@empiricalrun/test-gen 0.55.0 → 0.56.0

package/CHANGELOG.md

@@ -1,5 +1,22 @@
 # @empiricalrun/test-gen
 
+## 0.56.0
+
+### Minor Changes
+
+- b073084: feat: new APIs and refactor to introduce tool execute service
+
+### Patch Changes
+
+- 93d7a0b: feat: add tool call for downloading builds
+- a58ac3f: feat: add dashboard agent integration and optimize tsx handling
+- Updated dependencies [87af227]
+- Updated dependencies [3831109]
+- Updated dependencies [a58ac3f]
+- Updated dependencies [f77e33d]
+  - @empiricalrun/llm@0.15.1
+  - @empiricalrun/test-run@0.8.1
+
 ## 0.55.0
 
 ### Minor Changes

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":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,KAAK,EAAe,QAAQ,EAAE,MAAM,4BAA4B,CAAC;AAIxE,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAClC,OAAO,EAAE,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AAsBvD,wBAAgB,QAAQ,CAAC,GAAG,EAAE,GAAG,GAAG,GAAG,IAAI,MAAM,CAKhD;AAED,wBAAgB,wBAAwB,CAAC,KAAK,EAAE,MAAM,EAAE,UAIvD;AAiFD,wBAAsB,yBAAyB,CAAC,EAC9C,YAAY,EACZ,YAAY,EACZ,cAAc,GACf,EAAE;IACD,YAAY,EAAE,MAAM,CAAC;IACrB,YAAY,EAAE,MAAM,CAAC;IACrB,cAAc,EAAE,MAAM,EAAE,CAAC;CAC1B,iBA2BA;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;AAED,wBAAsB,yBAAyB,CAAC,EAC9C,QAAQ,EACR,QAAQ,EACR,KAAK,GACN,EAAE;IACD,QAAQ,EAAE,QAAQ,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB,GAAG,OAAO,CAAC,MAAM,CAAC,CAyDlB;AAyBD,wBAAsB,wBAAwB,CAAC,IAAI,EAAE,IAAI,iBA2HxD;AAED;;;GAGG;AACH,wBAAsB,oBAAoB,CACxC,OAAO,EAAE,MAAM,GACd,OAAO,CAAC,oBAAoB,CAAC,CAM/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;AAED,qBAAa,eAAe;IACd,OAAO,CAAC,SAAS;gBAAT,SAAS,EAAE,MAAM;IACrC,OAAO,CAAC,aAAa,CAAqB;YAE5B,mBAAmB;YAUnB,gBAAgB;IAsBjB,OAAO;IAoBb,SAAS;CAKjB"}
+{"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/utils.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,KAAK,EAAe,QAAQ,EAAE,MAAM,4BAA4B,CAAC;AAIxE,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAClC,OAAO,EAAE,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AAwBvD,wBAAgB,QAAQ,CAAC,GAAG,EAAE,GAAG,GAAG,GAAG,IAAI,MAAM,CAKhD;AAED,wBAAgB,wBAAwB,CAAC,KAAK,EAAE,MAAM,EAAE,UAIvD;AAiFD,wBAAsB,yBAAyB,CAAC,EAC9C,YAAY,EACZ,YAAY,EACZ,cAAc,GACf,EAAE;IACD,YAAY,EAAE,MAAM,CAAC;IACrB,YAAY,EAAE,MAAM,CAAC;IACrB,cAAc,EAAE,MAAM,EAAE,CAAC;CAC1B,iBA2BA;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;AAED,wBAAsB,yBAAyB,CAAC,EAC9C,QAAQ,EACR,QAAQ,EACR,KAAK,GACN,EAAE;IACD,QAAQ,EAAE,QAAQ,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB,GAAG,OAAO,CAAC,MAAM,CAAC,CAyDlB;AAyBD,wBAAsB,wBAAwB,CAAC,IAAI,EAAE,IAAI,iBA2HxD;AAED;;;GAGG;AACH,wBAAsB,oBAAoB,CACxC,OAAO,EAAE,MAAM,GACd,OAAO,CAAC,oBAAoB,CAAC,CA0B/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;AAED,qBAAa,eAAe;IACd,OAAO,CAAC,SAAS;gBAAT,SAAS,EAAE,MAAM;IACrC,OAAO,CAAC,aAAa,CAAqB;YAE5B,mBAAmB;YAUnB,gBAAgB;IAsBjB,OAAO;IAoBb,SAAS;CAKjB"}

package/dist/agent/browsing/utils.js

@@ -8,7 +8,8 @@ const fs_extra_1 = __importDefault(require("fs-extra"));
 const minimatch_1 = require("minimatch");
 const path_1 = __importDefault(require("path"));
 const ts_morph_1 = require("ts-morph");
-const api_1 = __importDefault(require("tsx/cjs/api"));
+// For TypeScript type safety
+let tsxImport = null;
 const logger_1 = require("../../bin/logger");
 const context_1 = require("../../bin/utils/context");
 const fs_1 = require("../../bin/utils/fs");
@@ -295,8 +296,25 @@ exports.injectPwLocatorGenerator = injectPwLocatorGenerator;
  * @return {*}  {Promise<PlaywrightTestConfig>}
  */
 async function readPlaywrightConfig(repoDir) {
+    if (typeof window !== "undefined") {
+        throw new Error("readPlaywrightConfig cannot be used in browser environments");
+    }
+    else {
+        // Only initialize on server side
+        // This will only execute on the server
+        await import("tsx/cjs/api")
+            .then((module) => {
+            tsxImport = module;
+        })
+            .catch((err) => {
+            console.error("Failed to import tsx:", err);
+        });
+    }
+    if (!tsxImport) {
+        throw new Error("tsx module not available");
+    }
     const [lastDir] = repoDir.split("/").reverse();
-    const playwrightConfig = (await api_1.default.require("./playwright.config.ts", `${repoDir}/${lastDir}`)).default;
+    const playwrightConfig = (await tsxImport.require("./playwright.config.ts", `${repoDir}/${lastDir}`)).default;
     return playwrightConfig;
 }
 exports.readPlaywrightConfig = readPlaywrightConfig;

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

@@ -1,9 +1,11 @@
 import { TraceClient } from "@empiricalrun/llm";
 import { IChatModel } from "@empiricalrun/llm/chat";
+import { ToolCallService } from "../../tool-call-service";
 import { ReporterFunction, SupportedChatModels } from "./types";
-export declare function chatAgentLoop({ chatModel, selectedModel, reporter, trace, }: {
+export declare function chatAgentLoop({ chatModel, selectedModel, reporter, trace, toolCallService, }: {
     chatModel: IChatModel<any>;
     selectedModel: SupportedChatModels;
+    toolCallService: ToolCallService;
     reporter: ReporterFunction;
     trace?: TraceClient;
 }): Promise<void>;

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;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"}
+{"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;AAGhC,OAAO,EAAE,eAAe,EAAE,MAAM,yBAAyB,CAAC;AAG1D,OAAO,EAAE,gBAAgB,EAAE,mBAAmB,EAAE,MAAM,SAAS,CAAC;AAgBhE,wBAAsB,aAAa,CAAC,EAClC,SAAS,EACT,aAAa,EACb,QAAQ,EACR,KAAK,EACL,eAAe,GAChB,EAAE;IACD,SAAS,EAAE,UAAU,CAAC,GAAG,CAAC,CAAC;IAC3B,aAAa,EAAE,mBAAmB,CAAC;IACnC,eAAe,EAAE,eAAe,CAAC;IACjC,QAAQ,EAAE,gBAAgB,CAAC;IAC3B,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB,iBAoCA"}

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

@@ -3,8 +3,6 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.chatAgentLoop = void 0;
 const chat_1 = require("@empiricalrun/llm/chat");
 const picocolors_1 = require("picocolors");
-const tool_call_service_1 = require("../../tool-call-service");
-const str_replace_editor_1 = require("../../tools/str_replace_editor");
 const prompt_1 = require("./prompt");
 const state_1 = require("./state");
 function getModelName(model) {
@@ -20,31 +18,18 @@ 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, }) {
+async function chatAgentLoop({ chatModel, selectedModel, reporter, trace, toolCallService, }) {
     const systemPrompt = await (0, prompt_1.buildSystemPrompt)();
     trace?.update({ input: { systemPrompt } });
-    const toolCallService = new tool_call_service_1.ToolCallService();
-    const { tools } = await toolCallService.getTools(selectedModel);
+    const { tools } = await toolCallService.getTools();
     while (!chatModel.askUserForInput) {
         const toolCalls = chatModel.getPendingToolCalls();
+        console.log("toolCalls", toolCalls);
         if (toolCalls.length) {
-            const toolResults = [];
-            for (const call of toolCalls) {
-                const args = JSON.stringify(call.input);
-                log(`Executing tool ${call.name} with args: ${args}`);
-                let callResponse = await toolCallService.execute({
-                    tool: {
-                        name: call.name,
-                        input: call.input,
-                    },
-                }, isRemote);
-                if (callResponse.isError) {
-                    log(`Tool ${call.name} failed: ${callResponse.result}`);
-                }
-                else {
-                    log(`Tool ${call.name} completed`);
-                }
-                toolResults.push(callResponse);
+            const toolResults = await toolCallService.execute(toolCalls, isRemote);
+            if (isRemote) {
+                log(`Tool call remote execution in progress`);
+                break;
             }
             chatModel.pushToolResultsMessage(toolCalls, toolResults);
         }
@@ -62,6 +47,5 @@ async function chatAgentLoop({ chatModel, selectedModel, reporter, trace, }) {
         const latest = chatModel.getHumanReadableLatestMessage();
         await reporter((0, state_1.chatStateFromModel)(chatModel, selectedModel), latest);
     }
-    (0, str_replace_editor_1.cleanupBackupFiles)(process.cwd());
 }
 exports.chatAgentLoop = chatAgentLoop;

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

@@ -0,0 +1,8 @@
+import { IChatModel } from "@empiricalrun/llm/chat";
+import { chatAgentLoop } from "./agent-loop";
+import { createChatModel } from "./model";
+import { ChatStateOnDisk } from "./state";
+import { ReporterFunction, SupportedChatModels } from "./types";
+export { chatAgentLoop, createChatModel };
+export type { ChatStateOnDisk, IChatModel, ReporterFunction, SupportedChatModels, };
+//# sourceMappingURL=exports.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"exports.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/exports.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAEpD,OAAO,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAC7C,OAAO,EAAE,eAAe,EAAE,MAAM,SAAS,CAAC;AAC1C,OAAO,EAAE,eAAe,EAAE,MAAM,SAAS,CAAC;AAC1C,OAAO,EAAE,gBAAgB,EAAE,mBAAmB,EAAE,MAAM,SAAS,CAAC;AAEhE,OAAO,EAAE,aAAa,EAAE,eAAe,EAAE,CAAC;AAE1C,YAAY,EACV,eAAe,EACf,UAAU,EACV,gBAAgB,EAChB,mBAAmB,GACpB,CAAC"}

package/dist/agent/chat/exports.js

@@ -0,0 +1,7 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.createChatModel = exports.chatAgentLoop = void 0;
+const agent_loop_1 = require("./agent-loop");
+Object.defineProperty(exports, "chatAgentLoop", { enumerable: true, get: function () { return agent_loop_1.chatAgentLoop; } });
+const model_1 = require("./model");
+Object.defineProperty(exports, "createChatModel", { enumerable: true, get: function () { return model_1.createChatModel; } });

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":"AAYA,OAAO,EAAoB,mBAAmB,EAAE,MAAM,SAAS,CAAC;AAiBhE,wBAAsB,kBAAkB,CAAC,EACvC,mBAAmB,EACnB,aAAa,EACb,oBAAoB,GACrB,EAAE;IACD,aAAa,EAAE,mBAAmB,CAAC;IACnC,mBAAmB,EAAE,OAAO,CAAC;IAC7B,oBAAoB,EAAE,MAAM,GAAG,SAAS,CAAC;CAC1C,iBAoFA;AAuBD,wBAAsB,wBAAwB,CAAC,EAC7C,aAAa,EACb,aAAa,GACd,EAAE;IACD,aAAa,EAAE,mBAAmB,CAAC;IACnC,aAAa,EAAE,MAAM,CAAC;CACvB,iBAyCA"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/index.ts"],"names":[],"mappings":"AAaA,OAAO,EAAoB,mBAAmB,EAAE,MAAM,SAAS,CAAC;AAiBhE,wBAAsB,kBAAkB,CAAC,EACvC,mBAAmB,EACnB,aAAa,EACb,oBAAoB,GACrB,EAAE;IACD,aAAa,EAAE,mBAAmB,CAAC;IACnC,mBAAmB,EAAE,OAAO,CAAC;IAC7B,oBAAoB,EAAE,MAAM,GAAG,SAAS,CAAC;CAC1C,iBAsFA;AAuBD,wBAAsB,wBAAwB,CAAC,EAC7C,aAAa,EACb,aAAa,GACd,EAAE;IACD,aAAa,EAAE,mBAAmB,CAAC;IACnC,aAAa,EAAE,MAAM,CAAC;CACvB,iBA2CA"}

package/dist/agent/chat/index.js

@@ -4,6 +4,7 @@ exports.runChatAgentForDashboard = exports.runChatAgentForCLI = void 0;
 const llm_1 = require("@empiricalrun/llm");
 const picocolors_1 = require("picocolors");
 const human_in_the_loop_1 = require("../../human-in-the-loop");
+const tool_call_service_1 = require("../../tool-call-service");
 const git_1 = require("../../utils/git");
 const agent_loop_1 = require("./agent-loop");
 const model_1 = require("./model");
@@ -83,11 +84,13 @@ async function runChatAgentForCLI({ useDiskForChatState, selectedModel, initialP
         }
         else {
             // TODO: Should we pass a loader function? That would allow us to show a spinner
+            const toolCallService = new tool_call_service_1.ToolCallService(null, selectedModel);
             await (0, agent_loop_1.chatAgentLoop)({
                 chatModel,
                 selectedModel,
                 reporter: reporterFunc,
                 trace,
+                toolCallService,
             });
         }
     }
@@ -125,6 +128,7 @@ async function runChatAgentForDashboard({ chatSessionId, selectedModel, }) {
             chatSessionId,
         },
     });
+    const toolCallService = new tool_call_service_1.ToolCallService(chatSessionId, selectedModel);
     await (0, git_1.checkoutBranch)(branchName);
     let chatModel = (0, model_1.createChatModel)(chatState.messages, selectedModel);
     let reporterFunc = async (chatState, latest) => {
@@ -147,6 +151,7 @@ async function runChatAgentForDashboard({ chatSessionId, selectedModel, }) {
         selectedModel,
         reporter: reporterFunc,
         trace,
+        toolCallService,
     });
     await (0, git_1.commitLocalAndPushBranchToRemote)(branchName);
 }

package/dist/bin/index.js

@@ -233,8 +233,10 @@ async function main() {
         generationId: testGenConfig.options?.metadata.generationId,
         projectRepoName: testGenConfig.options?.metadata.projectRepoName,
     });
-    // Download the build if repo has a download script
-    await (0, test_build_1.downloadBuild)(testGenConfig.build || {});
+    if (testGenConfig.build?.url) {
+        // Download the build if repo has a download script
+        await (0, test_build_1.downloadBuild)(testGenConfig.build.url);
+    }
     if (completedOptions.useChat) {
         await runChatAgent({
             chatSessionId: completedOptions.chatSessionId,

package/dist/test-build/index.d.ts

@@ -1,10 +1,3 @@
-import { Build } from "@empiricalrun/shared-types";
-/**
- * method to download the build from the URL provided in the build object
- * this is only used in cases like chrome extension as of yet.
- * @export
- * @param {Build} build
- * @return {*}  {Promise<void>}
- */
-export declare function downloadBuild(build: Build): Promise<void>;
+export declare function hasDownloadScript(): Promise<boolean>;
+export declare function downloadBuild(buildUrl: string): Promise<void>;
 //# sourceMappingURL=index.d.ts.map

package/dist/test-build/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/test-build/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAE,MAAM,4BAA4B,CAAC;AAUnD;;;;;;GAMG;AACH,wBAAsB,aAAa,CAAC,KAAK,EAAE,KAAK,GAAG,OAAO,CAAC,IAAI,CAAC,CAY/D"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/test-build/index.ts"],"names":[],"mappings":"AAeA,wBAAsB,iBAAiB,IAAI,OAAO,CAAC,OAAO,CAAC,CAG1D;AAED,wBAAsB,aAAa,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAUnE"}

package/dist/test-build/index.js

@@ -3,26 +3,27 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
     return (mod && mod.__esModule) ? mod : { "default": mod };
 };
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.downloadBuild = void 0;
+exports.downloadBuild = exports.hasDownloadScript = void 0;
 const fs_extra_1 = __importDefault(require("fs-extra"));
 const logger_1 = require("../bin/logger");
 const exec_1 = require("../utils/exec");
-/**
- * method to download the build from the URL provided in the build object
- * this is only used in cases like chrome extension as of yet.
- * @export
- * @param {Build} build
- * @return {*}  {Promise<void>}
- */
-async function downloadBuild(build) {
-    const logger = new logger_1.CustomLogger({ useReporter: false });
+async function getPackageJSON() {
     const packageJSONPath = "package.json";
     const packageJsonStr = await fs_extra_1.default.readFile(packageJSONPath, "utf-8");
-    const packageJSONData = JSON.parse(packageJsonStr);
-    const buildDownloadScript = packageJSONData.scripts["download"];
-    if (buildDownloadScript && build?.url) {
-        logger.log(`Downloading build from ${build.url}`);
-        await (0, exec_1.cmd)(`npm run download ${build.url}`.split(" "), {
+    return JSON.parse(packageJsonStr);
+}
+async function hasDownloadScript() {
+    const packageJSON = await getPackageJSON();
+    return !!packageJSON.scripts["download"];
+}
+exports.hasDownloadScript = hasDownloadScript;
+async function downloadBuild(buildUrl) {
+    const logger = new logger_1.CustomLogger({ useReporter: false });
+    const packageJSON = await getPackageJSON();
+    const buildDownloadScript = packageJSON.scripts["download"];
+    if (buildDownloadScript && buildUrl) {
+        logger.log(`Downloading build from ${buildUrl}`);
+        await (0, exec_1.cmd)(`npm run download ${buildUrl}`.split(" "), {
             env: { ...Object(process.env) },
         });
     }

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

@@ -1,4 +1,4 @@
-import { Tool, ToolResult } from "@empiricalrun/llm/chat";
+import { PendingToolCall, Tool, ToolResult } from "@empiricalrun/llm/chat";
 import { SupportedChatModels } from "../agent/chat/types";
 export type { SupportedChatModels };
 type ToolExecutors = {
@@ -7,15 +7,12 @@ type ToolExecutors = {
 export declare class ToolCallService {
     tools: Tool[];
     toolExecutors: ToolExecutors;
-    constructor();
-    getTools(selectedModel: SupportedChatModels): Promise<{
+    chatSessionId: number | null;
+    selectedModel: SupportedChatModels;
+    constructor(chatSessionId: number | null, selectedModel: SupportedChatModels);
+    getTools(): Promise<{
         tools: Tool[];
     }>;
-    execute(payload: {
-        tool: {
-            name: string;
-            input: any;
-        };
-    }, isRemote: boolean): Promise<ToolResult>;
+    execute(toolCalls: PendingToolCall[], isRemote: boolean): Promise<ToolResult[]>;
 }
 //# sourceMappingURL=index.d.ts.map

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

@@ -1 +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"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/tool-call-service/index.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,eAAe,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAE3E,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;AAyBF,qBAAa,eAAe;IAC1B,KAAK,EAAE,IAAI,EAAE,CAAM;IACnB,aAAa,EAAE,aAAa,CAAM;IAClC,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,aAAa,EAAE,mBAAmB,CAAC;gBAGjC,aAAa,EAAE,MAAM,GAAG,IAAI,EAC5B,aAAa,EAAE,mBAAmB;IAgB9B,QAAQ;;;IAaR,OAAO,CACX,SAAS,EAAE,eAAe,EAAE,EAC5B,QAAQ,EAAE,OAAO,GAChB,OAAO,CAAC,UAAU,EAAE,CAAC;CAqCzB"}

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

@@ -4,13 +4,14 @@ 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 download_build_1 = require("../tools/download-build");
 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) {
+async function sendToolRequestToRemoteQueue(payload) {
     const sqs = new client_sqs_1.SQSClient({
         region: process.env.AWS_REGION,
         credentials: {
@@ -18,18 +19,22 @@ async function sendToolRequestToRemoteQueue(toolName, input) {
             secretAccessKey: process.env.AWS_SECRET_ACCESS_KEY,
         },
     });
-    const queueUrl = "https://sqs.us-east-1.amazonaws.com/381492172454/toolRequests.fifo";
+    const queueUrl = process.env.TOOL_REQUEST_QUEUE_URL;
     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
+        MessageBody: JSON.stringify(payload),
+        MessageGroupId: payload.requestId,
+        MessageDeduplicationId: payload.requestId, // unique id for the tool request
     }));
 }
 class ToolCallService {
     tools = [];
     toolExecutors = {};
-    constructor() {
+    chatSessionId;
+    selectedModel;
+    constructor(chatSessionId, selectedModel) {
+        this.chatSessionId = chatSessionId;
+        this.selectedModel = selectedModel;
         this.tools = [
             grep_1.grepTool,
             test_run_1.runTestTool,
@@ -38,49 +43,58 @@ class ToolCallService {
             test_gen_browser_1.generateTestWithBrowserAgent,
             commit_and_create_pr_1.commitAndPushChangesTool,
             environment_crud_1.getEnvironmentTool,
+            download_build_1.downloadBuildTool,
         ];
     }
-    async getTools(selectedModel) {
-        if (!selectedModel.startsWith("claude")) {
+    async getTools() {
+        if (!this.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")) {
+        if (this.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`,
-            };
+    async execute(toolCalls, isRemote) {
+        if (isRemote && this.chatSessionId) {
+            console.log("Executing tool remotely", toolCalls);
+            await sendToolRequestToRemoteQueue({
+                toolCalls,
+                requestId: crypto.randomUUID(),
+                chatSessionId: this.chatSessionId,
+                selectedModel: this.selectedModel,
+            });
+            return toolCalls.map(() => ({
+                isError: false,
+                result: `Tool request sent to remote queue to execute.`,
+            }));
         }
-        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 {
+            const toolResults = [];
+            for (const toolCall of toolCalls) {
+                const toolExecutor = this.toolExecutors[toolCall.name];
+                if (!toolExecutor) {
+                    toolResults.push({
+                        isError: true,
+                        result: `Invalid function/tool call: invalid_tool_call not found`,
+                    });
+                    continue;
+                }
+                try {
+                    const result = await toolExecutor(toolCall.input);
+                    toolResults.push(result);
+                }
+                catch (error) {
+                    toolResults.push({
+                        isError: true,
+                        result: error instanceof Error ? error.message : String(error),
+                    });
+                }
             }
-            else {
-                return await toolExecutor(tool.input);
-            }
-        }
-        catch (error) {
-            return {
-                isError: true,
-                result: error instanceof Error ? error.message : String(error),
-            };
+            return toolResults;
         }
     }
 }

package/dist/tools/download-build.d.ts

@@ -0,0 +1,3 @@
+import type { Tool } from "@empiricalrun/llm/chat";
+export declare const downloadBuildTool: Tool;
+//# sourceMappingURL=download-build.d.ts.map

package/dist/tools/download-build.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"download-build.d.ts","sourceRoot":"","sources":["../../src/tools/download-build.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAKnD,eAAO,MAAM,iBAAiB,EAAE,IAkC/B,CAAC"}

package/dist/tools/download-build.js

@@ -0,0 +1,39 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.downloadBuildTool = void 0;
+const zod_1 = require("zod");
+const test_build_1 = require("../test-build");
+exports.downloadBuildTool = {
+    schema: {
+        name: "downloadBuild",
+        description: `Download a build from a build URL. If you do not have
+have a build URL, you can try getting the environment details with the getEnvironment tool.
+Environment details will include the build URL.`,
+        parameters: zod_1.z.object({
+            buildUrl: zod_1.z.string().describe("The URL of the build to download"),
+        }),
+    },
+    execute: async (input) => {
+        if (!(await (0, test_build_1.hasDownloadScript)())) {
+            return {
+                isError: true,
+                result: `This repo does not have a download script in package.json.
+You probably don't need to worry about this, since it means this repo does not have a build to download.`,
+            };
+        }
+        const { buildUrl } = input;
+        try {
+            await (0, test_build_1.downloadBuild)(buildUrl);
+            return {
+                isError: false,
+                result: "Build downloaded successfully",
+            };
+        }
+        catch (error) {
+            return {
+                isError: true,
+                result: `Failed to download build: ${error}`,
+            };
+        }
+    },
+};

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,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"}
+{"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,CAuBb;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/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@empiricalrun/test-gen",
-  "version": "0.55.0",
+  "version": "0.56.0",
   "publishConfig": {
     "registry": "https://registry.npmjs.org/",
     "access": "public"
@@ -18,6 +18,10 @@
       "types": "./dist/agent/chat/state.d.ts",
       "default": "./dist/agent/chat/state.js"
     },
+    "./chat": {
+      "types": "./dist/agent/chat/exports.d.ts",
+      "default": "./dist/agent/chat/exports.js"
+    },
     "./utils": {
       "types": "./dist/utils/index.d.ts",
       "default": "./dist/utils/index.js"
@@ -61,9 +65,9 @@
     "tsx": "^4.16.2",
     "typescript": "^5.3.3",
     "zod": "^3.23.8",
-    "@empiricalrun/llm": "^0.15.0",
+    "@empiricalrun/llm": "^0.15.1",
     "@empiricalrun/r2-uploader": "^0.3.8",
-    "@empiricalrun/test-run": "^0.8.0"
+    "@empiricalrun/test-run": "^0.8.1"
   },
   "devDependencies": {
     "@playwright/test": "1.47.1",