@empiricalrun/test-gen 0.53.4 → 0.53.5

package/CHANGELOG.md

@@ -1,5 +1,13 @@
 # @empiricalrun/test-gen
 
+## 0.53.5
+
+### Patch Changes
+
+- 9f3cb10: feat: automated tracing for LLM call overlay dismiss
+- Updated dependencies [9f3cb10]
+  - @empiricalrun/llm@0.14.4
+
 ## 0.53.4
 
 ### Patch Changes

package/dist/agent/cua/index.d.ts

@@ -1,8 +1,10 @@
+import { TraceClient } from "@empiricalrun/llm";
 import { Page } from "playwright";
 export declare function startPlaywrightCodegen(page: Page): Promise<void>;
-export declare function createTestUsingComputerUseAgent({ page, task, }: {
+export declare function createTestUsingComputerUseAgent({ page, task, trace, }: {
     page: Page;
     task: string;
+    trace?: TraceClient;
 }): Promise<{
     code: string;
     importPaths: string[];

package/dist/agent/cua/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/cua/index.ts"],"names":[],"mappings":"AASA,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAOlC,wBAAsB,sBAAsB,CAAC,IAAI,EAAE,IAAI,iBAoBtD;AAED,wBAAsB,+BAA+B,CAAC,EACpD,IAAI,EACJ,IAAI,GACL,EAAE;IACD,IAAI,EAAE,IAAI,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;CACd,GAAG,OAAO,CAAC;IACV,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,EAAE,CAAC;IACtB,cAAc,EAAE,MAAM,CAAC;CACxB,CAAC,CAmMD"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/cua/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAiB,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAS/D,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAMlC,wBAAsB,sBAAsB,CAAC,IAAI,EAAE,IAAI,iBAoBtD;AAED,wBAAsB,+BAA+B,CAAC,EACpD,IAAI,EACJ,IAAI,EACJ,KAAK,GACN,EAAE;IACD,IAAI,EAAE,IAAI,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB,GAAG,OAAO,CAAC;IACV,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,EAAE,CAAC;IACtB,cAAc,EAAE,MAAM,CAAC;CACxB,CAAC,CAqLD"}

package/dist/agent/cua/index.js

@@ -5,8 +5,7 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.createTestUsingComputerUseAgent = exports.startPlaywrightCodegen = void 0;
 const llm_1 = require("@empiricalrun/llm");
-const crypto_1 = __importDefault(require("crypto"));
-const logger_1 = require("../../bin/logger");
+const openai_1 = __importDefault(require("openai"));
 const utils_1 = require("../browsing/utils");
 const computer_1 = require("./computer");
 const model_1 = require("./model");
@@ -32,25 +31,18 @@ async function startPlaywrightCodegen(page) {
     await page.pause();
 }
 exports.startPlaywrightCodegen = startPlaywrightCodegen;
-async function createTestUsingComputerUseAgent({ page, task, }) {
+async function createTestUsingComputerUseAgent({ page, task, trace, }) {
     await (0, utils_1.injectPwLocatorGenerator)(page);
     const screenshotBytes = await (0, computer_1.getScreenshot)(page);
     const viewport = page.viewportSize();
     let screenWidth = viewport?.width || 1280;
     let screenHeight = viewport?.height || 720;
-    const logger = new logger_1.CustomLogger({ useReporter: false });
-    const trace = llm_1.langfuseInstance?.trace({
-        name: "computer-use-agent",
-        id: crypto_1.default.randomUUID(),
-        input: { task },
-    });
-    if (trace) {
-        const traceUrl = trace.getTraceUrl();
-        logger.log(`Starting computer use agent: ${traceUrl}`);
-    }
-    const span = trace?.span({
-        name: "initial-model-call",
-    });
+    const openAIClient = trace
+        ? (0, llm_1.observeOpenAI)(new openai_1.default(), {
+            generationName: `computer-use-agent`,
+            parent: trace,
+        })
+        : new openai_1.default();
     let response = await (0, model_1.callComputerUseModel)({
         input: [
             {
@@ -70,8 +62,8 @@ async function createTestUsingComputerUseAgent({ page, task, }) {
         ],
         screenWidth,
         screenHeight,
+        openAIClient,
     });
-    span?.end({ output: response });
     let isTaskDone = false;
     let maxIterations = 15;
     let generatedCode = "";
@@ -80,10 +72,6 @@ async function createTestUsingComputerUseAgent({ page, task, }) {
     while (!isTaskDone && iterationIndex < maxIterations) {
         actionsSummary.push(`\n# Agent iteration ${iterationIndex}`);
         iterationIndex++;
-        const iterationSpan = trace?.span({
-            name: `iteration-${iterationIndex}`,
-            input: { response },
-        });
         const computerCalls = response.output.filter((item) => item.type === "computer_call");
         const functionCalls = response.output.filter((item) => item.type === "function_call");
         if (computerCalls.length === 0 && functionCalls.length === 0) {
@@ -174,8 +162,8 @@ async function createTestUsingComputerUseAgent({ page, task, }) {
             ],
             screenWidth,
             screenHeight,
+            openAIClient,
         });
-        iterationSpan?.end({ output: response });
     }
     if (!isTaskDone) {
         actionsSummary.push(`Max iteration limit hit: Task not done after ${maxIterations} iterations`);

package/dist/agent/cua/model.d.ts

@@ -1,8 +1,10 @@
+import OpenAI from "openai";
 import { Response, ResponseInputItem } from "openai/resources/responses/responses.mjs";
-export declare function callComputerUseModel({ input, previousResponseId, screenWidth, screenHeight, }: {
+export declare function callComputerUseModel({ input, previousResponseId, screenWidth, screenHeight, openAIClient, }: {
     input: ResponseInputItem[];
     previousResponseId?: string;
     screenWidth: number;
     screenHeight: number;
+    openAIClient: OpenAI;
 }): Promise<Response>;
 //# sourceMappingURL=model.d.ts.map

package/dist/agent/cua/model.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"model.d.ts","sourceRoot":"","sources":["../../../src/agent/cua/model.ts"],"names":[],"mappings":"AACA,OAAO,EAEL,QAAQ,EACR,iBAAiB,EAClB,MAAM,0CAA0C,CAAC;AA8BlD,wBAAsB,oBAAoB,CAAC,EACzC,KAAK,EACL,kBAAkB,EAClB,WAAW,EACX,YAAY,GACb,EAAE;IACD,KAAK,EAAE,iBAAiB,EAAE,CAAC;IAC3B,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;CACtB,GAAG,OAAO,CAAC,QAAQ,CAAC,CAuBpB"}
+{"version":3,"file":"model.d.ts","sourceRoot":"","sources":["../../../src/agent/cua/model.ts"],"names":[],"mappings":"AAAA,OAAO,MAAM,MAAM,QAAQ,CAAC;AAC5B,OAAO,EAEL,QAAQ,EACR,iBAAiB,EAClB,MAAM,0CAA0C,CAAC;AA8BlD,wBAAsB,oBAAoB,CAAC,EACzC,KAAK,EACL,kBAAkB,EAClB,WAAW,EACX,YAAY,EACZ,YAAY,GACb,EAAE;IACD,KAAK,EAAE,iBAAiB,EAAE,CAAC;IAC3B,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,YAAY,EAAE,MAAM,CAAC;CACtB,GAAG,OAAO,CAAC,QAAQ,CAAC,CAuBpB"}

package/dist/agent/cua/model.js

@@ -1,10 +1,6 @@
 "use strict";
-var __importDefault = (this && this.__importDefault) || function (mod) {
-    return (mod && mod.__esModule) ? mod : { "default": mod };
-};
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.callComputerUseModel = void 0;
-const openai_1 = __importDefault(require("openai"));
 const INSTRUCTIONS = `You will be asked to execute some actions in a browser context.
 Don't ask the user for confirmations - just execute the actions.
 
@@ -30,9 +26,8 @@ const pageGotoTool = {
     },
     strict: true,
 };
-async function callComputerUseModel({ input, previousResponseId, screenWidth, screenHeight, }) {
-    const openai = new openai_1.default();
-    return await openai.responses.create({
+async function callComputerUseModel({ input, previousResponseId, screenWidth, screenHeight, openAIClient, }) {
+    const response = await openAIClient.responses.create({
         model: "computer-use-preview-2025-03-11",
         previous_response_id: previousResponseId,
         parallel_tool_calls: false,
@@ -53,5 +48,6 @@ async function callComputerUseModel({ input, previousResponseId, screenWidth, sc
         input,
         truncation: "auto",
     });
+    return response;
 }
 exports.callComputerUseModel = callComputerUseModel;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@empiricalrun/test-gen",
-  "version": "0.53.4",
+  "version": "0.53.5",
   "publishConfig": {
     "registry": "https://registry.npmjs.org/",
     "access": "public"
@@ -68,7 +68,7 @@
     "tsx": "^4.16.2",
     "typescript": "^5.3.3",
     "zod": "^3.23.8",
-    "@empiricalrun/llm": "^0.14.3",
+    "@empiricalrun/llm": "^0.14.4",
     "@empiricalrun/r2-uploader": "^0.3.8",
     "@empiricalrun/test-run": "^0.7.6"
   },