@empiricalrun/test-gen 0.51.4 → 0.51.5

package/CHANGELOG.md

@@ -1,5 +1,16 @@
 # @empiricalrun/test-gen
 
+## 0.51.5
+
+### Patch Changes
+
+- ae47eed: feat: add type checking to text editor tool call
+- beb704e: chore: move claude chat stuff to make space for gemini
+- 3d9c868: fix: system prompt for computer use agent
+- Updated dependencies [beb704e]
+- Updated dependencies [f283696]
+  - @empiricalrun/llm@0.11.4
+
 ## 0.51.4
 
 ### Patch Changes

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

@@ -1 +1 @@
-{"version":3,"file":"chat.d.ts","sourceRoot":"","sources":["../../src/agent/chat.ts"],"names":[],"mappings":"AA6FA,wBAAsB,SAAS,CAAC,EAC9B,SAAwC,EACxC,mBAAmB,GACpB,EAAE;IACD,SAAS,CAAC,EAAE,4BAA4B,GAAG,4BAA4B,CAAC;IACxE,mBAAmB,CAAC,EAAE,OAAO,CAAC;CAC/B,mBAkHA"}
+{"version":3,"file":"chat.d.ts","sourceRoot":"","sources":["../../src/agent/chat.ts"],"names":[],"mappings":"AA+FA,wBAAsB,SAAS,CAAC,EAC9B,SAAwC,EACxC,mBAAmB,GACpB,EAAE;IACD,SAAS,CAAC,EAAE,4BAA4B,GAAG,4BAA4B,CAAC;IACxE,mBAAmB,CAAC,EAAE,OAAO,CAAC;CAC/B,mBAkHA"}

package/dist/agent/chat.js

@@ -4,9 +4,10 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
 };
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.chatAgent = void 0;
-const claude_1 = require("@empiricalrun/llm/claude");
+const chat_1 = require("@empiricalrun/llm/chat");
 const path_1 = __importDefault(require("path"));
 const picocolors_1 = require("picocolors");
+const web_1 = require("../bin/utils/platform/web");
 const human_in_the_loop_1 = require("../human-in-the-loop");
 const diagnosis_fetcher_1 = require("../tools/diagnosis-fetcher");
 const grep_1 = require("../tools/grep");
@@ -84,12 +85,12 @@ const tools = [
 ];
 const toolExecutors = {
     ...Object.fromEntries(tools.map((tool) => [tool.schema.name, tool.execute])),
-    str_replace_editor: claude_1.strReplaceEditorTool,
+    str_replace_editor: (input) => (0, chat_1.strReplaceEditorTool)(input, web_1.validateTypescript),
 };
 async function chatAgent({ chatModel = "claude-3-7-sonnet-20250219", useDiskForChatState, }) {
     const ora = (await import("ora")).default;
     let userPrompt = undefined;
-    let chatState = useDiskForChatState ? claude_1.ChatState.load() : new claude_1.ChatState(false);
+    let chatState = useDiskForChatState ? chat_1.ChatState.load() : new chat_1.ChatState(false);
     const handleSigInt = () => {
         console.log(`\n${(0, picocolors_1.gray)("Usage summary -> " + chatState.getUsageSummary())}`);
         process.exit(0);
@@ -163,10 +164,10 @@ async function chatAgent({ chatModel = "claude-3-7-sonnet-20250219", useDiskForC
             continue;
         }
         const spinner = ora("Claude is working...").start();
-        const response = await (0, claude_1.createClaudeMessage)({
+        const response = await (0, chat_1.createClaudeMessage)({
             systemPrompt,
             messages: chatState.getMessagesForCreateCompletion(),
-            tools: tools.map((tool) => (0, claude_1.convertOpenAISchemaToAnthropic)((0, zod_schema_1.zodToOpenAITool)(tool.schema))),
+            tools: tools.map((tool) => (0, chat_1.convertOpenAISchemaToAnthropic)((0, zod_schema_1.zodToOpenAITool)(tool.schema))),
             model: chatModel,
             withStrReplaceEditor: true,
         });

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":"AAEA,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAWlC,wBAAsB,sBAAsB,CAAC,IAAI,EAAE,IAAI,iBAoBtD;AAED;;GAEG;AACH,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;CACvB,CAAC,CAkGD"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/cua/index.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAWlC,wBAAsB,sBAAsB,CAAC,IAAI,EAAE,IAAI,iBAoBtD;AAED;;GAEG;AACH,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;CACvB,CAAC,CAmGD"}

package/dist/agent/cua/index.js

@@ -110,6 +110,7 @@ async function createTestUsingComputerUseAgent({ page, task, }) {
                     environment: "browser",
                 },
             ],
+            instructions: INSTRUCTIONS,
             input: [
                 {
                     call_id: lastCallId,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@empiricalrun/test-gen",
-  "version": "0.51.4",
+  "version": "0.51.5",
   "publishConfig": {
     "registry": "https://registry.npmjs.org/",
     "access": "public"
@@ -74,7 +74,7 @@
     "tsx": "^4.16.2",
     "typescript": "^5.3.3",
     "zod": "^3.23.8",
-    "@empiricalrun/llm": "^0.11.3",
+    "@empiricalrun/llm": "^0.11.4",
     "@empiricalrun/r2-uploader": "^0.3.8",
     "@empiricalrun/reporter": "^0.23.2",
     "@empiricalrun/test-run": "^0.7.6"