",
-        `@ai-sdk/${provider}@<2`
+        `@ai-sdk/${provider}@<2`,
+        "ai@^4"
       ],
       classes: {
         modelRegistry: "MastraModelRegistry",
-        toolRegistry: "MastraToolRegistry",
         webhookHandler: "MastraAdapterWebhookHandler"
       }
     },
@@ -39,7 +38,6 @@ var createAdapterConfig = (provider) => {
       dependencies: ["@anthropic-ai/claude-agent-sdk@^0.1.0"],
       classes: {
         modelRegistry: "ClaudeAgentModelRegistry",
-        toolRegistry: "ClaudeAgentToolRegistry",
         webhookHandler: "ClaudeAgentWebhookHandler"
       }
     }
@@ -229,16 +227,22 @@ main();
 };
 
 // src/utils/examples/templates/env.ts
-var getEnvFileContent = (_modelProvider, apiKey = "", adapter = "ai-sdk") => {
+var getEnvFileContent = (_modelProvider, apiKey = "", adapter = "ai-sdk", deploymentMode = "cloud") => {
   const apiKeyValue = apiKey || "your_api_key_here";
   const apiKeyName = adapter === "claude-agent-sdk" ? "ANTHROPIC_API_KEY" : "OPENAI_API_KEY";
-  return `# Cloud deployment: Set these environment variables
+  const cloudEnvVars = deploymentMode === "cloud" ? `
+# AgentMark Cloud \u2014 required for managed deployments
+AGENTMARK_API_KEY=your_agentmark_api_key
+AGENTMARK_APP_ID=your_agentmark_app_id
+` : `
+# Cloud deployment: Set these environment variables
 # AGENTMARK_BASE_URL=https://api.agentmark.co
 # AGENTMARK_API_KEY=your_agentmark_api_key
 # AGENTMARK_APP_ID=your_agentmark_app_id
+`;
+  return `${apiKeyName}=${apiKeyValue}
+${cloudEnvVars}
 # Learn more: https://docs.agentmark.co/platform/getting_started/quickstart
-
-${apiKeyName}=${apiKeyValue}
 `;
 };
 
@@ -594,15 +598,7 @@ text_config:
   model_name: ${model}
   max_calls: 2
   tools:
-    search_knowledgebase:
-      description: Search the company knowledgebase for information about shipping, warranty, and returns policies.
-      parameters:
-        type: object
-        properties:
-          query:
-            type: string
-            description: The search query to find relevant information
-        required: [query]
+    - search_knowledgebase
 test_settings:
   dataset: customer-query.jsonl
   props:
@@ -752,8 +748,9 @@ var createExamplePrompts = (model, targetPath = ".", adapter = "ai-sdk") => {
 // src/utils/examples/templates/user-client-config.ts
 var getClientConfigContent = (options) => {
   const { provider, adapter, deploymentMode = "cloud" } = options;
+  const isMastra = adapter === "mastra";
   const adapterConfig = getAdapterConfig(adapter, provider);
-  const { modelRegistry, toolRegistry } = adapterConfig.classes;
+  const { modelRegistry } = adapterConfig.classes;
   const isClaudeAgentSdk = adapter === "claude-agent-sdk";
   const providerImport = isClaudeAgentSdk ? "" : `import { ${provider} } from '@ai-sdk/${provider}';`;
   const loaderImport = deploymentMode === "cloud" ? `import { ApiLoader } from "@agentmark-ai/loader-api";` : `import { ApiLoader } from "@agentmark-ai/loader-api";
@@ -812,64 +809,98 @@ const adapterOptions = {
   // allowedTools: ['Read', 'Write', 'Glob'],
   // disallowedTools: ['Bash'],
 };` : "";
-  const createClientCall = isClaudeAgentSdk ? `return createAgentMarkClient<AgentMarkTypes, typeof toolRegistry>({ loader, modelRegistry, toolRegistry, evalRegistry, adapterOptions });` : `return createAgentMarkClient<AgentMarkTypes, typeof toolRegistry>({ loader, modelRegistry, toolRegistry, evalRegistry });`;
+  const toolImport = isClaudeAgentSdk ? `import { tool, createSdkMcpServer } from "@anthropic-ai/claude-agent-sdk";
+import { z } from 'zod';` : isMastra ? `import { tool } from 'ai';
+import type { ToolsInput } from '@mastra/core/agent';
+import { z } from 'zod';` : `import { tool } from 'ai';
+import type { Tool } from 'ai';
+import { z } from 'zod';`;
+  const createClientCall = isClaudeAgentSdk ? `return createAgentMarkClient<AgentMarkTypes>({ loader, modelRegistry, evalRegistry, adapterOptions, mcpServers: { 'customer-support': customerSupportTools } });` : `return createAgentMarkClient<AgentMarkTypes>({ loader, modelRegistry, tools, evalRegistry });`;
+  const toolSchemaField = isMastra ? `parameters: z.object({ query: z.string().describe('The search query') })` : `inputSchema: z.object({ query: z.string().describe('The search query') })`;
+  const toolsReturnType = isMastra ? "ToolsInput" : "Record<string, Tool>";
+  const toolsSetup = isClaudeAgentSdk ? `
+// Custom tools exposed as an MCP server \u2014 the SDK's native tool mechanism.
+// The server name is used in mcpServers config; tool names are used in prompt files.
+const knowledgeBase = tool(
+  'search_knowledgebase',
+  'Search the knowledge base for relevant articles',
+  { query: z.string().describe('The search query') },
+  async ({ query }) => ({
+    content: [{ type: 'text' as const, text: JSON.stringify({
+      articles: [
+        { topic: 'shipping', content: 'Standard shipping takes 3\u20135 business days.' },
+        { topic: 'warranty', content: 'All products include a 1-year limited warranty.' },
+        { topic: 'returns', content: 'You can return items within 30 days of delivery.' },
+      ],
+    }) }],
+  })
+);
+
+const customerSupportTools = createSdkMcpServer({
+  name: 'customer-support',
+  tools: [knowledgeBase],
+});` : `
+function createTools(): ${toolsReturnType} {
+  return {
+    search_knowledgebase: tool({
+      description: 'Search the knowledge base for relevant articles',
+      ${toolSchemaField},
+      execute: async ({ query }) => {
+        // Simulate search delay
+        await new Promise(resolve => setTimeout(resolve, 500));
+
+        // Return all three knowledge base articles
+        // The LLM will select the relevant one based on the query
+        return {
+          articles: [
+            { topic: 'shipping', content: 'Standard shipping takes 3\u20135 business days.' },
+            { topic: 'warranty', content: 'All products include a 1-year limited warranty.' },
+            { topic: 'returns', content: 'You can return items within 30 days of delivery.' }
+          ]
+        };
+      },
+    }),
+  };
+}`;
+  const toolsVariable = isClaudeAgentSdk ? "" : `  const tools = createTools();`;
   return `// agentmark.client.ts
 import path from 'node:path';
 import dotenv from 'dotenv';
 dotenv.config({ path: path.resolve(__dirname, '.env') });
-import { createAgentMarkClient, ${modelRegistry}, ${toolRegistry}, EvalRegistry } from "${adapterConfig.package}";
+import { createAgentMarkClient, ${modelRegistry} } from "${adapterConfig.package}";
+import type { EvalRegistry } from "${adapterConfig.package}";
 ${loaderImport}
-import AgentMarkTypes, { Tools } from './agentmark.types';
+import AgentMarkTypes from './agentmark.types';
 ${providerImport}
+${toolImport}
 ${adapterOptionsImport}
 
 ${modelRegistrySetup}
+${toolsSetup}
 
-function createToolRegistry() {
-  const toolRegistry = new ${toolRegistry}<Tools>()
-    .register('search_knowledgebase', async ({ query }) => {
-      // Simulate search delay
-      await new Promise(resolve => setTimeout(resolve, 500));
-
-      // Return all three knowledge base articles
-      // The LLM will select the relevant one based on the query
+const evalRegistry: EvalRegistry = {
+  exact_match_json: ({ output, expectedOutput }) => {
+    if (!expectedOutput) {
+      return { score: 0, label: 'error', reason: 'No expected output provided', passed: false };
+    }
+    try {
+      const ok = JSON.stringify(output) === JSON.stringify(JSON.parse(expectedOutput));
       return {
-        articles: [
-          { topic: 'shipping', content: 'Standard shipping takes 3\u20135 business days.' },
-          { topic: 'warranty', content: 'All products include a 1-year limited warranty.' },
-          { topic: 'returns', content: 'You can return items within 30 days of delivery.' }
-        ]
+        score: ok ? 1 : 0,
+        label: ok ? 'correct' : 'incorrect',
+        reason: ok ? 'Exact match' : 'Mismatch',
+        passed: ok
       };
-    });
-  return toolRegistry;
-}
-
-function createEvalRegistry() {
-  const evalRegistry = new EvalRegistry()
-    .register('exact_match_json', ({ output, expectedOutput }) => {
-      if (!expectedOutput) {
-        return { score: 0, label: 'error', reason: 'No expected output provided', passed: false };
-      }
-      try {
-        const ok = JSON.stringify(output) === JSON.stringify(JSON.parse(expectedOutput));
-        return {
-          score: ok ? 1 : 0,
-          label: ok ? 'correct' : 'incorrect',
-          reason: ok ? 'Exact match' : 'Mismatch',
-          passed: ok
-        };
-      } catch (e) {
-        return { score: 0, label: 'error', reason: 'Failed to parse expected output as JSON', passed: false };
-      }
-    });
-  return evalRegistry;
-}
+    } catch (e) {
+      return { score: 0, label: 'error', reason: 'Failed to parse expected output as JSON', passed: false };
+    }
+  },
+};
 
 function createClient() {
 ${loaderSetup}
   const modelRegistry = createModelRegistry();
-  const toolRegistry = createToolRegistry();
-  const evalRegistry = createEvalRegistry();
+${toolsVariable}
   ${createClientCall}
 }
 
@@ -1113,6 +1144,10 @@ var createExampleApp = async (client, targetPath = ".", apiKey = "", adapter = "
       } else {
         envVars[apiKeyEnvVar] = adapter === "claude-agent-sdk" ? "your-anthropic-api-key" : "your-openai-api-key";
       }
+      if (deploymentMode === "cloud") {
+        envVars["AGENTMARK_API_KEY"] = "your_agentmark_api_key";
+        envVars["AGENTMARK_APP_ID"] = "your_agentmark_app_id";
+      }
       const result = appendEnv(targetPath, envVars);
       if (result.added.length > 0) {
         console.log(`\u2705 Added to .env: ${result.added.join(", ")}`);
@@ -1121,7 +1156,7 @@ var createExampleApp = async (client, targetPath = ".", apiKey = "", adapter = "
         console.log(`\u23ED\uFE0F  Skipped existing .env vars: ${result.skipped.join(", ")}`);
       }
     } else {
-      fs4.writeFileSync(`${targetPath}/.env`, getEnvFileContent(modelProvider, apiKey, adapter));
+      fs4.writeFileSync(`${targetPath}/.env`, getEnvFileContent(modelProvider, apiKey, adapter, deploymentMode));
     }
     const gitignoreEntries = ["node_modules/", ".env", "*.agentmark-outputs/", "dist/"];
     if (shouldMergeFile(".gitignore", projectInfo, resolutions)) {
@@ -1166,6 +1201,59 @@ var createExampleApp = async (client, targetPath = ".", apiKey = "", adapter = "
 export default interface AgentmarkTypes {}
 `);
     }
+    if (deploymentMode === "cloud") {
+      const handlerAdapterConfig = getAdapterConfig(adapter, modelProvider);
+      const { webhookHandler: handlerClass } = handlerAdapterConfig.classes;
+      const handlerContent = `import { ${handlerClass} } from '${handlerAdapterConfig.package}/runner';
+import { AgentMarkSDK } from '@agentmark-ai/sdk';
+import { client } from './agentmark.client';
+
+// Initialize tracing \u2014 sends traces to AgentMark Cloud
+const sdk = new AgentMarkSDK({
+  apiKey: process.env.AGENTMARK_API_KEY ?? '',
+  appId: process.env.AGENTMARK_APP_ID ?? '',
+  baseUrl: process.env.AGENTMARK_BASE_URL,
+});
+sdk.initTracing({ disableBatch: true });
+
+const adapter = new ${handlerClass}(client as any);
+
+export default async function handler(request: {
+  type: 'prompt-run' | 'dataset-run';
+  data: {
+    ast: any;
+    customProps?: Record<string, unknown>;
+    options?: { shouldStream?: boolean };
+    experimentId?: string;
+    datasetPath?: string;
+  };
+}) {
+  if (request.type === 'prompt-run') {
+    return adapter.runPrompt(request.data.ast, {
+      shouldStream: request.data.options?.shouldStream,
+      customProps: request.data.customProps,
+    });
+  }
+
+  if (request.type === 'dataset-run') {
+    return adapter.runExperiment(
+      request.data.ast,
+      request.data.experimentId ?? '',
+      request.data.datasetPath,
+    );
+  }
+
+  throw new Error(\`Unknown request type: \${request.type}\`);
+}
+`;
+      const handlerPath = path2.join(targetPath, "handler.ts");
+      if (fs4.existsSync(handlerPath)) {
+        console.log("\u23ED\uFE0F  Skipped handler.ts (already exists - preserving customizations)");
+      } else {
+        fs4.writeFileSync(handlerPath, handlerContent);
+        console.log(`\u2705 Created handler.ts for cloud deployment`);
+      }
+    }
     console.log("Creating development server entry point...");
     const adapterConfig = getAdapterConfig(adapter, modelProvider);
     const { webhookHandler } = adapterConfig.classes;
@@ -1450,7 +1538,7 @@ var getAgentmarkClientContent = (_deploymentMode, adapter) => {
     return `"""AgentMark client configuration.
 
 This file configures the AgentMark client with Claude Agent SDK adapter.
-Customize the model registry and tool registry as needed.
+Customize the model registry as needed.
 """
 
 import os
@@ -1461,7 +1549,6 @@ from agentmark.prompt_core import FileLoader
 from agentmark_claude_agent_sdk import (
     create_claude_agent_client,
     create_default_model_registry,
-    ClaudeAgentToolRegistry,
 )
 
 # Load environment variables
@@ -1471,26 +1558,15 @@ load_dotenv()
 # Supports: claude-* models
 model_registry = create_default_model_registry()
 
-# Configure tool registry for custom tools
-tool_registry = ClaudeAgentToolRegistry()
-
-# Example tool registration:
-# tool_registry.register(
-#     "search",
-#     lambda args, ctx: f"Search results for: {args['query']}",
-#     description="Search the web",
-#     parameters={"type": "object", "properties": {"query": {"type": "string"}}, "required": ["query"]},
-# )
-
 # Create file loader for local development
 # Uses the project root as base directory for resolving relative paths
 project_root = Path(__file__).parent.resolve()
 loader = FileLoader(base_dir=str(project_root))
 
 # Create the client
+# Claude Agent SDK handles tools natively through the SDK
 client = create_claude_agent_client(
     model_registry=model_registry,
-    tool_registry=tool_registry,
     loader=loader,
 )
 
@@ -1500,7 +1576,7 @@ __all__ = ["client"]
   return `"""AgentMark client configuration.
 
 This file configures the AgentMark client with Pydantic AI adapter.
-Customize the model registry and tool registry as needed.
+Customize the model registry and tools as needed.
 """
 
 import os
@@ -1511,7 +1587,6 @@ from agentmark.prompt_core import FileLoader
 from agentmark_pydantic_ai_v0 import (
     create_pydantic_ai_client,
     create_default_model_registry,
-    PydanticAIToolRegistry,
 )
 
 # Load environment variables
@@ -1521,14 +1596,12 @@ load_dotenv()
 # Supports: gpt-*, claude-*, gemini-*, etc.
 model_registry = create_default_model_registry()
 
-# Configure tool registry for custom tools
-tool_registry = PydanticAIToolRegistry()
-
-# Example tool registration:
-# @tool_registry.register("search")
-# async def search_web(args: dict, ctx: dict | None) -> str:
-#     query = args["query"]
+# Define tools as native pydantic-ai Tool objects or callables
+# Example:
+# def search(query: str) -> str:
 #     return f"Search results for: {query}"
+# tools = [search]
+tools = []
 
 # Create file loader for local development
 # Uses the project root as base directory for resolving relative paths
@@ -1538,7 +1611,7 @@ loader = FileLoader(base_dir=str(project_root))
 # Create the client
 client = create_pydantic_ai_client(
     model_registry=model_registry,
-    tool_registry=tool_registry,
+    tools=tools,
     loader=loader,
 )
 
@@ -2145,6 +2218,9 @@ var main = async () => {
     usedModels = await createExampleApp(client, targetPath, apiKey, adapter, deploymentMode, projectInfo, resolutions);
   }
   config.builtInModels = usedModels;
+  if (deploymentMode === "cloud") {
+    config.handler = "handler.ts";
+  }
   const agentmarkJsonPath = path6.join(targetPath, "agentmark.json");
   const agentmarkJsonResolution = resolutions.find((r) => r.path === "agentmark.json");
   if (!fs8.existsSync(agentmarkJsonPath) || agentmarkJsonResolution?.action === "overwrite") {