update samples

sdk/ai/ai-projects/samples-dev/agents/tools/agentToolboxWithGuardrail.ts

@@ -0,0 +1,94 @@
+// Copyright (c) Microsoft Corporation.
+// Licensed under the MIT License.
+
+/**
+ * This sample demonstrates how to create an agent that uses a toolbox with a guardrail (RAI policy).
+ * The toolbox is configured with a web_search tool and an RAI policy that filters responses.
+ *
+ * @summary Create an agent with a toolbox that has a guardrail (RAI policy) applied.
+ *
+ * @azsdk-weight 100
+ */
+
+import { DefaultAzureCredential } from "@azure/identity";
+import { AIProjectClient, type MCPTool } from "@azure/ai-projects";
+import "dotenv/config";
+
+const projectEndpoint = process.env["FOUNDRY_PROJECT_ENDPOINT"] || "<project endpoint>";
+const deploymentName = process.env["FOUNDRY_MODEL_NAME"] || "<model deployment name>";
+const raiPolicyName =
+  process.env["RAI_POLICY_NAME"] ||
+  "/subscriptions/<subscription-id>/resourceGroups/<resource-group>/providers/Microsoft.CognitiveServices/accounts/<account-name>/raiPolicies/<policy-name>";
+
+export async function main(): Promise<void> {
+  const credential = new DefaultAzureCredential();
+  const project = new AIProjectClient(projectEndpoint, credential);
+  const openAIClient = project.getOpenAIClient();
+
+  console.log("Creating toolbox with guardrail (RAI policy)...");
+  const toolboxVersion = await project.beta.toolboxes.createVersion(
+    "my-toolbox",
+    [{ type: "web_search" }],
+    {
+      description: "Toolbox with guardrail",
+      policies: {
+        rai_config: {
+          rai_policy_name: raiPolicyName,
+        },
+      },
+    },
+  );
+  console.log(`Toolbox version created: ${toolboxVersion.version}`);
+
+  console.log("\nCreating agent with deferred MCP tool and tool_search...");
+  const token = (await credential.getToken("https://ai.azure.com/.default")).token;
+  const toolboxMcpUrl = `${projectEndpoint}/toolboxes/my-toolbox/versions/${toolboxVersion.version}/mcp?api-version=v1`;
+
+  const agent = await project.agents.createVersion("MyGuardrailAgent", {
+    kind: "prompt",
+    model: deploymentName,
+    instructions: "You are a helpful assistant. Use web search when needed to answer questions.",
+    tools: [
+      {
+        type: "mcp",
+        server_label: "my-toolbox",
+        server_url: toolboxMcpUrl,
+        authorization: token,
+        headers: { "Foundry-Features": "Toolboxes=V1Preview" },
+        require_approval: "never",
+      } as MCPTool,
+    ],
+  });
+  console.log(`Agent created (id: ${agent.id}, name: ${agent.name}, version: ${agent.version})`);
+
+  console.log("\nSending request to the agent...");
+  try {
+    const response = await openAIClient.responses.create(
+      {
+        input: "What are the latest developments in responsible AI?",
+      },
+      {
+        body: {
+          agent_reference: { name: agent.name, type: "agent_reference" },
+        },
+      },
+    );
+    console.log(`\nResponse output: ${response.output_text}`);
+  } catch (e: any) {
+    if (e.status === 400 && e.message?.includes("blocked by policy: RAI")) {
+      console.log("\nThe RAI guardrail blocked the tool call as expected.");
+      console.log(`Details: ${e.message}`);
+    } else {
+      throw e;
+    }
+  }
+
+  console.log("\nCleaning up resources...");
+  await project.agents.deleteVersion(agent.name, agent.version);
+  await project.beta.toolboxes.deleteVersion("my-toolbox", toolboxVersion.version);
+  console.log("Agent and toolbox deleted");
+}
+
+main().catch((err) => {
+  console.error("The sample encountered an error:", err);
+});

sdk/ai/ai-projects/samples-dev/skills/skillWithToolbox.ts

@@ -0,0 +1,155 @@
+// Copyright (c) Microsoft Corporation.
+// Licensed under the MIT License.
+
+/**
+ * This sample demonstrates how to expose a Skill to a Prompt Agent via a
+ * Toolbox, using the AIProjectClient and the OpenAI-compatible client.
+ *
+ * It creates a Skill with inline content describing how to compute shipping
+ * cost, then creates a Toolbox version that references the skill. A Prompt
+ * Agent is created with an `MCPTool` pointed at the toolbox's versioned
+ * `/mcp` endpoint. The skill's instructions are injected into the agent's
+ * context, so when asked a shipping-cost question the agent answers directly
+ * using the skill's formula.
+ *
+ * Skills and Toolboxes are currently preview features. In the JS SDK, you
+ * access these operations via `project.beta.skills` and
+ * `project.beta.toolboxes`.
+ *
+ * @summary Demonstrates adding a skill to a toolbox and invoking it via a Prompt Agent.
+ *
+ * @azsdk-weight 100
+ */
+
+import type { MCPTool, ToolboxSkillReference } from "@azure/ai-projects";
+import { AIProjectClient, RestError } from "@azure/ai-projects";
+import { DefaultAzureCredential } from "@azure/identity";
+import "dotenv/config";
+
+const projectEndpoint = process.env["FOUNDRY_PROJECT_ENDPOINT"] || "<project endpoint>";
+const deploymentName = process.env["FOUNDRY_MODEL_NAME"] || "<model deployment name>";
+
+const SKILL_NAME = "shipping-cost-skill";
+const TOOLBOX_NAME = "toolbox_with_skill";
+const AGENT_NAME = "SkillToolboxAgent";
+
+export async function main(): Promise<void> {
+  const credential = new DefaultAzureCredential();
+  const project = new AIProjectClient(projectEndpoint, credential);
+  const openAIClient = project.getOpenAIClient();
+
+  // --- Clean up any prior runs ---
+  try {
+    await project.beta.toolboxes.delete(TOOLBOX_NAME);
+  } catch (e) {
+    if (!(e instanceof RestError && e.statusCode === 404)) throw e;
+  }
+  try {
+    await project.beta.skills.delete(SKILL_NAME);
+  } catch (e) {
+    if (!(e instanceof RestError && e.statusCode === 404)) throw e;
+  }
+
+  // --- 1. Create a skill ---
+  const skill = await project.beta.skills.create(SKILL_NAME, {
+    inlineContent: {
+      description: "Compute shipping cost for a package given weight and destination.",
+      instructions:
+        "You are a shipping cost calculator. When asked to compute " +
+        "shipping cost, use this formula: cost (USD) = 5 + 2 * weight_kg " +
+        "for domestic destinations, and cost (USD) = 15 + 4 * weight_kg " +
+        "for international destinations. Always state the formula you used.",
+      metadata: { revision: "1" },
+    },
+  });
+  console.log(`Created skill: ${skill.name} version=${skill.version}`);
+
+  // --- 2. Create a toolbox version with toolbox_search_preview and the skill ---
+  const skillRef: ToolboxSkillReference = {
+    type: "skill_reference",
+    name: skill.name,
+    version: skill.version,
+  };
+
+  const toolboxVersion = await project.beta.toolboxes.createVersion(
+    TOOLBOX_NAME,
+    [{ type: "toolbox_search_preview" }],
+    {
+      description: "Toolbox exposing a shipping-cost skill.",
+      skills: [skillRef],
+    },
+  );
+  console.log(`Created toolbox: ${toolboxVersion.name} version=${toolboxVersion.version}`);
+
+  // --- 3. Build the MCP tool reference to the toolbox endpoint ---
+  const toolboxMcpUrl = `${projectEndpoint}/toolboxes/${TOOLBOX_NAME}/versions/${toolboxVersion.version}/mcp?api-version=v1`;
+  const token = (await credential.getToken("https://ai.azure.com/.default"))!.token;
+
+  const toolboxMcpTool: MCPTool = {
+    type: "mcp",
+    server_label: "skill-toolbox",
+    server_url: toolboxMcpUrl,
+    authorization: token,
+    headers: { "Foundry-Features": "Toolboxes=V1Preview" },
+    require_approval: "never",
+  };
+
+  // --- 4. Create a prompt agent with the toolbox MCP tool ---
+  const agent = await project.agents.createVersion(AGENT_NAME, {
+    kind: "prompt",
+    model: deploymentName,
+    instructions:
+      "Answer the user using the `shipping-cost-skill` instructions " +
+      "available in your context. Do not call `tool_search`; the " +
+      "skill rules are already part of your knowledge. Apply the " +
+      "skill's formula exactly as given and state the formula in " +
+      "your answer.",
+    temperature: 0,
+    tools: [toolboxMcpTool],
+  });
+  console.log(`Agent created (id=${agent.id}, name=${agent.name}, version=${agent.version})`);
+
+  // --- 5. Send a query ---
+  const userInput = "Compute the shipping cost for a 3 kg package shipped domestically.";
+  console.log(`User: ${userInput}`);
+
+  const response = await openAIClient.responses.create(
+    { input: userInput },
+    {
+      body: {
+        agent_reference: { name: agent.name, type: "agent_reference" },
+      },
+    },
+  );
+
+  for (const item of response.output) {
+    if (item.type === "mcp_list_tools") {
+      console.log(
+        `mcp_list_tools server_label=${item.server_label} tools=${(item.tools || []).map((t: any) => t.name)}`,
+      );
+    } else if (item.type === "mcp_call") {
+      console.log(
+        `mcp_call server_label=${item.server_label} name=${item.name} error=${item.error}`,
+      );
+      if ("output" in item && item.output) {
+        console.log(`  output: ${item.output}`);
+      }
+    } else if (item.type === "mcp_approval_request") {
+      console.log(`mcp_approval_request server_label=${item.server_label} name=${item.name}`);
+    } else {
+      console.log(`output item type=${item.type}`);
+    }
+  }
+
+  console.log(`Response: ${response.output_text}`);
+
+  // --- 6. Clean up ---
+  await project.beta.toolboxes.delete(TOOLBOX_NAME);
+  console.log("Toolbox deleted");
+  await project.beta.skills.delete(SKILL_NAME);
+  console.log("Skill deleted");
+}
+
+main().catch((err) => {
+  console.error("The sample encountered an error:", err);
+});