ts-cua : lint + openai sdk types ; python-cua version bumps

index.ts

@@ -96,7 +96,7 @@ const INVOKE_SAMPLES: Record<
     [TEMPLATE_COMPUTER_USE]:
       'kernel invoke ts-cu cu-task --payload \'{"query": "Return the first url of a search result for NYC restaurant reviews Pete Wells"}\'',
     [TEMPLATE_CUA]:
-      'kernel invoke ts-cua cua-task --payload \'{"query": "Go to https://news.ycombinator.com and get the top 5 articles"}\'',
+      'kernel invoke ts-cua cua-task --payload \'{"task": "Go to https://news.ycombinator.com and get the top 5 articles"}\'',
   },
   [LANGUAGE_PYTHON]: {
     [TEMPLATE_SAMPLE_APP]:
@@ -108,7 +108,7 @@ const INVOKE_SAMPLES: Record<
     [TEMPLATE_COMPUTER_USE]:
       'kernel invoke python-cu cu-task --payload \'{"query": "Return the first url of a search result for NYC restaurant reviews Pete Wells"}\'',
     [TEMPLATE_CUA]:
-      'kernel invoke python-cua cua-task --payload \'{"query": "Go to https://news.ycombinator.com and get the top 5 articles"}\'',
+      'kernel invoke python-cua cua-task --payload \'{"task": "Go to https://news.ycombinator.com and get the top 5 articles"}\'',
   },
 };
 

templates/python/cua/pyproject.toml

@@ -6,24 +6,25 @@ readme = "README.md"
 requires-python = ">=3.11"
 dependencies = [
     "annotated-types==0.7.0",
-    "anyio==4.8.0",
-    "certifi==2025.1.31",
-    "charset-normalizer==3.4.1",
+    "anyio==4.9.0",
+    "certifi==2025.6.15",
+    "charset-normalizer==3.4.2",
     "distro==1.9.0",
-    "greenlet==3.1.1",
-    "h11==0.14.0",
-    "httpcore==1.0.7",
+    "greenlet==3.2.3",
+    "h11==0.16.0",
+    "httpcore==1.0.9",
     "httpx==0.28.1",
     "idna==3.10",
-    "jiter==0.8.2",
-    "pillow==11.1.0",
-    "playwright==1.50.0",
-    "pydantic==2.10.6",
-    "pydantic_core==2.27.2",
-    "pyee==12.1.1",
-    "python-dotenv==1.0.1",
-    "requests==2.32.3",
+    "jiter==0.10.0",
+    "pillow==11.2.1",
+    "kernel>=0.6.0",
+    "playwright==1.52.0",
+    "pydantic==2.11.7",
+    "pydantic_core==2.35.1",
+    "pyee==13.0.0",
+    "python-dotenv==1.1.0",
+    "requests==2.32.4",
     "sniffio==1.3.1",
-    "typing_extensions==4.12.2",
-    "urllib3==2.3.0",
+    "typing_extensions==4.14.0",
+    "urllib3==2.5.0",
 ]

templates/typescript/cua/.env.example

@@ -0,0 +1,2 @@
+OPENAI_API_KEY=YOUR_OPENAI_API_KEY
+# KERNEL_API_KEY=YOUR_KERNEL_KEY

templates/typescript/cua/.prettierrc

@@ -4,4 +4,4 @@
   "singleQuote": true,
   "printWidth": 100,
   "tabWidth": 2
-}
+}

templates/typescript/cua/README.md

@@ -5,4 +5,4 @@ This is a Kernel application that demonstrates using the Computer Using Agent (C
 It generally follows the [OpenAI CUA Sample App Reference](https://github.com/openai/openai-cua-sample-app) and uses Playwright via Kernel for browser automation.
 Also makes use of the latest OpenAI SDK format, and has local equivalent to Kernel methods for local testing before deploying on Kernel.
 
-See the [docs](https://docs.onkernel.com/quickstart) for information.
+See the [docs](https://docs.onkernel.com/quickstart) for information.

templates/typescript/cua/index.ts

@@ -1,14 +1,24 @@
-import "dotenv/config";
-import { Kernel, type KernelContext } from "@onkernel/sdk";
-import { Agent } from "./lib/agent";
-import computers from "./lib/computers";
+import 'dotenv/config';
+import { Kernel, type KernelContext } from '@onkernel/sdk';
+import { Agent } from './lib/agent';
+import computers from './lib/computers';
+import type { ResponseOutputMessage, ResponseItem } from 'openai/resources/responses/responses';
+
+interface CuaInput {
+  task: string;
+}
+interface CuaOutput {
+  elapsed: number;
+  answer: string | null;
+  logs?: ResponseItem[];
+}
 
 const kernel = new Kernel();
-const app = kernel.app("ts-cua");
+const app = kernel.app('ts-cua');
 
-// LLM API Keys are set in the environment during `kernel deploy <filename> -e ANTHROPIC_API_KEY=XXX`
-// See https://docs.onkernel.com/launch/deploy#environment-variables
-if (!process.env.OPENAI_API_KEY) throw new Error('OPENAI_API_KEY is not set');
+if (!process.env.OPENAI_API_KEY) {
+  throw new Error('OPENAI_API_KEY is not set');
+}
 
 /**
  * Example app that run an agent using openai CUA
@@ -24,88 +34,74 @@ if (!process.env.OPENAI_API_KEY) throw new Error('OPENAI_API_KEY is not set');
  *  kernel logs ts-cua -f # Open in separate tab
  */
 
-interface CuaInput {
-	task: string;
-}
-
-interface CuaOutput {
-	elapsed: number;
-	response?: Array<object>;
-	answer: object;
-}
-
 app.action<CuaInput, CuaOutput>(
-	"cua-task",
-	async (ctx: KernelContext, payload?: CuaInput): Promise<CuaOutput> => {
-		const startTime = Date.now();
-		const kernelBrowser = await kernel.browsers.create({
-			invocation_id: ctx.invocation_id,
-		});
-		console.log(
-			"> Kernel browser live view url: ",
-			kernelBrowser.browser_live_view_url,
-		);
-
-		if (!payload?.task){
-			throw new Error('task is required');
-		}
-
-		try {
+  'cua-task',
+  async (ctx: KernelContext, payload?: CuaInput): Promise<CuaOutput> => {
+    const start = Date.now();
+    if (!payload?.task) throw new Error('task is required');
 
-			// kernel browser
-			const { computer } = await computers.create({
-				type: "kernel", // for local testing before deploying to Kernel, you can use type: "local"
-				cdp_ws_url: kernelBrowser.cdp_ws_url,
-			});
+    try {
+      const kb = await kernel.browsers.create({ invocation_id: ctx.invocation_id });
+      console.log('> Kernel browser live view url:', kb.browser_live_view_url);
 
-			// setup agent
-			const agent = new Agent({
-				model: "computer-use-preview",
-				computer,
-				tools: [], // additional function_call tools to provide to the llm
-				acknowledge_safety_check_callback: (message: string) => {
-					console.log(`> safety check: ${message}`);
-					return true; // Auto-acknowledge all safety checks for testing
-				},
-			});
+      const { computer } = await computers.create({ type: 'kernel', cdp_ws_url: kb.cdp_ws_url });
+      const agent = new Agent({
+        model: 'computer-use-preview',
+        computer,
+        tools: [],
+        acknowledge_safety_check_callback: (m: string): boolean => {
+          console.log(`> safety check: ${m}`);
+          return true;
+        },
+      });
 
-			// start agent run
-			const response = await agent.runFullTurn({
-				messages: [
-					{
-						role: "system",
-						content: `- Current date and time: ${new Date().toISOString()} (${new Date().toLocaleDateString("en-US", { weekday: "long" })})`,
-					},
-					{
-						type: "message",
-						role: "user",
-						content: [
-							{
-								type: "input_text",
-								text: payload.task,
-								// text: "go to https://news.ycombinator.com , open top article , describe the target website design (in yaml format)"
-							},
-						],
-					},
-				],
-				print_steps: true, // log function_call and computer_call actions
-				debug: true, // show agent debug logs (llm messages and responses)
-				show_images: false, // if set to true, response messages stack will return base64 images (webp format) of screenshots, if false, replaced with "[omitted]""
-			});
+      // run agent and get response
+      const logs = await agent.runFullTurn({
+        messages: [
+          {
+            role: 'system',
+            content: `- Current date and time: ${new Date().toISOString()} (${new Date().toLocaleDateString(
+              'en-US',
+              { weekday: 'long' },
+            )})`,
+          },
+          {
+            type: 'message',
+            role: 'user',
+            content: [{ type: 'input_text', text: payload.task }],
+          },
+        ],
+        print_steps: true,
+        debug: true,
+        show_images: false,
+      });
 
-			console.log("> agent run done");
+      const elapsed = parseFloat(((Date.now() - start) / 1000).toFixed(2));
 
-			const endTime = Date.now();
-			const timeElapsed = (endTime - startTime) / 1000; // Convert to seconds
+      // filter only LLM messages
+      const messages = logs.filter(
+        (item): item is ResponseOutputMessage =>
+          item.type === 'message' &&
+          typeof (item as ResponseOutputMessage).role === 'string' &&
+          Array.isArray((item as ResponseOutputMessage).content),
+      );
+      const assistant = messages.find((m) => m.role === 'assistant');
+      const lastContentIndex = assistant?.content?.length ? assistant.content.length - 1 : -1;
+      const lastContent = lastContentIndex >= 0 ? assistant?.content?.[lastContentIndex] : null;
+      const answer = lastContent && 'text' in lastContent ? lastContent.text : null;
 
-			return {
-				// response, // full messages stack trace
-				elapsed: parseFloat(timeElapsed.toFixed(2)),
-				answer: response?.slice(-1)?.[0]?.content?.[0]?.text ?? null,
-			};
-		} finally {
-			// Note: KernelPlaywrightComputer handles browser cleanup internally
-			// No need to manually close browser here
-		}
-	},
+      return {
+        // logs, // optionally, get the full agent run messages logs
+        elapsed,
+        answer,
+      };
+    } catch (error) {
+      const elapsed = parseFloat(((Date.now() - start) / 1000).toFixed(2));
+      console.error('Error in cua-task:', error);
+      return {
+        elapsed,
+        answer: null,
+      };
+    }
+  },
 );