@empiricalrun/test-gen 0.54.1 → 0.55.0

package/CHANGELOG.md

@@ -1,5 +1,19 @@
 # @empiricalrun/test-gen
 
+## 0.55.0
+
+### Minor Changes
+
+- 508565d: feat: add support for openai chat model
+- 8da022c: feat: add environment fetching tool
+
+### Patch Changes
+
+- 99e4e6e: refactor: decoupled agentloop and toolcall
+- Updated dependencies [cc4cb5e]
+- Updated dependencies [508565d]
+  - @empiricalrun/llm@0.15.0
+
 ## 0.54.1
 
 ### Patch Changes

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

@@ -1 +1 @@
-{"version":3,"file":"agent-loop.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/agent-loop.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,EACL,UAAU,EAIX,MAAM,wBAAwB,CAAC;AAiBhC,OAAO,EAAE,gBAAgB,EAAE,mBAAmB,EAAE,MAAM,SAAS,CAAC;AAyChE,wBAAsB,aAAa,CAAC,EAClC,SAAS,EACT,aAAa,EACb,QAAQ,EACR,KAAK,GACN,EAAE;IACD,SAAS,EAAE,UAAU,CAAC,GAAG,CAAC,CAAC;IAC3B,aAAa,EAAE,mBAAmB,CAAC;IACnC,QAAQ,EAAE,gBAAgB,CAAC;IAC3B,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB,iBAiDA"}
+{"version":3,"file":"agent-loop.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/agent-loop.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,EACL,UAAU,EAIX,MAAM,wBAAwB,CAAC;AAOhC,OAAO,EAAE,gBAAgB,EAAE,mBAAmB,EAAE,MAAM,SAAS,CAAC;AAgBhE,wBAAsB,aAAa,CAAC,EAClC,SAAS,EACT,aAAa,EACb,QAAQ,EACR,KAAK,GACN,EAAE;IACD,SAAS,EAAE,UAAU,CAAC,GAAG,CAAC,CAAC;IAC3B,aAAa,EAAE,mBAAmB,CAAC;IACnC,QAAQ,EAAE,gBAAgB,CAAC;IAC3B,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB,iBAiDA"}

package/dist/agent/chat/agent-loop.js

@@ -3,55 +3,28 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.chatAgentLoop = void 0;
 const chat_1 = require("@empiricalrun/llm/chat");
 const picocolors_1 = require("picocolors");
-const web_1 = require("../../bin/utils/platform/web");
-const commit_and_create_pr_1 = require("../../tools/commit-and-create-pr");
-const diagnosis_fetcher_1 = require("../../tools/diagnosis-fetcher");
-const grep_1 = require("../../tools/grep");
+const tool_call_service_1 = require("../../tool-call-service");
 const str_replace_editor_1 = require("../../tools/str_replace_editor");
-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");
 const state_1 = require("./state");
-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,
-        commit_and_create_pr_1.commitAndPushChangesTool,
-    ];
-    if (selectedModel.startsWith("gemini")) {
-        // Claude will have its own built-in text editor tools
-        str_replace_editor_1.textEditorTools.forEach((tool) => {
-            const originalExecute = tool.execute;
-            tool.execute = (input) => originalExecute(input, web_1.validateTypescript);
-        });
-        tools.push(...str_replace_editor_1.textEditorTools);
-    }
-    const toolExecutors = {
-        ...Object.fromEntries(tools.map((tool) => [tool.schema.name, tool.execute])),
-    };
-    if (selectedModel.startsWith("claude")) {
-        toolExecutors.str_replace_editor = (input) => (0, str_replace_editor_1.strReplaceEditorExecutor)(input, web_1.validateTypescript);
-    }
-    return { tools, toolExecutors };
-}
 function getModelName(model) {
     if (model.startsWith("claude"))
         return "Claude";
     if (model.startsWith("gemini"))
         return "Gemini";
+    if (model.startsWith("o4"))
+        return "o4";
     return "AI";
 }
 const log = (...args) => {
     console.log((0, picocolors_1.gray)(args.join(" ")));
 };
+const isRemote = process.env.TOOL_EXECUTION_IS_REMOTE === "true" || false;
 async function chatAgentLoop({ chatModel, selectedModel, reporter, trace, }) {
     const systemPrompt = await (0, prompt_1.buildSystemPrompt)();
     trace?.update({ input: { systemPrompt } });
-    const { tools, toolExecutors } = getTools(selectedModel);
+    const toolCallService = new tool_call_service_1.ToolCallService();
+    const { tools } = await toolCallService.getTools(selectedModel);
     while (!chatModel.askUserForInput) {
         const toolCalls = chatModel.getPendingToolCalls();
         if (toolCalls.length) {
@@ -59,22 +32,17 @@ async function chatAgentLoop({ chatModel, selectedModel, reporter, trace, }) {
             for (const call of toolCalls) {
                 const args = JSON.stringify(call.input);
                 log(`Executing tool ${call.name} with args: ${args}`);
-                const toolExecutor = toolExecutors[call.name];
-                let callResponse;
-                if (!toolExecutor) {
-                    callResponse = {
-                        isError: true,
-                        result: `Invalid function/tool call: ${call.name} not found`,
-                    };
+                let callResponse = await toolCallService.execute({
+                    tool: {
+                        name: call.name,
+                        input: call.input,
+                    },
+                }, isRemote);
+                if (callResponse.isError) {
+                    log(`Tool ${call.name} failed: ${callResponse.result}`);
                 }
                 else {
-                    callResponse = await toolExecutor(call.input);
-                    if (callResponse.isError) {
-                        log(`Tool ${call.name} failed: ${callResponse.result}`);
-                    }
-                    else {
-                        log(`Tool ${call.name} completed`);
-                    }
+                    log(`Tool ${call.name} completed`);
                 }
                 toolResults.push(callResponse);
             }

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

@@ -1 +1 @@
-{"version":3,"file":"model.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/model.ts"],"names":[],"mappings":"AAAA,OAAO,EAGL,UAAU,EACX,MAAM,wBAAwB,CAAC;AAEhC,OAAO,EAAE,mBAAmB,EAAE,MAAM,SAAS,CAAC;AAE9C,wBAAgB,eAAe,CAC7B,QAAQ,EAAE,GAAG,EAAE,EACf,aAAa,EAAE,mBAAmB,GACjC,UAAU,CAAC,GAAG,CAAC,CAQjB"}
+{"version":3,"file":"model.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/model.ts"],"names":[],"mappings":"AAAA,OAAO,EAGL,UAAU,EAEX,MAAM,wBAAwB,CAAC;AAEhC,OAAO,EAAE,mBAAmB,EAAE,MAAM,SAAS,CAAC;AAE9C,wBAAgB,eAAe,CAC7B,QAAQ,EAAE,GAAG,EAAE,EACf,aAAa,EAAE,mBAAmB,GACjC,UAAU,CAAC,GAAG,CAAC,CAWjB"}

package/dist/agent/chat/model.js

@@ -9,6 +9,9 @@ function createChatModel(messages, selectedModel) {
     if (selectedModel.startsWith("gemini")) {
         return new chat_1.GeminiChatModel(messages);
     }
+    if (selectedModel.startsWith("o4")) {
+        return new chat_1.OpenAIChatModel(messages);
+    }
     throw new Error(`Unsupported model: ${selectedModel}`);
 }
 exports.createChatModel = createChatModel;

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

@@ -1,5 +1,5 @@
 import { ChatStateOnDisk } from "./state";
-export type SupportedChatModels = "claude-3-7-sonnet-20250219" | "claude-3-5-sonnet-20241022" | "gemini-2.5-pro-preview-03-25";
+export type SupportedChatModels = "claude-3-7-sonnet-20250219" | "claude-3-5-sonnet-20241022" | "gemini-2.5-pro-preview-03-25" | "o4-mini-2025-04-16";
 type LatestMessage = {
     role: string;
     textMessage: string;

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

@@ -1 +1 @@
-{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/types.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,MAAM,SAAS,CAAC;AAE1C,MAAM,MAAM,mBAAmB,GAC3B,4BAA4B,GAC5B,4BAA4B,GAC5B,8BAA8B,CAAC;AAEnC,KAAK,aAAa,GAAG;IACnB,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;CACrB,CAAC;AAEF,MAAM,MAAM,gBAAgB,GAAG,CAC7B,KAAK,EAAE,eAAe,CAAC,GAAG,CAAC,EAC3B,0BAA0B,EAAE,aAAa,GAAG,SAAS,KAClD,OAAO,CAAC,IAAI,CAAC,CAAC"}
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/types.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,MAAM,SAAS,CAAC;AAE1C,MAAM,MAAM,mBAAmB,GAC3B,4BAA4B,GAC5B,4BAA4B,GAC5B,8BAA8B,GAC9B,oBAAoB,CAAC;AAEzB,KAAK,aAAa,GAAG;IACnB,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;CACrB,CAAC;AAEF,MAAM,MAAM,gBAAgB,GAAG,CAC7B,KAAK,EAAE,eAAe,CAAC,GAAG,CAAC,EAC3B,0BAA0B,EAAE,aAAa,GAAG,SAAS,KAClD,OAAO,CAAC,IAAI,CAAC,CAAC"}

package/dist/bin/index.js

@@ -38,10 +38,11 @@ function setupProcessListeners(cleanup) {
 async function runChatAgent({ modelInput, chatSessionId, useDiskForChatState, initialPromptPath, }) {
     const MODEL_MAPPING = {
         "claude-3-7": "claude-3-7-sonnet-20250219",
-        "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-preview-03-25": "gemini-2.5-pro-preview-03-25",
+        "gemini-2.5-pro": "gemini-2.5-pro-preview-03-25",
+        "o4-mini": "o4-mini-2025-04-16",
+        "o4-mini-2025-04-16": "o4-mini-2025-04-16",
     };
     if (modelInput && !MODEL_MAPPING[modelInput]) {
         throw new Error(`Invalid chat model: ${modelInput}`);

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

@@ -8,7 +8,7 @@ export interface CliOptions {
     useDiskForChatState?: boolean;
     initialPrompt?: string;
     chatSessionId?: 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-preview-03-25";
+    chatModel?: "claude-3-7" | "claude-3-5" | "claude-3-7-sonnet-20250219" | "claude-3-5-sonnet-20241022" | "gemini-2.5-pro" | "gemini-2.5-pro-preview-03-25" | "o4-mini" | "o4-mini-2025-04-16";
 }
 export declare function validateAndCompleteCliOptions(options: CliOptions): Promise<CliOptions>;
 export declare function printBanner(): void;

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,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;AAED,wBAAgB,WAAW,SAgC1B"}
+{"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,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,SAAS,CAAC,EACN,YAAY,GACZ,YAAY,GACZ,4BAA4B,GAC5B,4BAA4B,GAC5B,gBAAgB,GAChB,8BAA8B,GAC9B,SAAS,GACT,oBAAoB,CAAC;CAC1B;AAQD,wBAAsB,6BAA6B,CACjD,OAAO,EAAE,UAAU,GAClB,OAAO,CAAC,UAAU,CAAC,CAyDrB;AAED,wBAAgB,WAAW,SAgC1B"}

package/dist/tool-call-service/index.d.ts

@@ -0,0 +1,21 @@
+import { Tool, ToolResult } from "@empiricalrun/llm/chat";
+import { SupportedChatModels } from "../agent/chat/types";
+export type { SupportedChatModels };
+type ToolExecutors = {
+    [key: string]: (input: any) => Promise<ToolResult>;
+};
+export declare class ToolCallService {
+    tools: Tool[];
+    toolExecutors: ToolExecutors;
+    constructor();
+    getTools(selectedModel: SupportedChatModels): Promise<{
+        tools: Tool[];
+    }>;
+    execute(payload: {
+        tool: {
+            name: string;
+            input: any;
+        };
+    }, isRemote: boolean): Promise<ToolResult>;
+}
+//# sourceMappingURL=index.d.ts.map

package/dist/tool-call-service/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/tool-call-service/index.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAE1D,OAAO,EAAE,mBAAmB,EAAE,MAAM,qBAAqB,CAAC;AAa1D,YAAY,EAAE,mBAAmB,EAAE,CAAC;AAEpC,KAAK,aAAa,GAAG;IACnB,CAAC,GAAG,EAAE,MAAM,GAAG,CAAC,KAAK,EAAE,GAAG,KAAK,OAAO,CAAC,UAAU,CAAC,CAAC;CACpD,CAAC;AAqBF,qBAAa,eAAe;IAC1B,KAAK,EAAE,IAAI,EAAE,CAAM;IACnB,aAAa,EAAE,aAAa,CAAM;;IAa5B,QAAQ,CAAC,aAAa,EAAE,mBAAmB;;;IAa3C,OAAO,CACX,OAAO,EAAE;QACP,IAAI,EAAE;YACJ,IAAI,EAAE,MAAM,CAAC;YACb,KAAK,EAAE,GAAG,CAAC;SACZ,CAAC;KACH,EACD,QAAQ,EAAE,OAAO,GAChB,OAAO,CAAC,UAAU,CAAC;CA8BvB"}

package/dist/tool-call-service/index.js

@@ -0,0 +1,87 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.ToolCallService = void 0;
+const client_sqs_1 = require("@aws-sdk/client-sqs");
+const commit_and_create_pr_1 = require("../tools/commit-and-create-pr");
+const diagnosis_fetcher_1 = require("../tools/diagnosis-fetcher");
+const environment_crud_1 = require("../tools/environment-crud");
+const grep_1 = require("../tools/grep");
+const str_replace_editor_1 = require("../tools/str_replace_editor");
+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");
+async function sendToolRequestToRemoteQueue(toolName, input) {
+    const sqs = new client_sqs_1.SQSClient({
+        region: process.env.AWS_REGION,
+        credentials: {
+            accessKeyId: process.env.AWS_ACCESS_KEY_ID,
+            secretAccessKey: process.env.AWS_SECRET_ACCESS_KEY,
+        },
+    });
+    const queueUrl = "https://sqs.us-east-1.amazonaws.com/381492172454/toolRequests.fifo";
+    await sqs.send(new client_sqs_1.SendMessageCommand({
+        QueueUrl: queueUrl,
+        MessageBody: JSON.stringify({ toolName, input }),
+        MessageGroupId: toolName, // tool request id
+        MessageDeduplicationId: toolName, // tool request id
+    }));
+}
+class ToolCallService {
+    tools = [];
+    toolExecutors = {};
+    constructor() {
+        this.tools = [
+            grep_1.grepTool,
+            test_run_1.runTestTool,
+            test_run_fetcher_1.fetchTestRunReportTool,
+            diagnosis_fetcher_1.fetchDiagnosisReportTool,
+            test_gen_browser_1.generateTestWithBrowserAgent,
+            commit_and_create_pr_1.commitAndPushChangesTool,
+            environment_crud_1.getEnvironmentTool,
+        ];
+    }
+    async getTools(selectedModel) {
+        if (!selectedModel.startsWith("claude")) {
+            this.tools.push(...str_replace_editor_1.textEditorTools);
+        }
+        this.tools.forEach((tool) => {
+            this.toolExecutors[tool.schema.name] = tool.execute;
+        });
+        if (selectedModel.startsWith("claude")) {
+            this.toolExecutors["str_replace_editor"] = str_replace_editor_1.strReplaceEditorExecutor;
+        }
+        return { tools: this.tools };
+    }
+    async execute(payload, isRemote) {
+        const { tool } = payload;
+        const toolExecutor = this.toolExecutors[tool.name];
+        if (!toolExecutor) {
+            return {
+                isError: true,
+                result: `Invalid function/tool call: invalid_tool_call not found`,
+            };
+        }
+        try {
+            if (isRemote) {
+                console.log("Executing tool remotely", tool.name, tool.input);
+                // push to sqs
+                await sendToolRequestToRemoteQueue(tool.name, tool.input);
+                // TODO: Need to stop the agent loop here
+                return {
+                    isError: false,
+                    result: `Tool request sent to remote queue to execute ${tool.name}.`,
+                };
+            }
+            else {
+                return await toolExecutor(tool.input);
+            }
+        }
+        catch (error) {
+            return {
+                isError: true,
+                result: error instanceof Error ? error.message : String(error),
+            };
+        }
+    }
+}
+exports.ToolCallService = ToolCallService;

package/dist/tools/environment-crud.d.ts

@@ -0,0 +1,4 @@
+import type { Tool } from "@empiricalrun/llm/chat";
+export declare const getEnvironmentTool: Tool;
+export declare const environmentTools: Tool[];
+//# sourceMappingURL=environment-crud.d.ts.map

package/dist/tools/environment-crud.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"environment-crud.d.ts","sourceRoot":"","sources":["../../src/tools/environment-crud.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAyDnD,eAAO,MAAM,kBAAkB,EAAE,IAwEhC,CAAC;AAGF,eAAO,MAAM,gBAAgB,EAAE,IAAI,EAAyB,CAAC"}

package/dist/tools/environment-crud.js

@@ -0,0 +1,98 @@
+"use strict";
+var __importDefault = (this && this.__importDefault) || function (mod) {
+    return (mod && mod.__esModule) ? mod : { "default": mod };
+};
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.environmentTools = exports.getEnvironmentTool = void 0;
+const fs_1 = __importDefault(require("fs"));
+const zod_1 = require("zod");
+const utils_1 = require("./utils");
+const getProjectRepoName = () => {
+    const packageJson = fs_1.default.readFileSync("package.json", "utf8");
+    if (!packageJson) {
+        throw new Error("Could not find or read package.json file");
+    }
+    const packageJsonData = JSON.parse(packageJson);
+    if (!packageJsonData.name) {
+        throw new Error("package.json does not contain a name field");
+    }
+    return packageJsonData.name;
+};
+const GetEnvironmentSchema = zod_1.z.object({
+    environment_slug: zod_1.z
+        .string()
+        .describe("The unique identifier (slug) for the environment you want to fetch details for. This is typically a URL-friendly version of the environment name."),
+});
+// Get Environment Tool
+exports.getEnvironmentTool = {
+    schema: {
+        name: "getEnvironment",
+        description: "Fetch details of an existing environment",
+        parameters: GetEnvironmentSchema,
+    },
+    execute: async (input) => {
+        // Get project repo name
+        let projectRepoName;
+        try {
+            projectRepoName = getProjectRepoName();
+        }
+        catch (error) {
+            return {
+                isError: true,
+                result: `Failed to get project repository name from package.json: ${error instanceof Error ? error.message : String(error)}`,
+            };
+        }
+        // Make API request
+        let response;
+        try {
+            const queryParams = new URLSearchParams({
+                project_repo_name: projectRepoName,
+                environment_slug: input.environment_slug,
+            });
+            response = await (0, utils_1.makeDashboardRequest)({
+                path: `/api/environments?${queryParams.toString()}`,
+                method: "GET",
+            });
+        }
+        catch (error) {
+            return {
+                isError: true,
+                result: `Failed to make API request to fetch environment: ${error instanceof Error ? error.message : String(error)}`,
+            };
+        }
+        // Parse and validate response
+        try {
+            const data = response;
+            if (data.error) {
+                return {
+                    isError: true,
+                    result: `API returned error while fetching environment: ${data.error.message}`,
+                    data: data.error,
+                };
+            }
+            if (!data.data?.environment) {
+                return {
+                    isError: true,
+                    result: `Environment "${input.environment_slug}" not found in project "${projectRepoName}"`,
+                    data: data.data,
+                };
+            }
+            return {
+                result: JSON.stringify({
+                    message: `Successfully fetched environment "${input.environment_slug}"`,
+                    data: data.data,
+                }),
+                isError: false,
+                data: data.data,
+            };
+        }
+        catch (error) {
+            return {
+                isError: true,
+                result: `Failed to parse or validate environment response data: ${error instanceof Error ? error.message : String(error)}`,
+            };
+        }
+    },
+};
+// Export all environment tools as a single array
+exports.environmentTools = [exports.getEnvironmentTool];

package/dist/tools/str_replace_editor.d.ts

@@ -17,7 +17,7 @@ export declare function cleanupBackupFiles(repoDir: string): number;
  * Our implementation of Claude's built-in text editor tool
  * https://docs.anthropic.com/en/docs/build-with-claude/tool-use/text-editor-tool
  */
-export declare function strReplaceEditorExecutor(input: StrReplaceInput, typeChecker?: (filePath: string) => string[]): Promise<ToolResult>;
+export declare function strReplaceEditorExecutor(input: StrReplaceInput): Promise<ToolResult>;
 export declare const textEditorTools: Tool[];
 export {};
 //# sourceMappingURL=str_replace_editor.d.ts.map

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

@@ -1 +1 @@
-{"version":3,"file":"str_replace_editor.d.ts","sourceRoot":"","sources":["../../src/tools/str_replace_editor.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAyB1D,UAAU,eAAe;IACvB,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,UAAU,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC9B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AAED;;;GAGG;AACH,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,CAwC1D;AAMD;;;GAGG;AACH,wBAAsB,wBAAwB,CAC5C,KAAK,EAAE,eAAe,EACtB,WAAW,CAAC,EAAE,CAAC,QAAQ,EAAE,MAAM,KAAK,MAAM,EAAE,GAC3C,OAAO,CAAC,UAAU,CAAC,CA8IrB;AA6GD,eAAO,MAAM,eAAe,EAAE,IAAI,EAKjC,CAAC"}
+{"version":3,"file":"str_replace_editor.d.ts","sourceRoot":"","sources":["../../src/tools/str_replace_editor.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AA2B1D,UAAU,eAAe;IACvB,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,UAAU,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC9B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AAED;;;GAGG;AACH,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,CAwC1D;AAMD;;;GAGG;AACH,wBAAsB,wBAAwB,CAC5C,KAAK,EAAE,eAAe,GACrB,OAAO,CAAC,UAAU,CAAC,CAwIrB;AA+FD,eAAO,MAAM,eAAe,EAAE,IAAI,EAKjC,CAAC"}

package/dist/tools/str_replace_editor.js

@@ -7,6 +7,7 @@ exports.textEditorTools = exports.strReplaceEditorExecutor = exports.cleanupBack
 const fs_1 = __importDefault(require("fs"));
 const path_1 = __importDefault(require("path"));
 const zod_1 = require("zod");
+const web_1 = require("../bin/utils/platform/web");
 function createBackup(filePath) {
     const backupPath = `${filePath}.bak`;
     if (fs_1.default.existsSync(filePath)) {
@@ -79,7 +80,7 @@ function escapeRegExp(text) {
  * Our implementation of Claude's built-in text editor tool
  * https://docs.anthropic.com/en/docs/build-with-claude/tool-use/text-editor-tool
  */
-async function strReplaceEditorExecutor(input, typeChecker) {
+async function strReplaceEditorExecutor(input) {
     const { path: filePath } = input;
     try {
         let content;
@@ -156,25 +157,19 @@ async function strReplaceEditorExecutor(input, typeChecker) {
                     }
                     newContent = content.replace(input.old_str, input.new_str);
                     fs_1.default.writeFileSync(filePath, newContent);
-                    if (typeChecker) {
-                        const errors = typeChecker(filePath);
-                        if (errors.length > 0) {
-                            return {
-                                result: `Edits to file ${filePath} have been applied. However, type checks are failing with errors:\n${errors.join("\n")}`,
-                                isError: true,
-                            };
-                        }
-                        else {
-                            return {
-                                result: `Edits to file ${filePath} have been applied. Type checks have also passed.`,
-                                isError: false,
-                            };
-                        }
+                    const errors = (0, web_1.validateTypescript)(filePath);
+                    if (errors.length > 0) {
+                        return {
+                            result: `Edits to file ${filePath} have been applied. However, type checks are failing with errors:\n${errors.join("\n")}`,
+                            isError: true,
+                        };
+                    }
+                    else {
+                        return {
+                            result: `Edits to file ${filePath} have been applied. Type checks have also passed.`,
+                            isError: false,
+                        };
                     }
-                    return {
-                        result: `Edits to file ${filePath} have been applied.`,
-                        isError: false,
-                    };
                 }
             case "insert":
                 if (input.insert_line === undefined || !input.new_str) {
@@ -245,12 +240,12 @@ const fileCreateTool = {
             file_text: zod_1.z.string().describe("The contents of the new file."),
         }),
     },
-    execute: async (input, typeChecker) => {
+    execute: async (input) => {
         return strReplaceEditorExecutor({
             command: "create",
             path: input.path,
             file_text: input.file_text,
-        }, typeChecker);
+        });
     },
 };
 const stringReplaceTool = {
@@ -264,13 +259,13 @@ in the file. If old_str is not unique, the tool will return an error.`,
             new_str: zod_1.z.string().describe("The string to replace old_str with."),
         }),
     },
-    execute: async (input, typeChecker) => {
+    execute: async (input) => {
         return strReplaceEditorExecutor({
             command: "str_replace",
             path: input.path,
             old_str: input.old_str,
             new_str: input.new_str,
-        }, typeChecker);
+        });
     },
 };
 const stringInsertTool = {
@@ -287,13 +282,13 @@ const stringInsertTool = {
             new_str: zod_1.z.string().describe("The string to insert."),
         }),
     },
-    execute: async (input, typeChecker) => {
+    execute: async (input) => {
         return strReplaceEditorExecutor({
             command: "insert",
             path: input.path,
             insert_line: input.insert_line,
             new_str: input.new_str,
-        }, typeChecker);
+        });
     },
 };
 exports.textEditorTools = [

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

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/utils/index.ts"],"names":[],"mappings":"AAAA,wBAAsB,oBAAoB,CAAC,CAAC,EAAE,EAC5C,IAAI,EACJ,MAAc,EACd,IAAI,GACL,EAAE;IACD,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,GAAG,CAAC;CACZ,GAAG,OAAO,CAAC,CAAC,CAAC,CAmBb;AAED,wBAAsB,eAAe,CAAC,EACpC,MAAM,EACN,GAAG,EACH,IAAI,GACL,EAAE;IACD,MAAM,EAAE,MAAM,CAAC;IACf,GAAG,EAAE,MAAM,CAAC;IACZ,IAAI,CAAC,EAAE,GAAG,CAAC;CACZ,oBAWA"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/tools/utils/index.ts"],"names":[],"mappings":"AAAA,wBAAsB,oBAAoB,CAAC,CAAC,EAAE,EAC5C,IAAI,EACJ,MAAc,EACd,IAAI,GACL,EAAE;IACD,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,GAAG,CAAC;CACZ,GAAG,OAAO,CAAC,CAAC,CAAC,CAwBb;AAED,wBAAsB,eAAe,CAAC,EACpC,MAAM,EACN,GAAG,EACH,IAAI,GACL,EAAE;IACD,MAAM,EAAE,MAAM,CAAC;IACf,GAAG,EAAE,MAAM,CAAC;IACZ,IAAI,CAAC,EAAE,GAAG,CAAC;CACZ,oBAWA"}

package/dist/tools/utils/index.js

@@ -2,12 +2,16 @@
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.callGitHubProxy = exports.makeDashboardRequest = void 0;
 async function makeDashboardRequest({ path, method = "GET", body, }) {
+    const API_KEY = process.env.EMPIRICALRUN_API_KEY;
+    if (!API_KEY) {
+        throw new Error("EMPIRICALRUN_API_KEY is not found");
+    }
     const requestHeaders = {
         "Content-Type": "application/json",
-        Authorization: `Bearer ${process.env.EMPIRICALRUN_API_KEY}`,
+        Authorization: `Bearer ${API_KEY}`,
         "User-Agent": "empiricalrun/test-gen",
     };
-    const baseUrl = "https://dash.empirical.run";
+    const baseUrl = process.env.DASHBOARD_DOMAIN || "https://dash.empirical.run";
     const response = await fetch(`${baseUrl}${path}`, {
         method,
         headers: requestHeaders,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@empiricalrun/test-gen",
-  "version": "0.54.1",
+  "version": "0.55.0",
   "publishConfig": {
     "registry": "https://registry.npmjs.org/",
     "access": "public"
@@ -25,6 +25,10 @@
     ".": {
       "types": "./dist/index.d.ts",
       "default": "./dist/index.js"
+    },
+    "./tool-call-service": {
+      "types": "./dist/tool-call-service/index.d.ts",
+      "default": "./dist/tool-call-service/index.js"
     }
   },
   "repository": {
@@ -33,6 +37,7 @@
   },
   "author": "Empirical Team <hey@empirical.run>",
   "dependencies": {
+    "@aws-sdk/client-sqs": "^3.787.0",
     "@babel/parser": "^7.26.3",
     "@types/sanitize-html": "^2.11.0",
     "async-retry": "^1.3.3",
@@ -56,7 +61,7 @@
     "tsx": "^4.16.2",
     "typescript": "^5.3.3",
     "zod": "^3.23.8",
-    "@empiricalrun/llm": "^0.14.8",
+    "@empiricalrun/llm": "^0.15.0",
     "@empiricalrun/r2-uploader": "^0.3.8",
     "@empiricalrun/test-run": "^0.8.0"
   },