@empiricalrun/test-gen 0.52.2 → 0.52.4

package/CHANGELOG.md

@@ -1,5 +1,25 @@
 # @empiricalrun/test-gen
 
+## 0.52.4
+
+### Patch Changes
+
+- 85665ef: feat: simplify text editor tool schemas for gemini
+- Updated dependencies [4f8137b]
+- Updated dependencies [85665ef]
+  - @empiricalrun/llm@0.13.2
+
+## 0.52.3
+
+### Patch Changes
+
+- 6a19298: feat: changed gemini-2.5pro from exp to preview
+- cbe5823: fix: removed headed default from runTest tool schema and added function in role of toolRes Gemini
+- Updated dependencies [f4f4c5d]
+- Updated dependencies [6a19298]
+- Updated dependencies [cbe5823]
+  - @empiricalrun/llm@0.13.1
+
 ## 0.52.2
 
 ### Patch Changes

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

@@ -1,5 +1,5 @@
 export declare function chatAgent({ selectedModel, useDiskForChatState, initialPromptContent, }: {
-    selectedModel?: "claude-3-7-sonnet-20250219" | "claude-3-5-sonnet-20241022" | "gemini-2.5-pro-exp-03-25";
+    selectedModel?: "claude-3-7-sonnet-20250219" | "claude-3-5-sonnet-20241022" | "gemini-2.5-pro-preview-03-25";
     useDiskForChatState?: boolean;
     initialPromptContent?: string;
 }): Promise<string>;

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":"AAgEA,wBAAsB,SAAS,CAAC,EAC9B,aAA4C,EAC5C,mBAA2B,EAC3B,oBAAoB,GACrB,EAAE;IACD,aAAa,CAAC,EACV,4BAA4B,GAC5B,4BAA4B,GAC5B,0BAA0B,CAAC;IAC/B,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B,oBAAoB,CAAC,EAAE,MAAM,CAAC;CAC/B,mBAsFA"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/index.ts"],"names":[],"mappings":"AAiFA,wBAAsB,SAAS,CAAC,EAC9B,aAA4C,EAC5C,mBAA2B,EAC3B,oBAAoB,GACrB,EAAE;IACD,aAAa,CAAC,EACV,4BAA4B,GAC5B,4BAA4B,GAC5B,8BAA8B,CAAC;IACnC,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B,oBAAoB,CAAC,EAAE,MAAM,CAAC;CAC/B,mBAuFA"}

package/dist/agent/chat/index.js

@@ -11,18 +11,27 @@ const test_gen_browser_1 = require("../../tools/test-gen-browser");
 const test_run_1 = require("../../tools/test-run");
 const test_run_fetcher_1 = require("../../tools/test-run-fetcher");
 const prompt_1 = require("./prompt");
-// TODO: Add strReplaceEditor for non-Claude models
-const tools = [
-    test_run_1.runTestTool,
-    test_gen_browser_1.generateTestWithBrowserAgent,
-    diagnosis_fetcher_1.diagnosisTool,
-    grep_1.grepTool,
-    test_run_fetcher_1.testRunTool,
-];
-const toolExecutors = {
-    ...Object.fromEntries(tools.map((tool) => [tool.schema.name, tool.execute])),
-    str_replace_editor: (input) => (0, chat_1.strReplaceEditorExecutor)(input, web_1.validateTypescript),
-};
+function getTools(selectedModel) {
+    let tools = [
+        grep_1.grepTool,
+        test_run_1.runTestTool,
+        test_run_fetcher_1.fetchTestRunReportTool,
+        diagnosis_fetcher_1.fetchDiagnosisReportTool,
+        test_gen_browser_1.generateTestWithBrowserAgent,
+    ];
+    if (selectedModel.startsWith("gemini")) {
+        // Claude will have its own built-in text editor tools
+        tools.push(...chat_1.textEditorTools);
+    }
+    const toolExecutors = {
+        // TODO: Add validateTypescript
+        ...Object.fromEntries(tools.map((tool) => [tool.schema.name, tool.execute])),
+    };
+    if (selectedModel.startsWith("claude")) {
+        toolExecutors.str_replace_editor = (input) => (0, chat_1.strReplaceEditorExecutor)(input, web_1.validateTypescript);
+    }
+    return { tools, toolExecutors };
+}
 function createChatModel(useDiskForChatState, selectedModel) {
     if (selectedModel.startsWith("claude")) {
         return new chat_1.ClaudeChatModel(useDiskForChatState);
@@ -68,6 +77,7 @@ async function chatAgent({ selectedModel = "claude-3-7-sonnet-20250219", useDisk
         }
     }
     const systemPrompt = await (0, prompt_1.buildSystemPrompt)();
+    const { tools, toolExecutors } = getTools(selectedModel);
     while (!userPrompt?.toLowerCase().includes("stop")) {
         if (chatModel.askUserForInput) {
             try {

package/dist/bin/index.js

@@ -41,7 +41,7 @@ async function runChatAgent(modelInput, useDiskForChatState, initialPromptPath)
         "3-7": "claude-3-7-sonnet-20250219",
         "claude-3-5": "claude-3-5-sonnet-20241022",
         "3-5": "claude-3-5-sonnet-20241022",
-        "gemini-2.5-pro-exp-03-25": "gemini-2.5-pro-exp-03-25",
+        "gemini-2.5-pro-preview-03-25": "gemini-2.5-pro-preview-03-25",
     };
     if (modelInput && !MODEL_MAPPING[modelInput]) {
         throw new Error(`Invalid chat model: ${modelInput}`);
@@ -199,7 +199,7 @@ async function runAgentsWorkflow(testGenConfig, testGenToken) {
         .option("--suites <suites>", "Comma separated list of describe blocks")
         .option("--use-chat", "Use chat agent (and not the workflow)")
         .option("--use-disk-for-chat-state", "Save and load chat state from disk")
-        .option("--chat-model <model>", "Chat model to use (claude-3-7-sonnet-20250219 or claude-3-5-sonnet-20241022 or gemini-2.5-pro-exp-03-25)")
+        .option("--chat-model <model>", "Chat model to use (claude-3-7-sonnet-20250219 or claude-3-5-sonnet-20241022 or gemini-2.5-pro-preview-03-25)")
         .option("--initial-prompt <path>", "Path to an initial prompt file (e.g. prompt.md)")
         .parse(process.argv);
     const options = program.opts();

package/dist/bin/utils/index.d.ts

@@ -7,7 +7,7 @@ export interface CliOptions {
     useChat?: boolean;
     useDiskForChatState?: boolean;
     initialPrompt?: string;
-    chatModel?: "claude-3-7" | "3-7" | "claude-3-5" | "3-5" | "claude-3-7-sonnet-20250219" | "claude-3-5-sonnet-20241022" | "gemini-2.5-pro-exp-03-25";
+    chatModel?: "claude-3-7" | "3-7" | "claude-3-5" | "3-5" | "claude-3-7-sonnet-20250219" | "claude-3-5-sonnet-20241022" | "gemini-2.5-pro-preview-03-25";
 }
 export declare function validateAndCompleteCliOptions(options: CliOptions): Promise<CliOptions>;
 //# sourceMappingURL=index.d.ts.map

package/dist/bin/utils/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/bin/utils/index.ts"],"names":[],"mappings":"AAEA,MAAM,WAAW,UAAU;IACzB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,SAAS,CAAC,EACN,YAAY,GACZ,KAAK,GACL,YAAY,GACZ,KAAK,GACL,4BAA4B,GAC5B,4BAA4B,GAC5B,0BAA0B,CAAC;CAChC;AAQD,wBAAsB,6BAA6B,CACjD,OAAO,EAAE,UAAU,GAClB,OAAO,CAAC,UAAU,CAAC,CAyDrB"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/bin/utils/index.ts"],"names":[],"mappings":"AAEA,MAAM,WAAW,UAAU;IACzB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,SAAS,CAAC,EACN,YAAY,GACZ,KAAK,GACL,YAAY,GACZ,KAAK,GACL,4BAA4B,GAC5B,4BAA4B,GAC5B,8BAA8B,CAAC;CACpC;AAQD,wBAAsB,6BAA6B,CACjD,OAAO,EAAE,UAAU,GAClB,OAAO,CAAC,UAAU,CAAC,CAyDrB"}

package/dist/tools/diagnosis-fetcher.d.ts

@@ -1,3 +1,3 @@
 import type { Tool } from "@empiricalrun/llm/chat";
-export declare const diagnosisTool: Tool;
+export declare const fetchDiagnosisReportTool: Tool;
 //# sourceMappingURL=diagnosis-fetcher.d.ts.map

package/dist/tools/diagnosis-fetcher.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"diagnosis-fetcher.d.ts","sourceRoot":"","sources":["../../src/tools/diagnosis-fetcher.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAanD,eAAO,MAAM,aAAa,EAAE,IAgF3B,CAAC"}
+{"version":3,"file":"diagnosis-fetcher.d.ts","sourceRoot":"","sources":["../../src/tools/diagnosis-fetcher.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAanD,eAAO,MAAM,wBAAwB,EAAE,IAgFtC,CAAC"}

package/dist/tools/diagnosis-fetcher.js

@@ -3,7 +3,7 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
     return (mod && mod.__esModule) ? mod : { "default": mod };
 };
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.diagnosisTool = void 0;
+exports.fetchDiagnosisReportTool = void 0;
 const promises_1 = __importDefault(require("fs/promises"));
 const path_1 = __importDefault(require("path"));
 const zod_1 = require("zod");
@@ -12,7 +12,7 @@ const DiagnosisSchema = zod_1.z.object({
         .string()
         .describe("The full URL of the diagnosis (e.g. https://dash.empirical.run/shopflo-tests/diagnosis/split-cod-place-the-order--byynrPjCml57)"),
 });
-exports.diagnosisTool = {
+exports.fetchDiagnosisReportTool = {
     schema: {
         name: "fetchDiagnosisDetails",
         description: "Fetch details about a test case diagnosis using its URL or slug",

package/dist/tools/test-run-fetcher/index.d.ts

@@ -1,4 +1,4 @@
 import type { Tool } from "@empiricalrun/llm/chat";
 export declare function extractPathAfterSourceRepo(fullPath: string): string;
-export declare const testRunTool: Tool;
+export declare const fetchTestRunReportTool: Tool;
 //# sourceMappingURL=index.d.ts.map

package/dist/tools/test-run-fetcher/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/test-run-fetcher/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAanD,wBAAgB,0BAA0B,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,CAOnE;AAED,eAAO,MAAM,WAAW,EAAE,IA4HzB,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/test-run-fetcher/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAanD,wBAAgB,0BAA0B,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,CAOnE;AAED,eAAO,MAAM,sBAAsB,EAAE,IA4HpC,CAAC"}

package/dist/tools/test-run-fetcher/index.js

@@ -1,6 +1,6 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.testRunTool = exports.extractPathAfterSourceRepo = void 0;
+exports.fetchTestRunReportTool = exports.extractPathAfterSourceRepo = void 0;
 const zod_1 = require("zod");
 const TestRunSchema = zod_1.z.object({
     testRunUrl: zod_1.z
@@ -17,7 +17,7 @@ function extractPathAfterSourceRepo(fullPath) {
     }
 }
 exports.extractPathAfterSourceRepo = extractPathAfterSourceRepo;
-exports.testRunTool = {
+exports.fetchTestRunReportTool = {
     schema: {
         name: "fetchTestRunDetails",
         description: "Fetch details about a test run using its URL",

package/dist/tools/test-run.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"test-run.d.ts","sourceRoot":"","sources":["../../src/tools/test-run.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAwBnD,eAAO,MAAM,WAAW,EAAE,IA8BzB,CAAC"}
+{"version":3,"file":"test-run.d.ts","sourceRoot":"","sources":["../../src/tools/test-run.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAuBnD,eAAO,MAAM,WAAW,EAAE,IA8BzB,CAAC"}

package/dist/tools/test-run.js

@@ -15,8 +15,7 @@ const RunTestSchema = zod_1.z.object({
     headed: zod_1.z
         .boolean()
         .describe("Whether to run the test in headed mode (default is false, which is headless)")
-        .optional()
-        .default(false),
+        .optional(),
 });
 exports.runTestTool = {
     schema: {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@empiricalrun/test-gen",
-  "version": "0.52.2",
+  "version": "0.52.4",
   "publishConfig": {
     "registry": "https://registry.npmjs.org/",
     "access": "public"
@@ -75,7 +75,7 @@
     "tsx": "^4.16.2",
     "typescript": "^5.3.3",
     "zod": "^3.23.8",
-    "@empiricalrun/llm": "^0.13.0",
+    "@empiricalrun/llm": "^0.13.2",
     "@empiricalrun/r2-uploader": "^0.3.8",
     "@empiricalrun/test-run": "^0.7.6"
   },