@empiricalrun/test-gen 0.48.1 → 0.50.0

package/CHANGELOG.md

@@ -1,5 +1,24 @@
 # @empiricalrun/test-gen
 
+## 0.50.0
+
+### Minor Changes
+
+- 0eeff70: feat: add test-run-fetcher tool call, response body changes
+
+### Patch Changes
+
+- b14d5bf: feat: support headed executions of test run tool
+- Updated dependencies [b14d5bf]
+  - @empiricalrun/test-run@0.7.3
+
+## 0.49.0
+
+### Minor Changes
+
+- 3d050ec: feat: Add grep tool for case-insensitive code search
+- ec8f23b: feat: Convert agent tools to use Zod schemas
+
 ## 0.48.1
 
 ### 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":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,0BAA0B,CAAC;AA6D1D,wBAAsB,SAAS,CAAC,EAC9B,MAAM,GACP,EAAE;IACD,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB,8CAmEA"}
+{"version":3,"file":"chat.d.ts","sourceRoot":"","sources":["../../src/agent/chat.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,0BAA0B,CAAC;AAwE1D,wBAAsB,SAAS,CAAC,EAC9B,MAAM,GACP,EAAE;IACD,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB,8CAqEA"}

package/dist/agent/chat.js

@@ -9,7 +9,10 @@ const path_1 = __importDefault(require("path"));
 const human_in_the_loop_1 = require("../human-in-the-loop");
 const browser_agent_1 = require("../tools/browser-agent");
 const diagnosis_fetcher_1 = require("../tools/diagnosis-fetcher");
+const grep_1 = require("../tools/grep");
 const test_run_1 = require("../tools/test-run");
+const test_run_fetcher_1 = require("../tools/test-run-fetcher");
+const zod_schema_1 = require("../tools/zod-schema");
 const repo_tree_1 = require("../utils/repo-tree");
 const systemPrompt = `
 You are a helpful assistant that can answer questions and help with tasks.
@@ -23,6 +26,9 @@ Summarize the results in a few sentences.
 If the user provides a diagnosis URL, you can use the fetchDiagnosisDetails tool
 to get more information about the test case and its results.
 
+If the user provides a test run URL, you can use the fetchTestRunDetails tool
+to get detailed information about a specific test run.
+
 Or if the user asks you to modify a test, you could use the generateTestWithBrowserAgent tool. If you suspect
 that a UI selector needs to be updated, using the browser agent is a good idea.
 
@@ -38,6 +44,7 @@ The position of the comment is important: the browser agent will look for this c
 the actual code to click on the login button. If you are fixing a failing test, your comment should be 
 around the failing line of code, so that it can be replaced/modified.
 
+# Repo context
 You are running as a CLI tool inside the directory of the repo where this test file is located. Here is
 the repo directory structure:
 
@@ -47,9 +54,15 @@ While specifying paths to files, use relative paths from the current working dir
 - Correct path: "tests/lesson.spec.ts"
 - Incorrect path: "/repo/tests/lesson.spec.ts" or "${path_1.default.basename(process.cwd())}/tests/lesson.spec.ts"
 `;
-const tools = [test_run_1.runTestTool, browser_agent_1.browserAgentTool, diagnosis_fetcher_1.diagnosisTool];
+const tools = [
+    test_run_1.runTestTool,
+    browser_agent_1.browserAgentTool,
+    diagnosis_fetcher_1.diagnosisTool,
+    grep_1.grepTool,
+    test_run_fetcher_1.testRunTool,
+];
 const toolExecutors = {
-    ...Object.fromEntries(tools.map((tool) => [tool.schema.function.name, tool.execute])),
+    ...Object.fromEntries(tools.map((tool) => [tool.schema.name, tool.execute])),
     str_replace_editor: claude_1.strReplaceEditorTool,
 };
 async function chatAgent({ prompt, }) {
@@ -84,7 +97,7 @@ async function chatAgent({ prompt, }) {
             continue;
         }
         const response = await (0, claude_1.createChatCompletion)(systemPrompt, chatState.getMessages(), [
-            ...tools.map((tool) => (0, claude_1.convertOpenAISchemaToAnthropic)(tool.schema)),
+            ...tools.map((tool) => (0, claude_1.convertOpenAISchemaToAnthropic)((0, zod_schema_1.zodToOpenAITool)(tool.schema))),
             {
                 type: "text_editor_20250124",
                 name: "str_replace_editor",

package/dist/tools/browser-agent.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"browser-agent.d.ts","sourceRoot":"","sources":["../../src/tools/browser-agent.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,SAAS,CAAC;AAUpC,eAAO,MAAM,gBAAgB,EAAE,IAkH9B,CAAC"}
+{"version":3,"file":"browser-agent.d.ts","sourceRoot":"","sources":["../../src/tools/browser-agent.ts"],"names":[],"mappings":"AAMA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,SAAS,CAAC;AAmDpC,eAAO,MAAM,gBAAgB,EAAE,IA6C9B,CAAC"}

package/dist/tools/browser-agent.js

@@ -1,16 +1,25 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.browserAgentTool = void 0;
+const zod_1 = require("zod");
 const run_1 = require("../agent/browsing/run");
 const utils_1 = require("../agent/browsing/utils");
 const scenarios_1 = require("../bin/utils/scenarios");
 const git_1 = require("../utils/git");
-exports.browserAgentTool = {
-    schema: {
-        type: "function",
-        function: {
-            name: "generateTestWithBrowserAgent",
-            description: `
+const BrowserAgentSchema = zod_1.z.object({
+    testName: zod_1.z.string().describe("The name of the test to create or modify"),
+    testSuites: zod_1.z
+        .array(zod_1.z.string())
+        .describe("The suites (describe blocks) where the test is located"),
+    fileName: zod_1.z
+        .string()
+        .describe("The name of the file where the test is located. File name must end with .spec.ts"),
+    project: zod_1.z
+        .string()
+        .describe("The Playwright project to run tests against (e.g. 'chromium' or 'firefox')"),
+    changeToMake: zod_1.z.string().describe("The change to make to the test"),
+});
+const BROWSER_AGENT_DESCRIPTION = `
 Create or modify a test case with browser agent. The browser agent can take user interactions in a web browser
 and generate Playwright code for that actions. This is a useful tool when the modifications require knowing the 
 locator/selector for an element on the page.
@@ -39,43 +48,12 @@ test("Example test code", async ({ page }) => {
   await page.getByRole("button", { name: "Login" }).click();
 });
 \`\`\`
-      `,
-            parameters: {
-                type: "object",
-                properties: {
-                    testName: {
-                        type: "string",
-                        description: "The name of the test to create or modify",
-                    },
-                    testSuites: {
-                        type: "array",
-                        description: "The suites (describe blocks) where the test is located",
-                        items: {
-                            type: "string",
-                        },
-                    },
-                    fileName: {
-                        type: "string",
-                        description: "The name of the file where the test is located. File name must end with .spec.ts",
-                    },
-                    project: {
-                        type: "string",
-                        description: "The Playwright project to run tests against (e.g. 'chromium' or 'firefox')",
-                    },
-                    changeToMake: {
-                        type: "string",
-                        description: "The change to make to the test",
-                    },
-                },
-                required: [
-                    "testName",
-                    "testSuites",
-                    "fileName",
-                    "changeToMake",
-                    "project",
-                ],
-            },
-        },
+`;
+exports.browserAgentTool = {
+    schema: {
+        name: "generateTestWithBrowserAgent",
+        description: BROWSER_AGENT_DESCRIPTION,
+        parameters: BrowserAgentSchema,
     },
     execute: async (input) => {
         const { testName, testSuites, fileName, changeToMake, project } = input;

package/dist/tools/codegen-agent.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"codegen-agent.d.ts","sourceRoot":"","sources":["../../src/tools/codegen-agent.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,SAAS,CAAC;AASpC,eAAO,MAAM,WAAW,EAAE,IAsDzB,CAAC"}
+{"version":3,"file":"codegen-agent.d.ts","sourceRoot":"","sources":["../../src/tools/codegen-agent.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,SAAS,CAAC;AAepC,eAAO,MAAM,WAAW,EAAE,IAyBzB,CAAC"}

package/dist/tools/codegen-agent.js

@@ -1,39 +1,23 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.codegenTool = void 0;
+const zod_1 = require("zod");
 const run_1 = require("../agent/codegen/run");
+const CodegenSchema = zod_1.z.object({
+    testName: zod_1.z.string().describe("The name of the test to create or modify"),
+    testSuites: zod_1.z
+        .array(zod_1.z.string())
+        .describe("The suites (describe blocks) where the test is located"),
+    fileName: zod_1.z
+        .string()
+        .describe("The name of the file where the test is located. File name must end with .spec.ts"),
+    changeToMake: zod_1.z.string().describe("The change to make to the test"),
+});
 exports.codegenTool = {
     schema: {
-        type: "function",
-        function: {
-            name: "generateTestWithCodegen",
-            description: "Create or modify a test case with code generation. This is useful when modifications can be done with TypeScript only, and don't require any browser interactions or element selectors.",
-            parameters: {
-                type: "object",
-                properties: {
-                    testName: {
-                        type: "string",
-                        description: "The name of the test to create or modify",
-                    },
-                    testSuites: {
-                        type: "array",
-                        description: "The suites (describe blocks) where the test is located",
-                        items: {
-                            type: "string",
-                        },
-                    },
-                    fileName: {
-                        type: "string",
-                        description: "The name of the file where the test is located. File name must end with .spec.ts",
-                    },
-                    changeToMake: {
-                        type: "string",
-                        description: "The change to make to the test",
-                    },
-                },
-                required: ["testName", "testSuites", "fileName", "changeToMake"],
-            },
-        },
+        name: "generateTestWithCodegen",
+        description: "Create or modify a test case with code generation. This is useful when modifications can be done with TypeScript only, and don't require any browser interactions or element selectors.",
+        parameters: CodegenSchema,
     },
     execute: async (input) => {
         const { testName, testSuites, fileName, changeToMake } = input;

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":"AAGA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,SAAS,CAAC;AAMpC,eAAO,MAAM,aAAa,EAAE,IA4F3B,CAAC"}
+{"version":3,"file":"diagnosis-fetcher.d.ts","sourceRoot":"","sources":["../../src/tools/diagnosis-fetcher.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,SAAS,CAAC;AAUpC,eAAO,MAAM,aAAa,EAAE,IAgF3B,CAAC"}

package/dist/tools/diagnosis-fetcher.js

@@ -6,23 +6,17 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.diagnosisTool = void 0;
 const promises_1 = __importDefault(require("fs/promises"));
 const path_1 = __importDefault(require("path"));
+const zod_1 = require("zod");
+const DiagnosisSchema = zod_1.z.object({
+    diagnosisUrl: zod_1.z
+        .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 = {
     schema: {
-        type: "function",
-        function: {
-            name: "fetchDiagnosisDetails",
-            description: "Fetch details about a test case diagnosis using its URL or slug",
-            parameters: {
-                type: "object",
-                properties: {
-                    diagnosisUrl: {
-                        type: "string",
-                        description: "The full URL of the diagnosis (e.g. https://dash.empirical.run/shopflo-tests/diagnosis/split-cod-place-the-order--byynrPjCml57)",
-                    },
-                },
-                required: ["diagnosisUrl"],
-            },
-        },
+        name: "fetchDiagnosisDetails",
+        description: "Fetch details about a test case diagnosis using its URL or slug",
+        parameters: DiagnosisSchema,
     },
     execute: async (input) => {
         const { diagnosisUrl } = input;
@@ -31,11 +25,10 @@ exports.diagnosisTool = {
         if (!slug) {
             throw new Error("Invalid diagnosis URL - could not extract slug");
         }
-        // Make the API call to fetch diagnosis details
         const response = await fetch(`https://dash.empirical.run/api/diagnosis/${slug}/detailed`, {
             method: "GET",
             headers: {
-                Authorization: "weQPMWKT", // Using the auth token from test-endpoint.mdc
+                Authorization: "weQPMWKT",
             },
         });
         if (!response.ok) {
@@ -46,8 +39,10 @@ exports.diagnosisTool = {
         }
         const data = await response.json();
         const { test_case, diagnosis } = data.data;
-        const project = diagnosis[0]?.test_project || "unknown";
+        const project = diagnosis?.test_project || "unknown";
         const sourceContext = await promises_1.default.readFile(path_1.default.join("tests", test_case.file_path), "utf-8");
+        const repoName = path_1.default.basename(process.cwd());
+        const cleanErrorStack = diagnosis?.failed_run_metadata?.stack?.replace(`"/runner/_work/${repoName}/${repoName}/source-repo/"`, "");
         // Format the response as markdown
         const markdownResponse = `
 # Test Case Diagnosis
@@ -64,21 +59,21 @@ ${sourceContext}
 ## What Happened in the Test Run
 
 ### Failure Details
-- **Failing Line**: ${diagnosis[0]?.failing_line || "No failing line available"}
+- **Failing Line**: ${diagnosis?.failing_line || "No failing line available"}
 
 #### Error Stack
 \`\`\`
-${diagnosis[0]?.failed_run_metadata?.stack?.replace("/runner/_work/shopflo-tests/shopflo-tests/source-repo/", "") || "No error stack available"}
+${cleanErrorStack || "No error stack available"}
 \`\`\`
 
 #### Error Summary
-${diagnosis[0]?.error_stack_summary?.content || "No error summary available"}
+${diagnosis?.error_stack_summary?.content || "No error summary available"}
 
 #### Visual Analysis
-${diagnosis[0]?.visual_diff_summary?.summary || "No visual analysis available"}
+${diagnosis?.visual_diff_summary?.summary || "No visual analysis available"}
 
 #### Merged Summary
-${diagnosis[0]?.merged_summary?.content || "No merged summary available"}
+${diagnosis?.merged_summary?.content || "No merged summary available"}
 `;
         return {
             result: markdownResponse,

package/dist/tools/grep.d.ts

@@ -0,0 +1,3 @@
+import { Tool } from "./types";
+export declare const grepTool: Tool;
+//# sourceMappingURL=grep.d.ts.map

package/dist/tools/grep.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"grep.d.ts","sourceRoot":"","sources":["../../src/tools/grep.ts"],"names":[],"mappings":"AAKA,OAAO,EAAE,IAAI,EAAc,MAAM,SAAS,CAAC;AAgB3C,eAAO,MAAM,QAAQ,EAAE,IA+CtB,CAAC"}

package/dist/tools/grep.js

@@ -0,0 +1,62 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.grepTool = void 0;
+const child_process_1 = require("child_process");
+const util_1 = require("util");
+const zod_1 = require("zod");
+const repo_tree_1 = require("../utils/repo-tree");
+const execAsync = (0, util_1.promisify)(child_process_1.exec);
+const GrepInputSchema = zod_1.z.object({
+    pattern: zod_1.z.string().describe("The pattern to search for"),
+    directory: zod_1.z
+        .string()
+        .optional()
+        .describe("The directory to search in (defaults to current directory)"),
+    filePattern: zod_1.z
+        .string()
+        .optional()
+        .describe("File pattern to search in (e.g., '*.ts' for TypeScript files)"),
+});
+exports.grepTool = {
+    schema: {
+        name: "grep",
+        description: "Search for a pattern in files using grep (case insensitive)",
+        parameters: GrepInputSchema,
+    },
+    execute: async (input) => {
+        try {
+            const dir = input.directory || process.cwd();
+            // Create exclude pattern for grep
+            const excludePatterns = repo_tree_1.DEFAULT_EXCLUDE.map((pattern) => typeof pattern === "string" ? pattern : pattern.source)
+                .map((pattern) => `--exclude-dir="${pattern}"`)
+                .join(" ");
+            // Using -n to show line numbers in output
+            let cmd = `grep -rin ${excludePatterns} "${input.pattern}" ${dir}`;
+            if (input.filePattern) {
+                // For file pattern searches, we'll use find with exclusions
+                const excludeFind = repo_tree_1.DEFAULT_EXCLUDE.map((pattern) => typeof pattern === "string" ? pattern : pattern.source)
+                    .map((pattern) => `-not -path "*/${pattern}/*"`)
+                    .join(" ");
+                // Using -n to show line numbers and removed -l to show actual matches
+                cmd = `find ${dir} ${excludeFind} -name "${input.filePattern}" -exec grep -rin "${input.pattern}" {} \\;`;
+            }
+            const { stdout, stderr } = await execAsync(cmd);
+            if (stderr) {
+                return {
+                    isError: true,
+                    result: stderr,
+                };
+            }
+            return {
+                isError: false,
+                result: stdout,
+            };
+        }
+        catch (error) {
+            return {
+                isError: true,
+                result: error instanceof Error ? error.message : String(error),
+            };
+        }
+    },
+};

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

@@ -0,0 +1,3 @@
+import type { Tool } from "./types";
+export declare const testRunTool: Tool;
+//# sourceMappingURL=test-run-fetcher.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"test-run-fetcher.d.ts","sourceRoot":"","sources":["../../src/tools/test-run-fetcher.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,SAAS,CAAC;AAUpC,eAAO,MAAM,WAAW,EAAE,IA0DzB,CAAC"}

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

@@ -0,0 +1,59 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.testRunTool = void 0;
+const zod_1 = require("zod");
+const TestRunSchema = zod_1.z.object({
+    testRunUrl: zod_1.z
+        .string()
+        .describe("The full URL of the test run (e.g. https://dash.empirical.run/sortment-tests/test-runs/20269 or with query params like ?status=failed)"),
+});
+exports.testRunTool = {
+    schema: {
+        name: "fetchTestRunDetails",
+        description: "Fetch details about a test run using its URL",
+        parameters: TestRunSchema,
+    },
+    execute: async (input) => {
+        const { testRunUrl } = input;
+        // Remove query parameters if they exist
+        const urlWithoutParams = testRunUrl.split("?")[0] || testRunUrl;
+        // Extract the run ID and repo name from the URL
+        const urlParts = urlWithoutParams.split("/");
+        const runId = urlParts.pop(); // Last part is the run ID
+        const repoName = urlParts[urlParts.length - 2]; // Second to last part is the repo name
+        if (!runId || !repoName) {
+            throw new Error("Invalid test run URL - could not extract run ID or repo name");
+        }
+        // Make the API call to fetch test run details
+        const response = await fetch(`https://dash.empirical.run/api/test-runs/${runId}?repo_name=${repoName}`, {
+            method: "GET",
+            headers: {
+                Authorization: "weQPMWKT",
+            },
+        });
+        if (!response.ok) {
+            return {
+                result: `Failed to fetch test run details: ${response.statusText}`,
+                isError: true,
+            };
+        }
+        const data = await response.json();
+        // Format the response as markdown
+        const markdownResponse = `
+# Test Run Details
+
+## Run Information
+- **Run ID**: ${runId}
+- **Repository**: ${repoName}
+
+## Test Run Data
+\`\`\`json
+${JSON.stringify(data, null, 2)}
+\`\`\`
+`;
+        return {
+            result: markdownResponse,
+            isError: false,
+        };
+    },
+};

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":"AAEA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,SAAS,CAAC;AASpC,eAAO,MAAM,WAAW,EAAE,IA8CzB,CAAC"}
+{"version":3,"file":"test-run.d.ts","sourceRoot":"","sources":["../../src/tools/test-run.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,SAAS,CAAC;AAsBpC,eAAO,MAAM,WAAW,EAAE,IAoBzB,CAAC"}

package/dist/tools/test-run.js

@@ -2,44 +2,36 @@
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.runTestTool = void 0;
 const test_run_1 = require("@empiricalrun/test-run");
+const zod_1 = require("zod");
+const RunTestSchema = zod_1.z.object({
+    testName: zod_1.z.string().describe("The name of the test to run"),
+    suites: zod_1.z
+        .array(zod_1.z.string())
+        .describe("The suites (describe blocks) where the test is located."),
+    fileName: zod_1.z
+        .string()
+        .describe("The name of the file where the test is located. File name must end with .spec.ts"),
+    project: zod_1.z.string().describe("The project to run the test on"),
+    headed: zod_1.z
+        .boolean()
+        .describe("Whether to run the test in headed mode (default is false, which is headless)")
+        .optional()
+        .default(false),
+});
 exports.runTestTool = {
     schema: {
-        type: "function",
-        function: {
-            name: "runTest",
-            description: "Run a test",
-            parameters: {
-                type: "object",
-                properties: {
-                    testName: {
-                        type: "string",
-                        description: "The name of the test to run",
-                    },
-                    suites: {
-                        type: "array",
-                        description: "The suites (describe blocks) where the test is located.",
-                        items: { type: "string" },
-                    },
-                    fileName: {
-                        type: "string",
-                        description: "The name of the file where the test is located. File name must end with .spec.ts",
-                    },
-                    project: {
-                        type: "string",
-                        description: "The project to run the test on",
-                    },
-                },
-                required: ["testName", "suites", "fileName", "project"],
-            },
-        },
+        name: "runTest",
+        description: "Run a test",
+        parameters: RunTestSchema,
     },
     execute: async (input) => {
-        const { testName, suites, fileName, project } = input;
+        const { testName, suites, fileName, project, headed } = input;
         const result = await (0, test_run_1.runSingleTest)({
             testName,
             suites,
             fileName,
             projects: [project],
+            headed,
         });
         return {
             result: JSON.stringify(result),

package/dist/tools/types.d.ts

@@ -1,11 +1,38 @@
-import type { OpenAI } from "openai";
-export type ToolSchema = OpenAI.Chat.Completions.ChatCompletionTool;
-export type ToolResult = {
-    result: string;
-    isError: boolean;
-};
-export type Tool = {
-    schema: ToolSchema;
+import { z } from "zod";
+/**
+ * Base schema for all tools. Each tool should extend this with their specific parameters.
+ */
+export declare const BaseToolSchema: z.ZodObject<{
+    name: z.ZodString;
+    description: z.ZodString;
+    parameters: z.ZodObject<{}, "passthrough", z.ZodTypeAny, z.objectOutputType<{}, z.ZodTypeAny, "passthrough">, z.objectInputType<{}, z.ZodTypeAny, "passthrough">>;
+}, "strip", z.ZodTypeAny, {
+    name: string;
+    description: string;
+    parameters: {} & {
+        [k: string]: unknown;
+    };
+}, {
+    name: string;
+    description: string;
+    parameters: {} & {
+        [k: string]: unknown;
+    };
+}>;
+export type ToolSchema = z.infer<typeof BaseToolSchema>;
+/**
+ * Interface for creating a tool with its schema and execute function
+ */
+export interface Tool {
+    schema: {
+        name: string;
+        description: string;
+        parameters: z.ZodType;
+    };
     execute: (input: any) => Promise<ToolResult>;
-};
+}
+export interface ToolResult {
+    isError: boolean;
+    result: string;
+}
 //# sourceMappingURL=types.d.ts.map

package/dist/tools/types.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/tools/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,QAAQ,CAAC;AAErC,MAAM,MAAM,UAAU,GAAG,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,kBAAkB,CAAC;AAEpE,MAAM,MAAM,UAAU,GAAG;IACvB,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,OAAO,CAAC;CAClB,CAAC;AAEF,MAAM,MAAM,IAAI,GAAG;IAIjB,MAAM,EAAE,UAAU,CAAC;IACnB,OAAO,EAAE,CAAC,KAAK,EAAE,GAAG,KAAK,OAAO,CAAC,UAAU,CAAC,CAAC;CAC9C,CAAC"}
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/tools/types.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB;;GAEG;AACH,eAAO,MAAM,cAAc;;;;;;;;;;;;;;;;EAIzB,CAAC;AAEH,MAAM,MAAM,UAAU,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,cAAc,CAAC,CAAC;AAExD;;GAEG;AACH,MAAM,WAAW,IAAI;IACnB,MAAM,EAAE;QACN,IAAI,EAAE,MAAM,CAAC;QACb,WAAW,EAAE,MAAM,CAAC;QACpB,UAAU,EAAE,CAAC,CAAC,OAAO,CAAC;KACvB,CAAC;IACF,OAAO,EAAE,CAAC,KAAK,EAAE,GAAG,KAAK,OAAO,CAAC,UAAU,CAAC,CAAC;CAC9C;AAED,MAAM,WAAW,UAAU;IACzB,OAAO,EAAE,OAAO,CAAC;IACjB,MAAM,EAAE,MAAM,CAAC;CAChB"}

package/dist/tools/types.js

@@ -1,2 +1,12 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
+exports.BaseToolSchema = void 0;
+const zod_1 = require("zod");
+/**
+ * Base schema for all tools. Each tool should extend this with their specific parameters.
+ */
+exports.BaseToolSchema = zod_1.z.object({
+    name: zod_1.z.string(),
+    description: zod_1.z.string(),
+    parameters: zod_1.z.object({}).passthrough(),
+});

package/dist/tools/zod-schema.d.ts

@@ -0,0 +1,19 @@
+import type OpenAI from "openai";
+import { z } from "zod";
+/**
+ * Convert a tool schema to OpenAI tool format
+ */
+export declare function zodToOpenAITool(schema: {
+    name: string;
+    description: string;
+    parameters: z.ZodType;
+}): OpenAI.Chat.Completions.ChatCompletionTool;
+/**
+ * Convert Zod schema to JSON Schema
+ */
+export declare function zodToJsonSchema(schema: z.ZodType): any;
+/**
+ * Convert specific Zod type to JSON Schema
+ */
+export declare function zodTypeToJsonSchema(zodType: z.ZodType): any;
+//# sourceMappingURL=zod-schema.d.ts.map

package/dist/tools/zod-schema.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"zod-schema.d.ts","sourceRoot":"","sources":["../../src/tools/zod-schema.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,MAAM,MAAM,QAAQ,CAAC;AACjC,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB;;GAEG;AACH,wBAAgB,eAAe,CAAC,MAAM,EAAE;IACtC,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,CAAC,CAAC,OAAO,CAAC;CACvB,GAAG,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,kBAAkB,CAS7C;AAED;;GAEG;AACH,wBAAgB,eAAe,CAAC,MAAM,EAAE,CAAC,CAAC,OAAO,GAAG,GAAG,CAuBtD;AAED;;GAEG;AACH,wBAAgB,mBAAmB,CAAC,OAAO,EAAE,CAAC,CAAC,OAAO,GAAG,GAAG,CAoD3D"}

package/dist/tools/zod-schema.js

@@ -0,0 +1,95 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.zodTypeToJsonSchema = exports.zodToJsonSchema = exports.zodToOpenAITool = void 0;
+const zod_1 = require("zod");
+/**
+ * Convert a tool schema to OpenAI tool format
+ */
+function zodToOpenAITool(schema) {
+    return {
+        type: "function",
+        function: {
+            name: schema.name,
+            description: schema.description,
+            parameters: zodToJsonSchema(schema.parameters),
+        },
+    };
+}
+exports.zodToOpenAITool = zodToOpenAITool;
+/**
+ * Convert Zod schema to JSON Schema
+ */
+function zodToJsonSchema(schema) {
+    if (schema instanceof zod_1.z.ZodObject) {
+        const shape = schema._def.shape();
+        const properties = {};
+        const required = [];
+        Object.entries(shape).forEach(([key, value]) => {
+            properties[key] = zodTypeToJsonSchema(value);
+            // Check if this field is required
+            if (!(value instanceof zod_1.z.ZodOptional)) {
+                required.push(key);
+            }
+        });
+        return {
+            type: "object",
+            properties,
+            ...(required.length > 0 ? { required } : {}),
+        };
+    }
+    return { type: "string" }; // Fallback
+}
+exports.zodToJsonSchema = zodToJsonSchema;
+/**
+ * Convert specific Zod type to JSON Schema
+ */
+function zodTypeToJsonSchema(zodType) {
+    // Handle string types
+    if (zodType instanceof zod_1.z.ZodString) {
+        const schema = { type: "string" };
+        if (zodType.description)
+            schema.description = zodType.description;
+        return schema;
+    }
+    // Handle number types
+    if (zodType instanceof zod_1.z.ZodNumber) {
+        const schema = { type: "number" };
+        if (zodType.description)
+            schema.description = zodType.description;
+        return schema;
+    }
+    // Handle boolean
+    if (zodType instanceof zod_1.z.ZodBoolean) {
+        const schema = { type: "boolean" };
+        if (zodType.description)
+            schema.description = zodType.description;
+        return schema;
+    }
+    // Handle arrays
+    if (zodType instanceof zod_1.z.ZodArray) {
+        return {
+            type: "array",
+            items: zodTypeToJsonSchema(zodType._def.type),
+            ...(zodType.description ? { description: zodType.description } : {}),
+        };
+    }
+    // Handle objects
+    if (zodType instanceof zod_1.z.ZodObject) {
+        return zodToJsonSchema(zodType);
+    }
+    // Handle enums
+    if (zodType instanceof zod_1.z.ZodEnum) {
+        return {
+            type: "string",
+            enum: zodType._def.values,
+            ...(zodType.description ? { description: zodType.description } : {}),
+        };
+    }
+    // Handle optional types
+    if (zodType instanceof zod_1.z.ZodOptional) {
+        return zodTypeToJsonSchema(zodType._def.innerType);
+    }
+    // Default fallback
+    return { type: "string" };
+}
+exports.zodTypeToJsonSchema = zodTypeToJsonSchema;

package/dist/utils/repo-tree.d.ts

@@ -1,2 +1,3 @@
+export declare const DEFAULT_EXCLUDE: (string | RegExp)[];
 export declare function generateAsciiTree(dirPath: string, options?: {}): string;
 //# sourceMappingURL=repo-tree.d.ts.map

package/dist/utils/repo-tree.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"repo-tree.d.ts","sourceRoot":"","sources":["../../src/utils/repo-tree.ts"],"names":[],"mappings":"AAYA,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,KAAK,UAsE9D"}
+{"version":3,"file":"repo-tree.d.ts","sourceRoot":"","sources":["../../src/utils/repo-tree.ts"],"names":[],"mappings":"AAGA,eAAO,MAAM,eAAe,qBAO3B,CAAC;AAEF,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,KAAK,UAsE9D"}

package/dist/utils/repo-tree.js

@@ -3,10 +3,10 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
     return (mod && mod.__esModule) ? mod : { "default": mod };
 };
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.generateAsciiTree = void 0;
+exports.generateAsciiTree = exports.DEFAULT_EXCLUDE = void 0;
 const fs_1 = __importDefault(require("fs"));
 const path_1 = __importDefault(require("path"));
-const DEFAULT_EXCLUDE = [
+exports.DEFAULT_EXCLUDE = [
     "node_modules",
     "dist",
     "build",
@@ -17,7 +17,7 @@ const DEFAULT_EXCLUDE = [
 function generateAsciiTree(dirPath, options = {}) {
     const defaultOptions = {
         showHidden: false,
-        exclude: DEFAULT_EXCLUDE,
+        exclude: exports.DEFAULT_EXCLUDE,
         maxDepth: 10,
     };
     const opts = { ...defaultOptions, ...options };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@empiricalrun/test-gen",
-  "version": "0.48.1",
+  "version": "0.50.0",
   "publishConfig": {
     "registry": "https://registry.npmjs.org/",
     "access": "public"
@@ -73,10 +73,11 @@
     "ts-morph": "^23.0.0",
     "tsx": "^4.16.2",
     "typescript": "^5.3.3",
+    "zod": "^3.23.8",
     "@empiricalrun/llm": "^0.10.1",
     "@empiricalrun/r2-uploader": "^0.3.8",
     "@empiricalrun/reporter": "^0.23.1",
-    "@empiricalrun/test-run": "^0.7.2"
+    "@empiricalrun/test-run": "^0.7.3"
   },
   "devDependencies": {
     "@playwright/test": "1.47.1",