@empiricalrun/test-gen 0.79.4 → 0.79.6

package/CHANGELOG.md

@@ -1,5 +1,25 @@
 # @empiricalrun/test-gen
 
+## 0.79.6
+
+### Patch Changes
+
+- Updated dependencies [b4e2fff]
+  - @empiricalrun/cua@0.3.0
+  - @empiricalrun/llm@0.26.0
+  - videostil@0.3.5
+
+## 0.79.5
+
+### Patch Changes
+
+- b58777e: refactor: extract CUA agent into standalone @empiricalrun/cua package
+- fda9bc3: feat: extract DashboardAPIClient into standalone @empiricalrun/dashboard-client package
+- Updated dependencies [b58777e]
+- Updated dependencies [fda9bc3]
+  - @empiricalrun/cua@0.2.0
+  - @empiricalrun/dashboard-client@0.2.0
+
 ## 0.79.4
 
 ### Patch Changes

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

@@ -1,5 +1,5 @@
 import { type IChatModel } from "@empiricalrun/llm/chat";
-import type { CanonicalMessage, ChatState, ChatStateGeneric, MessageCheckpoint, PendingToolCall, ReporterFunction, StreamingMessageReporterFunc, SupportedChatModels, TextMessagePart, TextMessageWithAttachmentsPart, ToolResult, ToolsForLLM } from "@empiricalrun/shared-types/chat-agent";
+import type { CanonicalMessage, ChatState, ChatStateGeneric, MessageCheckpoint, PendingToolCall, ProcessToolResultsResult, ReporterFunction, StreamingMessageReporterFunc, SupportedChatModels, TextMessagePart, TextMessageWithAttachmentsPart, ToolResult, ToolsForLLM } from "@empiricalrun/shared-types/chat-agent";
 import type { TraceClient } from "@empiricalrun/shared-types/common";
 import type { FileInfo } from "@empiricalrun/shared-types/test-gen";
 import type { IToolExecutor } from "../../tools/executor/types";
@@ -50,7 +50,7 @@ export declare abstract class BaseAgent<T> {
     setChatState(chatState: ChatStateGeneric<T>): void;
     changeModel(newModel: SupportedChatModels): void;
     stop(): ChatStateGeneric<T>;
-    processToolResults(toolCalls: PendingToolCall[], toolResults: ToolResult[], checkpoint: MessageCheckpoint | null): ChatState | undefined;
+    processToolResults(toolCalls: PendingToolCall[], toolResults: ToolResult[], checkpoint: MessageCheckpoint | null): ProcessToolResultsResult;
     toolCallHandler(toolCalls: PendingToolCall[]): Promise<void>;
     runLoop(params: RunLoopParams<any>): Promise<void>;
     getHumanReadableLatestMessage(): {

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

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/base/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAmB,KAAK,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAC1E,OAAO,KAAK,EACV,gBAAgB,EAChB,SAAS,EACT,gBAAgB,EAChB,iBAAiB,EACjB,eAAe,EACf,gBAAgB,EAChB,4BAA4B,EAC5B,mBAAmB,EACnB,eAAe,EACf,8BAA8B,EAC9B,UAAU,EACV,WAAW,EACZ,MAAM,uCAAuC,CAAC;AAC/C,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,mCAAmC,CAAC;AACrE,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,qCAAqC,CAAC;AAEpE,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,4BAA4B,CAAC;AAMhE,KAAK,aAAa,CAAC,CAAC,IAAI;IACtB,MAAM,CAAC,EAAE,WAAW,CAAC;IACrB,QAAQ,EAAE,gBAAgB,CAAC,CAAC,CAAC,CAAC;IAC9B,wBAAwB,CAAC,EAAE,4BAA4B,CAAC;IACxD,aAAa,CAAC,EAAE,CACd,QAAQ,EAAE,CAAC,EACX,aAAa,EAAE,mBAAmB,KAC/B,OAAO,CAAC,IAAI,CAAC,CAAC;CACpB,CAAC;AAEF,KAAK,wBAAwB,CAAC,CAAC,IAAI;IACjC,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,SAAS,CAAC,EAAE,gBAAgB,CAAC,CAAC,CAAC,CAAC;IAChC,SAAS,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACnC,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;IACxB,YAAY,CAAC,EAAE,aAAa,CAAC;IAC7B,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,aAAa,EAAE,mBAAmB,CAAC;IACnC,eAAe,CAAC,EAAE,MAAM,OAAO,CAAC,QAAQ,CAAC,CAAC;CAC3C,CAAC;AAEF,MAAM,MAAM,WAAW,CAAC,CAAC,IAAI,wBAAwB,CAAC,CAAC,CAAC,GACtD,CAAC;IAAE,YAAY,EAAE,MAAM,CAAA;CAAE,GAAG;IAAE,eAAe,EAAE,MAAM,OAAO,CAAC,QAAQ,CAAC,CAAA;CAAE,CAAC,CAAC;AAE5E,MAAM,MAAM,8BAA8B,GAAG,CAC3C,YAAY,EAAE,SAAS,KACpB,OAAO,CAAC,IAAI,CAAC,CAAC;AAEnB,MAAM,MAAM,kBAAkB,GAAG,CAC/B,SAAS,EAAE,eAAe,EAAE,KACzB,OAAO,CAAC,IAAI,CAAC,CAAC;AAEnB,MAAM,MAAM,qBAAqB,CAAC,CAAC,IAAI,CACrC,QAAQ,EAAE,CAAC,EACX,aAAa,EAAE,mBAAmB,KAC/B,OAAO,CAAC,IAAI,CAAC,CAAC;AAEnB,8BAAsB,SAAS,CAAC,CAAC;IAC/B,YAAY,EAAE,MAAM,EAAE,CAAM;IAC5B,KAAK,EAAE,WAAW,CAAC;IACnB,aAAa,EAAE,mBAAmB,CAAC;IACnC,SAAS,EAAE,UAAU,CAAC,CAAC,CAAC,CAAC;IACzB,SAAS,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACnC,YAAY,CAAC,EAAE,aAAa,CAAC;IAC7B,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,YAAY,EAAE,MAAM,GAAG,SAAS,CAAC;IACjC,eAAe,EAAE,CAAC,MAAM,OAAO,CAAC,QAAQ,CAAC,CAAC,GAAG,SAAS,CAAC;IACvD,sBAAsB,CAAC,EAAE,8BAA8B,CAAC;IACxD,UAAU,CAAC,EAAE,kBAAkB,CAAC;IAChC,aAAa,CAAC,EAAE,qBAAqB,CAAC,CAAC,CAAC,CAAC;gBAE7B,EACV,YAAY,EACZ,aAAa,EACb,SAAS,EACT,SAAS,EACT,YAAY,EACZ,KAAK,EACL,YAAY,EACZ,eAAe,GAChB,EAAE,WAAW,CAAC,CAAC,CAAC;IAuBjB,SAAS,CAAC,QAAQ,CAAC,QAAQ,IAAI,WAAW;IAC1C,SAAS,CAAC,QAAQ,CAAC,iBAAiB,CAClC,eAAe,EAAE,MAAM,OAAO,CAAC,QAAQ,CAAC,GACvC,OAAO,CAAC,MAAM,CAAC;IAElB,IAAI,QAAQ,IAAI,gBAAgB,EAAE,CAEjC;IAED,IAAI,eAAe,IAAI,OAAO,CAE7B;IAED,IAAI,SAAS,IAAI,gBAAgB,CAAC,CAAC,CAAC,CAKnC;IAED,eAAe,CACb,KAAK,EAAE,KAAK,CAAC,eAAe,GAAG,8BAA8B,CAAC,GAC7D,gBAAgB,CAAC,CAAC,CAAC;IAKtB,YAAY,CAAC,SAAS,EAAE,gBAAgB,CAAC,CAAC,CAAC;IAQ3C,WAAW,CAAC,QAAQ,EAAE,mBAAmB,GAAG,IAAI;IAUhD,IAAI,IAAI,gBAAgB,CAAC,CAAC,CAAC;IA6B3B,kBAAkB,CAChB,SAAS,EAAE,eAAe,EAAE,EAC5B,WAAW,EAAE,UAAU,EAAE,EACzB,UAAU,EAAE,iBAAiB,GAAG,IAAI,GACnC,SAAS,GAAG,SAAS;IA4BlB,eAAe,CAAC,SAAS,EAAE,eAAe,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC;IAoB5D,OAAO,CAAC,MAAM,EAAE,aAAa,CAAC,GAAG,CAAC;IA2BxC,6BAA6B,IACzB;QACE,IAAI,EAAE,MAAM,CAAC;QACb,WAAW,EAAE,MAAM,CAAC;KACrB,GACD,SAAS;CAGd"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/base/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAmB,KAAK,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAC1E,OAAO,KAAK,EACV,gBAAgB,EAChB,SAAS,EACT,gBAAgB,EAChB,iBAAiB,EACjB,eAAe,EACf,wBAAwB,EACxB,gBAAgB,EAChB,4BAA4B,EAC5B,mBAAmB,EACnB,eAAe,EACf,8BAA8B,EAC9B,UAAU,EACV,WAAW,EACZ,MAAM,uCAAuC,CAAC;AAC/C,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,mCAAmC,CAAC;AACrE,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,qCAAqC,CAAC;AAEpE,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,4BAA4B,CAAC;AAMhE,KAAK,aAAa,CAAC,CAAC,IAAI;IACtB,MAAM,CAAC,EAAE,WAAW,CAAC;IACrB,QAAQ,EAAE,gBAAgB,CAAC,CAAC,CAAC,CAAC;IAC9B,wBAAwB,CAAC,EAAE,4BAA4B,CAAC;IACxD,aAAa,CAAC,EAAE,CACd,QAAQ,EAAE,CAAC,EACX,aAAa,EAAE,mBAAmB,KAC/B,OAAO,CAAC,IAAI,CAAC,CAAC;CACpB,CAAC;AAEF,KAAK,wBAAwB,CAAC,CAAC,IAAI;IACjC,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,SAAS,CAAC,EAAE,gBAAgB,CAAC,CAAC,CAAC,CAAC;IAChC,SAAS,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACnC,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;IACxB,YAAY,CAAC,EAAE,aAAa,CAAC;IAC7B,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,aAAa,EAAE,mBAAmB,CAAC;IACnC,eAAe,CAAC,EAAE,MAAM,OAAO,CAAC,QAAQ,CAAC,CAAC;CAC3C,CAAC;AAEF,MAAM,MAAM,WAAW,CAAC,CAAC,IAAI,wBAAwB,CAAC,CAAC,CAAC,GACtD,CAAC;IAAE,YAAY,EAAE,MAAM,CAAA;CAAE,GAAG;IAAE,eAAe,EAAE,MAAM,OAAO,CAAC,QAAQ,CAAC,CAAA;CAAE,CAAC,CAAC;AAE5E,MAAM,MAAM,8BAA8B,GAAG,CAC3C,YAAY,EAAE,SAAS,KACpB,OAAO,CAAC,IAAI,CAAC,CAAC;AAEnB,MAAM,MAAM,kBAAkB,GAAG,CAC/B,SAAS,EAAE,eAAe,EAAE,KACzB,OAAO,CAAC,IAAI,CAAC,CAAC;AAEnB,MAAM,MAAM,qBAAqB,CAAC,CAAC,IAAI,CACrC,QAAQ,EAAE,CAAC,EACX,aAAa,EAAE,mBAAmB,KAC/B,OAAO,CAAC,IAAI,CAAC,CAAC;AAEnB,8BAAsB,SAAS,CAAC,CAAC;IAC/B,YAAY,EAAE,MAAM,EAAE,CAAM;IAC5B,KAAK,EAAE,WAAW,CAAC;IACnB,aAAa,EAAE,mBAAmB,CAAC;IACnC,SAAS,EAAE,UAAU,CAAC,CAAC,CAAC,CAAC;IACzB,SAAS,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACnC,YAAY,CAAC,EAAE,aAAa,CAAC;IAC7B,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,YAAY,EAAE,MAAM,GAAG,SAAS,CAAC;IACjC,eAAe,EAAE,CAAC,MAAM,OAAO,CAAC,QAAQ,CAAC,CAAC,GAAG,SAAS,CAAC;IACvD,sBAAsB,CAAC,EAAE,8BAA8B,CAAC;IACxD,UAAU,CAAC,EAAE,kBAAkB,CAAC;IAChC,aAAa,CAAC,EAAE,qBAAqB,CAAC,CAAC,CAAC,CAAC;gBAE7B,EACV,YAAY,EACZ,aAAa,EACb,SAAS,EACT,SAAS,EACT,YAAY,EACZ,KAAK,EACL,YAAY,EACZ,eAAe,GAChB,EAAE,WAAW,CAAC,CAAC,CAAC;IAuBjB,SAAS,CAAC,QAAQ,CAAC,QAAQ,IAAI,WAAW;IAC1C,SAAS,CAAC,QAAQ,CAAC,iBAAiB,CAClC,eAAe,EAAE,MAAM,OAAO,CAAC,QAAQ,CAAC,GACvC,OAAO,CAAC,MAAM,CAAC;IAElB,IAAI,QAAQ,IAAI,gBAAgB,EAAE,CAEjC;IAED,IAAI,eAAe,IAAI,OAAO,CAE7B;IAED,IAAI,SAAS,IAAI,gBAAgB,CAAC,CAAC,CAAC,CAKnC;IAED,eAAe,CACb,KAAK,EAAE,KAAK,CAAC,eAAe,GAAG,8BAA8B,CAAC,GAC7D,gBAAgB,CAAC,CAAC,CAAC;IAKtB,YAAY,CAAC,SAAS,EAAE,gBAAgB,CAAC,CAAC,CAAC;IAQ3C,WAAW,CAAC,QAAQ,EAAE,mBAAmB,GAAG,IAAI;IAUhD,IAAI,IAAI,gBAAgB,CAAC,CAAC,CAAC;IA6B3B,kBAAkB,CAChB,SAAS,EAAE,eAAe,EAAE,EAC5B,WAAW,EAAE,UAAU,EAAE,EACzB,UAAU,EAAE,iBAAiB,GAAG,IAAI,GACnC,wBAAwB;IAwErB,eAAe,CAAC,SAAS,EAAE,eAAe,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC;IAoB5D,OAAO,CAAC,MAAM,EAAE,aAAa,CAAC,GAAG,CAAC;IA2BxC,6BAA6B,IACzB;QACE,IAAI,EAAE,MAAM,CAAC;QACb,WAAW,EAAE,MAAM,CAAC;KACrB,GACD,SAAS;CAGd"}

package/dist/agent/base/index.js

@@ -97,12 +97,43 @@ class BaseAgent {
         if (filteredToolCalls.length > 0) {
             this.chatModel.pushToolResultsMessage(filteredToolCalls, filteredToolResults, checkpoint);
             return {
-                ...this.chatState,
-                messages: this.chatModel.messages,
-                askUserForInput: this.chatModel.askUserForInput,
+                status: "accepted",
+                chatState: {
+                    ...this.chatState,
+                    messages: this.chatModel.messages,
+                    askUserForInput: this.chatModel.askUserForInput,
+                },
             };
         }
-        return undefined;
+        // Check if these tool calls were already completed (idempotent retry)
+        const messages = this.chatModel.messages;
+        const completedToolCallIds = new Set();
+        for (const msg of messages) {
+            if (msg.role === "tool") {
+                for (const part of msg.parts) {
+                    if ("toolCallId" in part && part.toolCallId) {
+                        completedToolCallIds.add(part.toolCallId);
+                    }
+                }
+            }
+        }
+        if (toolCalls.every((tc) => completedToolCallIds.has(tc.id))) {
+            console.log(`[processToolResults] Tool calls already completed (idempotent retry): ${toolCalls.map((tc) => `${tc.id} (${tc.name})`).join(", ")}`);
+            return { status: "already_accepted" };
+        }
+        // Diagnose why no tool calls matched
+        for (const call of toolCalls) {
+            if (completedToolCallIds.has(call.id)) {
+                console.log(`[processToolResults] Tool call ${call.id} (${call.name}) already has a result`);
+            }
+            else {
+                console.log(`[processToolResults] Tool call ${call.id} (${call.name}) not found in messages (${messages.length} messages)`);
+            }
+        }
+        if (pendingToolCalls.length > 0) {
+            console.log(`[processToolResults] Pending tool calls: ${pendingToolCalls.map((p) => `${p.id} (${p.name})`).join(", ")}`);
+        }
+        return { status: "rejected" };
     }
     async toolCallHandler(toolCalls) {
         if (!this.toolExecutor) {
@@ -113,9 +144,9 @@ class BaseAgent {
         }
         const result = await this.toolExecutor.execute(toolCalls);
         if (result.status === "completed") {
-            const newChatState = this.processToolResults(toolCalls, result.toolResults, result.checkpoint);
-            if (newChatState && this.onToolResultsProcessed) {
-                await this.onToolResultsProcessed(newChatState);
+            const { status, chatState } = this.processToolResults(toolCalls, result.toolResults, result.checkpoint);
+            if (status === "accepted" && chatState && this.onToolResultsProcessed) {
+                await this.onToolResultsProcessed(chatState);
             }
         }
     }

package/dist/agent/browsing/run.d.ts

@@ -1,5 +1,5 @@
+import { BrowserAgentResult } from "@empiricalrun/cua";
 import { Usage } from "@empiricalrun/shared-types/chat-agent";
-import { BrowserAgentResult } from "../cua";
 type GenerateTestsType = {
     testCaseName: string;
     testCaseSuites: string[];

package/dist/agent/browsing/run.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/run.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAE,MAAM,uCAAuC,CAAC;AAa9D,OAAO,EAAE,kBAAkB,EAAE,MAAM,QAAQ,CAAC;AAG5C,KAAK,iBAAiB,GAAG;IACvB,YAAY,EAAE,MAAM,CAAC;IACrB,cAAc,EAAE,MAAM,EAAE,CAAC;IACzB,YAAY,EAAE,MAAM,CAAC;IACrB,gBAAgB,EAAE,MAAM,CAAC;IACzB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,yBAAyB,EAAE,OAAO,CAAC;IACnC,YAAY,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CACtC,CAAC;AAEF,wBAAsB,8BAA8B,CAAC,EACnD,gBAAgB,EAChB,OAAO,EACP,YAAY,GACb,EAAE;IACD,gBAAgB,EAAE,MAAM,EAAE,CAAC;IAC3B,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;CACtB,GAAG,OAAO,CAAC,MAAM,CAAC,CAQlB;AAED,wBAAsB,gBAAgB,CAAC,EACrC,YAAY,EACZ,cAAc,EACd,YAAY,EACZ,gBAAgB,EAChB,WAAW,EACX,YAAY,EACZ,OAAO,EACP,OAAO,EACP,yBAAyB,EACzB,YAAY,GACb,EAAE,iBAAiB,GAAG,OAAO,CAAC;IAC7B,OAAO,EAAE,OAAO,CAAC;IACjB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,kBAAkB,GAAG,SAAS,CAAC;IACvC,KAAK,EAAE,KAAK,GAAG,SAAS,CAAC;CAC1B,CAAC,CA+FD"}
+{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/run.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,kBAAkB,EAAE,MAAM,mBAAmB,CAAC;AACvD,OAAO,EAAE,KAAK,EAAE,MAAM,uCAAuC,CAAC;AAc9D,KAAK,iBAAiB,GAAG;IACvB,YAAY,EAAE,MAAM,CAAC;IACrB,cAAc,EAAE,MAAM,EAAE,CAAC;IACzB,YAAY,EAAE,MAAM,CAAC;IACrB,gBAAgB,EAAE,MAAM,CAAC;IACzB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,yBAAyB,EAAE,OAAO,CAAC;IACnC,YAAY,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CACtC,CAAC;AAEF,wBAAsB,8BAA8B,CAAC,EACnD,gBAAgB,EAChB,OAAO,EACP,YAAY,GACb,EAAE;IACD,gBAAgB,EAAE,MAAM,EAAE,CAAC;IAC3B,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;CACtB,GAAG,OAAO,CAAC,MAAM,CAAC,CAQlB;AAED,wBAAsB,gBAAgB,CAAC,EACrC,YAAY,EACZ,cAAc,EACd,YAAY,EACZ,gBAAgB,EAChB,WAAW,EACX,YAAY,EACZ,OAAO,EACP,OAAO,EACP,yBAAyB,EACzB,YAAY,GACb,EAAE,iBAAiB,GAAG,OAAO,CAAC;IAC7B,OAAO,EAAE,OAAO,CAAC;IACjB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,kBAAkB,GAAG,SAAS,CAAC;IACvC,KAAK,EAAE,KAAK,GAAG,SAAS,CAAC;CAC1B,CAAC,CA+FD"}

package/dist/agent/browsing/utils.d.ts

@@ -1,21 +1,14 @@
 import { Page } from "playwright";
 import { PlaywrightTestConfig } from "playwright/test";
 export declare function isRegExp(obj: any): obj is RegExp;
-export declare function prepareBrowsingAgentTask(steps: string[]): string;
 export declare function addImportForMethod(testFilePath: string, methodName: string): void;
 export declare function replaceTodoWithCreateTest(testFilePath: string, repoDir: string): string | undefined;
-export declare function markTestAsOnly({ testCaseName, testCaseSuites, specPath, }: {
-    testCaseName: string;
-    testCaseSuites: string[];
-    specPath: string;
-}): Promise<void>;
 export declare function injectPwLocatorGenerator(page: Page): Promise<void>;
 /**
  * function to read playwright config from the source repo
  * @return {*}  {Promise<PlaywrightTestConfig>}
  */
 export declare function readPlaywrightConfig(repoDir: string): Promise<PlaywrightTestConfig>;
-export declare function getValidProjectNames(playwrightConfig: PlaywrightTestConfig): Promise<string[]>;
 /**
  * detect the project name for the given file in playwright test repo
  * if project and test file path for running test don't match, then playwright throws error

package/dist/agent/browsing/utils.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/utils.ts"],"names":[],"mappings":"AAGA,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAClC,OAAO,EAAE,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AAYvD,wBAAgB,QAAQ,CAAC,GAAG,EAAE,GAAG,GAAG,GAAG,IAAI,MAAM,CAKhD;AAED,wBAAgB,wBAAwB,CAAC,KAAK,EAAE,MAAM,EAAE,UAIvD;AAED,wBAAgB,kBAAkB,CAAC,YAAY,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,QAe1E;AAED,wBAAgB,yBAAyB,CACvC,YAAY,EAAE,MAAM,EACpB,OAAO,EAAE,MAAM,GACd,MAAM,GAAG,SAAS,CAsBpB;AAED,wBAAsB,cAAc,CAAC,EACnC,YAAY,EACZ,cAAc,EACd,QAAQ,GACT,EAAE;IACD,YAAY,EAAE,MAAM,CAAC;IACrB,cAAc,EAAE,MAAM,EAAE,CAAC;IACzB,QAAQ,EAAE,MAAM,CAAC;CAClB,iBAoBA;AAyBD,wBAAsB,wBAAwB,CAAC,IAAI,EAAE,IAAI,iBA6HxD;AAED;;;GAGG;AACH,wBAAsB,oBAAoB,CACxC,OAAO,EAAE,MAAM,GACd,OAAO,CAAC,oBAAoB,CAAC,CA2B/B;AAWD,wBAAsB,oBAAoB,CACxC,gBAAgB,EAAE,oBAAoB,GACrC,OAAO,CAAC,MAAM,EAAE,CAAC,CAQnB;AAED;;;;;GAKG;AACH,wBAAsB,iBAAiB,CACrC,YAAY,EAAE,MAAM,EACpB,gBAAgB,EAAE,oBAAoB,EACtC,gBAAgB,GAAE,MAAM,EAAU,GACjC,OAAO,CAAC,MAAM,CAAC,CA+CjB"}
+{"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/utils.ts"],"names":[],"mappings":"AAGA,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAClC,OAAO,EAAE,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AAOvD,wBAAgB,QAAQ,CAAC,GAAG,EAAE,GAAG,GAAG,GAAG,IAAI,MAAM,CAKhD;AAED,wBAAgB,kBAAkB,CAAC,YAAY,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,QAmB1E;AAED,wBAAgB,yBAAyB,CACvC,YAAY,EAAE,MAAM,EACpB,OAAO,EAAE,MAAM,GACd,MAAM,GAAG,SAAS,CAsBpB;AAED,wBAAsB,wBAAwB,CAAC,IAAI,EAAE,IAAI,iBA0GxD;AAED;;;GAGG;AACH,wBAAsB,oBAAoB,CACxC,OAAO,EAAE,MAAM,GACd,OAAO,CAAC,oBAAoB,CAAC,CA2B/B;AAWD;;;;;GAKG;AACH,wBAAsB,iBAAiB,CACrC,YAAY,EAAE,MAAM,EACpB,gBAAgB,EAAE,oBAAoB,EACtC,gBAAgB,GAAE,MAAM,EAAU,GACjC,OAAO,CAAC,MAAM,CAAC,CA+CjB"}

package/dist/agent/browsing/utils.js

@@ -4,13 +4,10 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
 };
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.isRegExp = isRegExp;
-exports.prepareBrowsingAgentTask = prepareBrowsingAgentTask;
 exports.addImportForMethod = addImportForMethod;
 exports.replaceTodoWithCreateTest = replaceTodoWithCreateTest;
-exports.markTestAsOnly = markTestAsOnly;
 exports.injectPwLocatorGenerator = injectPwLocatorGenerator;
 exports.readPlaywrightConfig = readPlaywrightConfig;
-exports.getValidProjectNames = getValidProjectNames;
 exports.detectProjectName = detectProjectName;
 const fs_1 = __importDefault(require("fs"));
 const minimatch_1 = require("minimatch");
@@ -22,15 +19,15 @@ function isRegExp(obj) {
     return (obj instanceof RegExp ||
         Object.prototype.toString.call(obj) === "[object RegExp]");
 }
-function prepareBrowsingAgentTask(steps) {
-    const sanitizedSteps = steps.map((step) => step.replace(/`/g, "\\`"));
-    const task = `${sanitizedSteps.join("\n")}\n`;
-    return task;
-}
 function addImportForMethod(testFilePath, methodName) {
-    // Instead of using "@empiricalrun/test-gen", we use the local dist file
-    // This is to avoid assuming that the test-gen package is installed in the project
-    const importSource = path_1.default.join(__dirname, "../../../dist/index.js");
+    let importSource;
+    try {
+        importSource = require.resolve("@empiricalrun/test-gen");
+    }
+    catch {
+        // Fallback for bundled context (e.g. sandbox-wrapper)
+        importSource = path_1.default.join(__dirname, "index.js");
+    }
     if (!fs_1.default.existsSync(importSource)) {
         throw new Error(`createTest import source not found at ${importSource}`);
     }
@@ -50,50 +47,12 @@ function replaceTodoWithCreateTest(testFilePath, repoDir) {
     addImportForMethod(absoluteTestFilePath, "createTest");
     return todoContent;
 }
-async function markTestAsOnly({ testCaseName, testCaseSuites, specPath, }) {
-    const testFileContent = fs_1.default.readFileSync(specPath, "utf-8");
-    const { testBlock, testNode } = (0, web_1.getTypescriptTestBlock)({
-        scenarioName: testCaseName,
-        content: testFileContent,
-        suites: testCaseSuites,
-    });
-    const parentDescribe = (0, web_1.findFirstSerialDescribeBlock)(testNode);
-    const isFileMarkedSerial = await (0, web_1.hasTopLevelDescribeConfigureWithSerialMode)(specPath);
-    // add test.only / describe.only to the spec file so that only that block is executed
-    if (!isFileMarkedSerial) {
-        const updatedTestFileContent = newContentsWithTestOnly(testFileContent, testBlock, testBlock, parentDescribe?.getText() || "");
-        fs_1.default.writeFileSync(specPath, updatedTestFileContent);
-    }
-}
-function newContentsWithTestOnly(existingContents, originalTestBlock, updatedTestBlock, parentDescribeBlock) {
-    if (!parentDescribeBlock) {
-        const testMarkedAsOnly = updatedTestBlock.replace("test(", "test.only(");
-        return existingContents.replace(originalTestBlock, testMarkedAsOnly);
-    }
-    else {
-        const updatedDescribeBlock = parentDescribeBlock.replace(originalTestBlock, updatedTestBlock);
-        // TODO: this should only happen when the describe block has "serial" execution
-        const describeMarkedAsOnly = updatedDescribeBlock.replace("test.describe(", "test.describe.only(");
-        return existingContents.replace(parentDescribeBlock, describeMarkedAsOnly);
-    }
-}
 async function injectPwLocatorGenerator(page) {
-    let pathToInstalledTestGen = require.resolve(".").split("dist")[0];
-    if (process.env.RUNNING_BROWSER_TESTS_FOR_TEST_GEN) {
-        pathToInstalledTestGen = process.cwd();
-    }
-    const annotateElementPath = path_1.default.join(pathToInstalledTestGen, "dist", "browser-injected-scripts", "annotate-elements.js");
-    if (!fs_1.default.existsSync(annotateElementPath)) {
-        throw new Error(`annotate-elements.js not found at path: ${annotateElementPath}`);
-    }
     const remoteScriptResponses = await Promise.all([
         "https://assets-test.empirical.run/pw-selector.js",
         "https://code.jquery.com/jquery-3.7.1.min.js",
     ].map((url) => fetch(url)));
-    const scripts = await Promise.all([
-        ...remoteScriptResponses.map((r) => r.text()),
-        fs_1.default.readFileSync(annotateElementPath, "utf-8"),
-    ]);
+    const scripts = await Promise.all(remoteScriptResponses.map((r) => r.text()));
     page.on("load", async () => {
         try {
             await Promise.all(scripts.map((s) => page.addScriptTag({ content: s })));
@@ -219,15 +178,6 @@ function matchAgainstPattern(pattern, filePathToTest) {
         return (0, minimatch_1.minimatch)(filePathToTest, pattern);
     }
 }
-async function getValidProjectNames(playwrightConfig) {
-    if (!playwrightConfig.projects) {
-        return [];
-    }
-    const filteredProjectNames = playwrightConfig.projects
-        .map((p) => p.name)
-        .filter((p) => !!p);
-    return filteredProjectNames;
-}
 /**
  * detect the project name for the given file in playwright test repo
  * if project and test file path for running test don't match, then playwright throws error

package/dist/agent/chat/exports.d.ts

@@ -9,6 +9,6 @@ export { CodeReviewAgent } from "../code-review";
 export { FastTriageAgent } from "../fast-triage";
 export { TriageAgent } from "../triage";
 export { ChatAgent } from "./index";
-export { fetchToolCallFromId, getLatestDownloadBuildUrl, LATEST_CHAT_STATE_VERSION, } from "./state";
+export { fetchToolCallFromId, LATEST_CHAT_STATE_VERSION } from "./state";
 export { extractAttachments } from "./utils";
 //# sourceMappingURL=exports.d.ts.map

package/dist/agent/chat/exports.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"exports.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/exports.ts"],"names":[],"mappings":"AAAA,YAAY,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AACzD,OAAO,EAAE,eAAe,EAAE,MAAM,wBAAwB,CAAC;AACzD,OAAO,EAAE,qBAAqB,EAAE,MAAM,kCAAkC,CAAC;AACzE,OAAO,EACL,qBAAqB,EACrB,mBAAmB,GACpB,MAAM,iCAAiC,CAAC;AAEzC,YAAY,EAAE,WAAW,EAAE,kBAAkB,EAAE,MAAM,SAAS,CAAC;AAC/D,OAAO,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AACpC,YAAY,EACV,kBAAkB,EAClB,kBAAkB,EAClB,kBAAkB,EAClB,yBAAyB,GAC1B,MAAM,gBAAgB,CAAC;AACxB,OAAO,EAAE,eAAe,EAAE,MAAM,gBAAgB,CAAC;AACjD,OAAO,EAAE,eAAe,EAAE,MAAM,gBAAgB,CAAC;AACjD,OAAO,EAAE,WAAW,EAAE,MAAM,WAAW,CAAC;AACxC,OAAO,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AACpC,OAAO,EACL,mBAAmB,EACnB,yBAAyB,EACzB,yBAAyB,GAC1B,MAAM,SAAS,CAAC;AACjB,OAAO,EAAE,kBAAkB,EAAE,MAAM,SAAS,CAAC"}
+{"version":3,"file":"exports.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/exports.ts"],"names":[],"mappings":"AAAA,YAAY,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AACzD,OAAO,EAAE,eAAe,EAAE,MAAM,wBAAwB,CAAC;AACzD,OAAO,EAAE,qBAAqB,EAAE,MAAM,kCAAkC,CAAC;AACzE,OAAO,EACL,qBAAqB,EACrB,mBAAmB,GACpB,MAAM,iCAAiC,CAAC;AAEzC,YAAY,EAAE,WAAW,EAAE,kBAAkB,EAAE,MAAM,SAAS,CAAC;AAC/D,OAAO,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AACpC,YAAY,EACV,kBAAkB,EAClB,kBAAkB,EAClB,kBAAkB,EAClB,yBAAyB,GAC1B,MAAM,gBAAgB,CAAC;AACxB,OAAO,EAAE,eAAe,EAAE,MAAM,gBAAgB,CAAC;AACjD,OAAO,EAAE,eAAe,EAAE,MAAM,gBAAgB,CAAC;AACjD,OAAO,EAAE,WAAW,EAAE,MAAM,WAAW,CAAC;AACxC,OAAO,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AACpC,OAAO,EAAE,mBAAmB,EAAE,yBAAyB,EAAE,MAAM,SAAS,CAAC;AACzE,OAAO,EAAE,kBAAkB,EAAE,MAAM,SAAS,CAAC"}

package/dist/agent/chat/exports.js

@@ -1,6 +1,6 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.extractAttachments = exports.LATEST_CHAT_STATE_VERSION = exports.getLatestDownloadBuildUrl = exports.fetchToolCallFromId = exports.ChatAgent = exports.TriageAgent = exports.FastTriageAgent = exports.CodeReviewAgent = exports.BaseAgent = exports.viewFileUsingGitHub = exports.getFileInfoFromGitHub = exports.SUPPORTED_CHAT_MODELS = exports.createChatModel = void 0;
+exports.extractAttachments = exports.LATEST_CHAT_STATE_VERSION = exports.fetchToolCallFromId = exports.ChatAgent = exports.TriageAgent = exports.FastTriageAgent = exports.CodeReviewAgent = exports.BaseAgent = exports.viewFileUsingGitHub = exports.getFileInfoFromGitHub = exports.SUPPORTED_CHAT_MODELS = exports.createChatModel = void 0;
 var chat_1 = require("@empiricalrun/llm/chat");
 Object.defineProperty(exports, "createChatModel", { enumerable: true, get: function () { return chat_1.createChatModel; } });
 var constants_1 = require("@empiricalrun/llm/chat/constants");
@@ -20,7 +20,6 @@ var index_1 = require("./index");
 Object.defineProperty(exports, "ChatAgent", { enumerable: true, get: function () { return index_1.ChatAgent; } });
 var state_1 = require("./state");
 Object.defineProperty(exports, "fetchToolCallFromId", { enumerable: true, get: function () { return state_1.fetchToolCallFromId; } });
-Object.defineProperty(exports, "getLatestDownloadBuildUrl", { enumerable: true, get: function () { return state_1.getLatestDownloadBuildUrl; } });
 Object.defineProperty(exports, "LATEST_CHAT_STATE_VERSION", { enumerable: true, get: function () { return state_1.LATEST_CHAT_STATE_VERSION; } });
 var utils_1 = require("./utils");
 Object.defineProperty(exports, "extractAttachments", { enumerable: true, get: function () { return utils_1.extractAttachments; } });

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

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,uCAAuC,CAAC;AACzE,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,qCAAqC,CAAC;AAsBpE,OAAO,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AAMpC,qBAAa,SAAS,CAAC,CAAC,CAAE,SAAQ,SAAS,CAAC,CAAC,CAAC;IAC5C,SAAS,CAAC,QAAQ,IAAI,WAAW;IA0B3B,iBAAiB,CACrB,eAAe,EAAE,MAAM,OAAO,CAAC,QAAQ,CAAC,GACvC,OAAO,CAAC,MAAM,CAAC;CAyGnB"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,uCAAuC,CAAC;AACzE,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,qCAAqC,CAAC;AAsBpE,OAAO,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AAMpC,qBAAa,SAAS,CAAC,CAAC,CAAE,SAAQ,SAAS,CAAC,CAAC,CAAC;IAC5C,SAAS,CAAC,QAAQ,IAAI,WAAW;IA2B3B,iBAAiB,CACrB,eAAe,EAAE,MAAM,OAAO,CAAC,QAAQ,CAAC,GACvC,OAAO,CAAC,MAAM,CAAC;CAwGnB"}

package/dist/agent/chat/index.js

@@ -2,8 +2,8 @@
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.ChatAgent = void 0;
 const tools_1 = require("../../tools");
+const api_client_1 = require("../../tools/api-client");
 const analyse_video_1 = require("../../tools/definitions/analyse-video");
-const download_build_1 = require("../../tools/definitions/download-build");
 const grep_1 = require("../../tools/definitions/grep");
 const list_tests_and_projects_1 = require("../../tools/definitions/list-tests-and-projects");
 const run_test_1 = require("../../tools/definitions/run-test");
@@ -18,17 +18,18 @@ const test_case_def_1 = require("./prompt/test-case-def");
 const trace_utils_docs_1 = require("./prompt/trace-utils-docs");
 class ChatAgent extends base_1.BaseAgent {
     getTools() {
+        const useAPIClientTool = this.featureFlags.includes("useAPIClientTool");
         const custom = [
             analyse_video_1.analyseVideo,
             run_test_1.runTestTool,
             grep_1.grepTool,
-            list_environments_1.listEnvironmentsTool,
-            download_build_1.downloadBuildTool,
+            ...(useAPIClientTool
+                ? [api_client_1.apiClientTool]
+                : [list_environments_1.listEnvironmentsTool, test_run_fetcher_1.fetchTestRunDetailsTool]),
             fetch_file_1.fetchFileTool,
             list_tests_and_projects_1.listProjectsTool,
             list_tests_and_projects_1.listTestsForProjectTool,
             ...tools_1.testGenerationTools,
-            test_run_fetcher_1.fetchTestRunDetailsTool,
             diagnosis_fetcher_1.fetchDiagnosisReportTool,
             ...(0, tools_1.textEditorToolsForModel)(this.selectedModel),
             tools_1.safeBashTool,
@@ -45,7 +46,6 @@ class ChatAgent extends base_1.BaseAgent {
     async buildSystemPrompt(repoInfoBuilder) {
         const repoInfo = await repoInfoBuilder();
         const repoContext = await (0, repo_1.getRepoContextPrompt)(repoInfo);
-        const hasTraceUtils = this.featureFlags.includes("enableTraceUtilsViaBash");
         const preamble = `
 You are a helpful assistant that can answer questions and help with tasks related to writing and maintaining Playwright tests.
 

package/dist/agent/chat/prompt/repo.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"repo.d.ts","sourceRoot":"","sources":["../../../../src/agent/chat/prompt/repo.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,qCAAqC,CAAC;AAuD/D,wBAAsB,oBAAoB,CAAC,QAAQ,EAAE,QAAQ,mBA4B5D"}
+{"version":3,"file":"repo.d.ts","sourceRoot":"","sources":["../../../../src/agent/chat/prompt/repo.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,qCAAqC,CAAC;AAoF/D,wBAAsB,oBAAoB,CAAC,QAAQ,EAAE,QAAQ,mBA8B5D"}

package/dist/agent/chat/prompt/repo.js

@@ -6,6 +6,28 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.getRepoContextPrompt = getRepoContextPrompt;
 const path_1 = __importDefault(require("path"));
 const repo_tree_1 = require("../../../utils/repo-tree");
+async function environmentsYamlSection(directory) {
+    if (directory.type !== "directory") {
+        return "";
+    }
+    const empiricalrunDir = directory.children.find((child) => child.type === "directory" && child.path === ".empiricalrun");
+    if (!empiricalrunDir || empiricalrunDir.type !== "directory") {
+        return "";
+    }
+    const envFile = empiricalrunDir.children.find((child) => child.type === "file" &&
+        path_1.default.basename(child.path) === "ENVIRONMENTS.yaml");
+    if (!envFile || envFile.type !== "file") {
+        return "";
+    }
+    const content = await envFile.getContent();
+    return `## Environments
+
+The repo is configured to run tests for these environments with .empiricalrun/ENVIRONMENTS.yaml (see docs: https://docs.empirical.run/agents/env-vars)
+
+\`\`\`yaml
+${content}
+\`\`\``;
+}
 async function getAllMarkdownFiles(directory) {
     if (directory.type !== "directory") {
         return [];
@@ -60,6 +82,8 @@ While specifying paths to files, prefer using absolute paths with the /repo/ pre
 - Preferred: "/repo/tests/lesson.spec.ts" (absolute path with /repo/ prefix)
 - Also supported: "tests/lesson.spec.ts" (relative path from repository root)
 - Incorrect: "/${repoInfo.name}/tests/lesson.spec.ts" (wrong prefix)
+
+${await environmentsYamlSection(repoInfo)}
 `;
     const knowledge = await knowledgeContext(repoInfo);
     if (knowledge.length > 0) {

package/dist/agent/chat/state.d.ts

@@ -18,7 +18,6 @@ export declare function loadChatState<T>({ resetChat, }: {
     resetChat: boolean;
 }): ChatStateGeneric<T> | undefined;
 export declare function saveToDisk<T>(chatState: ChatStateGeneric<T>): void;
-export declare function getLatestDownloadBuildUrl(messages: CanonicalMessage[]): string | null;
 export declare function fetchToolCallFromId(toolCallId: String, messages: CanonicalMessage[]): {
     toolCallQueuedAt: Date | undefined;
     toolCall: PendingToolCall | undefined;

package/dist/agent/chat/state.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"state.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/state.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AACpD,OAAO,EACL,gBAAgB,EAChB,cAAc,EACd,gBAAgB,EAChB,eAAe,EAEf,UAAU,EACX,MAAM,uCAAuC,CAAC;AAM/C,eAAO,MAAM,kCAAkC,EAAE,MAAM,CACrD,MAAM,EACN,CAAC,KAAK,EAAE,GAAG,KAAK,GAAG,CAGpB,CAAC;AAEF,eAAO,MAAM,yBAAyB,QAAQ,CAAC;AAE/C,eAAO,MAAM,eAAe,QAI3B,CAAC;AAEF,wBAAgB,kBAAkB,CAAC,CAAC,EAAE,EACpC,SAAS,EACT,KAAK,GACN,EAAE;IACD,SAAS,EAAE,UAAU,CAAC,CAAC,CAAC,CAAC;IACzB,KAAK,EAAE,cAAc,GAAG,IAAI,CAAC;CAC9B;;;;;;EAQA;AAED,wBAAgB,eAAe,CAAC,CAAC,GAAG,gBAAgB,EAClD,SAAS,EAAE,gBAAgB,CAAC,CAAC,CAAC,GAC7B,MAAM,CA4BR;AAED,wBAAgB,aAAa,CAAC,CAAC,EAAE,EAC/B,SAAS,GACV,EAAE;IACD,SAAS,EAAE,OAAO,CAAC;CACpB,GAAG,gBAAgB,CAAC,CAAC,CAAC,GAAG,SAAS,CAoBlC;AAED,wBAAgB,UAAU,CAAC,CAAC,EAAE,SAAS,EAAE,gBAAgB,CAAC,CAAC,CAAC,QAQ3D;AA2BD,wBAAgB,yBAAyB,CACvC,QAAQ,EAAE,gBAAgB,EAAE,GAC3B,MAAM,GAAG,IAAI,CAef;AAED,wBAAgB,mBAAmB,CACjC,UAAU,EAAE,MAAM,EAClB,QAAQ,EAAE,gBAAgB,EAAE,GAC3B;IACD,gBAAgB,EAAE,IAAI,GAAG,SAAS,CAAC;IACnC,QAAQ,EAAE,eAAe,GAAG,SAAS,CAAC;IACtC,UAAU,EAAE,UAAU,GAAG,SAAS,CAAC;CACpC,CA0BA"}
+{"version":3,"file":"state.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/state.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AACpD,OAAO,EACL,gBAAgB,EAChB,cAAc,EACd,gBAAgB,EAChB,eAAe,EAEf,UAAU,EACX,MAAM,uCAAuC,CAAC;AAI/C,eAAO,MAAM,kCAAkC,EAAE,MAAM,CACrD,MAAM,EACN,CAAC,KAAK,EAAE,GAAG,KAAK,GAAG,CAGpB,CAAC;AAEF,eAAO,MAAM,yBAAyB,QAAQ,CAAC;AAE/C,eAAO,MAAM,eAAe,QAI3B,CAAC;AAEF,wBAAgB,kBAAkB,CAAC,CAAC,EAAE,EACpC,SAAS,EACT,KAAK,GACN,EAAE;IACD,SAAS,EAAE,UAAU,CAAC,CAAC,CAAC,CAAC;IACzB,KAAK,EAAE,cAAc,GAAG,IAAI,CAAC;CAC9B;;;;;;EAQA;AAED,wBAAgB,eAAe,CAAC,CAAC,GAAG,gBAAgB,EAClD,SAAS,EAAE,gBAAgB,CAAC,CAAC,CAAC,GAC7B,MAAM,CA4BR;AAED,wBAAgB,aAAa,CAAC,CAAC,EAAE,EAC/B,SAAS,GACV,EAAE;IACD,SAAS,EAAE,OAAO,CAAC;CACpB,GAAG,gBAAgB,CAAC,CAAC,CAAC,GAAG,SAAS,CAoBlC;AAED,wBAAgB,UAAU,CAAC,CAAC,EAAE,SAAS,EAAE,gBAAgB,CAAC,CAAC,CAAC,QAQ3D;AAED,wBAAgB,mBAAmB,CACjC,UAAU,EAAE,MAAM,EAClB,QAAQ,EAAE,gBAAgB,EAAE,GAC3B;IACD,gBAAgB,EAAE,IAAI,GAAG,SAAS,CAAC;IACnC,QAAQ,EAAE,eAAe,GAAG,SAAS,CAAC;IACtC,UAAU,EAAE,UAAU,GAAG,SAAS,CAAC;CACpC,CA0BA"}

package/dist/agent/chat/state.js

@@ -8,7 +8,6 @@ exports.chatStateFromModel = chatStateFromModel;
 exports.getUsageSummary = getUsageSummary;
 exports.loadChatState = loadChatState;
 exports.saveToDisk = saveToDisk;
-exports.getLatestDownloadBuildUrl = getLatestDownloadBuildUrl;
 exports.fetchToolCallFromId = fetchToolCallFromId;
 const fs_1 = __importDefault(require("fs"));
 const path_1 = __importDefault(require("path"));
@@ -77,33 +76,6 @@ function saveToDisk(chatState) {
     }
     fs_1.default.writeFileSync(statePath, JSON.stringify(chatState, null, 2));
 }
-function findLatestToolCall(messages, toolName) {
-    return messages
-        .sort((a, b) => new Date(b.timestamp).getTime() - new Date(a.timestamp).getTime())
-        .find((m) => m.parts.some((p) => "toolCall" in p && p.toolCall?.name === toolName));
-}
-function findSuccessfulToolResponse(messages, toolCallId) {
-    return messages.find((m) => m.parts.some((p) => {
-        if ("toolResult" in p) {
-            return p.toolCallId === toolCallId && !p.toolResult.isError;
-        }
-        return false;
-    }));
-}
-function getLatestDownloadBuildUrl(messages) {
-    const toolCallMessage = findLatestToolCall(messages, "downloadBuild");
-    if (!toolCallMessage)
-        return null;
-    const toolCallPart = toolCallMessage.parts.find((p) => "toolCall" in p && p.toolCall?.name === "downloadBuild");
-    if (!toolCallPart || !("toolCall" in toolCallPart))
-        return null;
-    const toolCallId = toolCallPart.toolCall.id;
-    const toolResponseMessage = findSuccessfulToolResponse(messages, toolCallId);
-    if (!toolResponseMessage)
-        return null;
-    const input = toolCallPart.toolCall.input;
-    return input.buildUrl;
-}
 function fetchToolCallFromId(toolCallId, messages) {
     const toolCallMessage = messages.find((message) => message.parts.some((part) => {
         return "toolCall" in part && part.toolCallId === toolCallId;

package/dist/agent/master/browser-tests/cua.spec.js

@@ -1,10 +1,10 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
-const run_1 = require("../run");
+const cua_1 = require("@empiricalrun/cua");
 const fixtures_1 = require("./fixtures");
 fixtures_1.test.skip("computer use agent generates code for iframes", async ({ page, server, }) => {
     await page.goto(`${server.baseURL}/iframe-elements.html`);
-    const result = await (0, run_1.createTestUsingComputerUseAgent)({
+    const result = await (0, cua_1.createTestUsingComputerUseAgent)({
         page,
         task: `Click on both of the "Enter your name" input fields`,
     });
@@ -13,7 +13,7 @@ fixtures_1.test.skip("computer use agent generates code for iframes", async ({ p
 });
 fixtures_1.test.skip("computer use agent generates code for google login", async ({ page, }) => {
     await page.goto(`https://dash.empirical.run`);
-    const result = await (0, run_1.createTestUsingComputerUseAgent)({
+    const result = await (0, cua_1.createTestUsingComputerUseAgent)({
         page,
         task: `Click on the "Sign in with Google" button and enter dpdzero-test-user@empirical.run
 as the email and click continue. then enter password as 1234556789`,

package/dist/agent/master/browser-tests/index.spec.js

@@ -1,81 +1,11 @@
 "use strict";
-var __importDefault = (this && this.__importDefault) || function (mod) {
-    return (mod && mod.__esModule) ? mod : { "default": mod };
-};
 Object.defineProperty(exports, "__esModule", { value: true });
-const fs_1 = __importDefault(require("fs"));
-const path_1 = __importDefault(require("path"));
-const utils_1 = require("../../browsing/utils");
-const element_annotation_1 = require("../element-annotation");
-const run_1 = require("../run");
+const cua_1 = require("@empiricalrun/cua");
 const fixtures_1 = require("./fixtures");
-fixtures_1.test.skip("able to scroll and interact with elements", async ({ page, server, }) => {
-    await page.goto(`${server.baseURL}/blog-page.html`);
-    const response = await (0, run_1.createTestUsingMasterAgent)({
-        task: `fill test@test.com into the email field and click the submit`,
-        page,
-        options: {},
-    });
-    console.log(response);
-    (0, fixtures_1.expect)(response.importPaths.length).toBe(0);
-    (0, fixtures_1.expect)(response.code).toContain("await page.getByPlaceholder('Enter your email').fill(\"test@test.com\")");
-    (0, fixtures_1.expect)(response.code).toContain("await page.getByRole('button', { name: 'Subscribe' }).click()");
-});
-fixtures_1.test.skip("scroll when element does not exist", async ({ page, server }) => {
-    await page.goto(`${server.baseURL}/blog-page.html`);
-    const response = await (0, run_1.createTestUsingMasterAgent)({
-        task: `click search button`,
-        page,
-        options: {},
-    });
-    console.log(response);
-    (0, fixtures_1.expect)(response.importPaths.length).toBe(0);
-    (0, fixtures_1.expect)(response.code.length).toBe(0);
-});
-fixtures_1.test.skip("scroll and click inside div elements", async ({ page, server }) => {
-    await page.goto(`${server.baseURL}/dropdown-scrolls.html`);
-    const response = await (0, run_1.createTestUsingMasterAgent)({
-        task: `click on x-3 inside bmw dropdown, and then,
-click on maverick inside ford dropdown`,
-        page,
-        options: {},
-    });
-    console.log(response);
-    (0, fixtures_1.expect)(response.importPaths.length).toBe(0);
-    (0, fixtures_1.expect)(response.code.length).toBeGreaterThan(0);
-    const lines = response.code.split("\n");
-    (0, fixtures_1.expect)(lines.find((l) => l.match(/^await page.+BMW.+.click/))).toBeTruthy();
-    (0, fixtures_1.expect)(lines.find((l) => l.match(/^await page.+X3.+.click/))).toBeTruthy();
-    (0, fixtures_1.expect)(lines.find((l) => l.match(/^await page.+Ford.+.click/))).toBeTruthy();
-    (0, fixtures_1.expect)(lines.find((l) => l.match(/^await page.+Maverick.+.click/))).toBeTruthy();
-});
-fixtures_1.test.skip("master agent can click icons accurately", async ({ page, server, }) => {
-    await page.goto(`${server.baseURL}/icons-navbar.html`);
-    await (0, fixtures_1.expect)(page.getByText("select an icon")).toBeVisible();
-    const response = await (0, run_1.createTestUsingMasterAgent)({
-        task: `click on the gear icon`,
-        page,
-        options: {},
-    });
-    // Validate code generated and action performed
-    await (0, fixtures_1.expect)(page.getByText("you clicked Gear")).toBeVisible();
-    console.log(response);
-    (0, fixtures_1.expect)(response.importPaths.length).toBe(0);
-    (0, fixtures_1.expect)(response.code).toContain("page.locator");
-    (0, fixtures_1.expect)(response.code).toContain("click()");
-    // Validate icons registry
-    const iconsRegistryFile = path_1.default.join(process.cwd(), ".empiricalrun", "icons.json");
-    const icons = JSON.parse(fs_1.default.readFileSync(iconsRegistryFile, "utf-8"));
-    (0, fixtures_1.expect)(icons.length).toBeGreaterThan(0);
-    // Commenting out this check since with parallel test executions,
-    // we can have icons from other tests in the same file
-    // expect(icons.length).toBe(4); // 1 for each unique icon
-    fs_1.default.unlinkSync(iconsRegistryFile);
-});
 fixtures_1.test.skip("cua agent can click icons accurately", async ({ page, server }) => {
     await page.goto(`${server.baseURL}/icons-navbar.html`);
     await (0, fixtures_1.expect)(page.getByText("select an icon")).toBeVisible();
-    const response = await (0, run_1.createTestUsingComputerUseAgent)({
+    const response = await (0, cua_1.createTestUsingComputerUseAgent)({
         task: `click on the gear icon`,
         page,
     });
@@ -86,42 +16,3 @@ fixtures_1.test.skip("cua agent can click icons accurately", async ({ page, serv
     (0, fixtures_1.expect)(response.code).toContain("page.locator");
     (0, fixtures_1.expect)(response.code).toContain("click()");
 });
-fixtures_1.test.skip("annotate and enrich annotations correctly", async ({ page, server, }) => {
-    await (0, utils_1.injectPwLocatorGenerator)(page);
-    await page.goto(`${server.baseURL}/iframe-elements.html`);
-    const { annotationKeys: keys } = await (0, element_annotation_1.getAnnotationKeys)({
-        page,
-        preference: {
-            actionType: "all",
-        },
-    });
-    console.log(keys);
-    (0, fixtures_1.expect)(keys.length).toBe(6);
-    // 2 icons: 1 in main frame, 1 in iframe
-    (0, fixtures_1.expect)(keys.filter((k) => k.text.includes("icon") && k.text.includes("close"))
-        .length).toBe(2);
-    // 2 text inputs: 1 in main frame, 1 in iframe
-    (0, fixtures_1.expect)(keys.filter((k) => k.text.includes("Enter your name")).length).toBe(2);
-    // 2 clickable divs: 1 in main frame, 1 in iframe
-    (0, fixtures_1.expect)(keys.filter((k) => k.text.includes("Lorem Ipsum")).length).toBe(2);
-});
-fixtures_1.test.skip("fill action with multiple pages", async ({ context }) => {
-    const page1 = await context.newPage();
-    const page2 = await context.newPage();
-    const response = await (0, run_1.createTestUsingMasterAgent)({
-        task: `goto empirical.run on page1 and goto github.com/search in page2. Enter text empirical on page2 and click on search.`,
-        page: page2,
-        options: {},
-        scopeVars: {
-            context,
-            page1,
-            page2,
-        },
-    });
-    (0, fixtures_1.expect)(response.code).toBeTruthy();
-    console.log(response.code);
-    const lines = response.code.split("\n");
-    (0, fixtures_1.expect)(lines.find((l) => l.match(/^await page1\.goto(.+)empirical\.run/))).toBeTruthy();
-    (0, fixtures_1.expect)(lines.find((l) => l.match(/^await page2\.goto(.+)github\.com\/search/))).toBeTruthy();
-    (0, fixtures_1.expect)(lines.find((l) => l.match(/^await page2(.+)fill(.+)empirical/))).toBeTruthy();
-});

package/dist/agent/master/run.d.ts

@@ -2,7 +2,6 @@ import type { TestGenConfigOptions } from "@empiricalrun/shared-types/agent-work
 import type { TestCaseWithSuitesAsArray } from "@empiricalrun/shared-types/api/test-cases";
 import { Page } from "playwright";
 import { ScopeVars } from "../../types";
-export { createTestUsingComputerUseAgent } from "../cua";
 export declare const IS_ALLOWED_TO_USE_SKILLS = false;
 export declare function createTestUsingMasterAgent({ task, page, testCase, specPath, options, scopeVars, }: {
     task: string;

package/dist/agent/master/run.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/agent/master/run.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,2CAA2C,CAAC;AACtF,OAAO,KAAK,EAAE,yBAAyB,EAAE,MAAM,2CAA2C,CAAC;AAC3F,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAYlC,OAAO,EAAE,SAAS,EAAE,MAAM,aAAa,CAAC;AAQxC,OAAO,EAAE,+BAA+B,EAAE,MAAM,QAAQ,CAAC;AAKzD,eAAO,MAAM,wBAAwB,QAAQ,CAAC;AAqB9C,wBAAsB,0BAA0B,CAAC,EAC/C,IAAI,EACJ,IAAI,EACJ,QAAQ,EACR,QAAQ,EACR,OAAO,EACP,SAAS,GACV,EAAE;IACD,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,IAAI,CAAC;IACX,QAAQ,CAAC,EAAE,yBAAyB,CAAC;IACrC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,OAAO,CAAC,oBAAoB,CAAC,CAAC;IACvC,SAAS,CAAC,EAAE,SAAS,CAAC;CACvB;;;GAwJA"}
+{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/agent/master/run.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,2CAA2C,CAAC;AACtF,OAAO,KAAK,EAAE,yBAAyB,EAAE,MAAM,2CAA2C,CAAC;AAC3F,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAYlC,OAAO,EAAE,SAAS,EAAE,MAAM,aAAa,CAAC;AAWxC,eAAO,MAAM,wBAAwB,QAAQ,CAAC;AAqB9C,wBAAsB,0BAA0B,CAAC,EAC/C,IAAI,EACJ,IAAI,EACJ,QAAQ,EACR,QAAQ,EACR,OAAO,EACP,SAAS,GACV,EAAE;IACD,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,IAAI,CAAC;IACX,QAAQ,CAAC,EAAE,yBAAyB,CAAC;IACrC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,OAAO,CAAC,oBAAoB,CAAC,CAAC;IACvC,SAAS,CAAC,EAAE,SAAS,CAAC;CACvB;;;GAwJA"}