@empiricalrun/test-gen 0.53.8 → 0.53.10

package/CHANGELOG.md

@@ -1,5 +1,22 @@
 # @empiricalrun/test-gen
 
+## 0.53.10
+
+### Patch Changes
+
+- 088545c: feat: upload test reports from run-test tool calls
+- d003ea8: feat: add description to PRs created by chat agent
+- Updated dependencies [088545c]
+  - @empiricalrun/test-run@0.7.7
+
+## 0.53.9
+
+### Patch Changes
+
+- 2e5d412: fix: grep tool should first return stdout, not stderr
+- 2619c58: fix: add selected model info to chat state
+- 58d170d: fix: skip all serial tests after browser agent is done
+
 ## 0.53.8
 
 ### Patch Changes

package/dist/agent/browsing/run.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/run.ts"],"names":[],"mappings":"AAiBA,KAAK,iBAAiB,GAAG;IACvB,YAAY,EAAE,MAAM,CAAC;IACrB,gBAAgB,EAAE,MAAM,CAAC;IACzB,gBAAgB,EAAE,MAAM,EAAE,CAAC;IAC3B,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,MAAM,CAAC;IAChB,yBAAyB,EAAE,OAAO,CAAC;CACpC,CAAC;AAEF,wBAAsB,6BAA6B,CAAC,EAClD,YAAY,EACZ,gBAAgB,EAChB,gBAAgB,EAChB,YAAY,EACZ,OAAO,EACP,yBAAyB,GAC1B,EAAE,iBAAiB,GAAG,OAAO,CAAC;IAC7B,OAAO,EAAE,OAAO,CAAC;IACjB,KAAK,EAAE,MAAM,CAAC;IACd,cAAc,CAAC,EAAE,MAAM,CAAC;CACzB,CAAC,CAqFD"}
+{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/agent/browsing/run.ts"],"names":[],"mappings":"AAiBA,KAAK,iBAAiB,GAAG;IACvB,YAAY,EAAE,MAAM,CAAC;IACrB,gBAAgB,EAAE,MAAM,CAAC;IACzB,gBAAgB,EAAE,MAAM,EAAE,CAAC;IAC3B,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,MAAM,CAAC;IAChB,yBAAyB,EAAE,OAAO,CAAC;CACpC,CAAC;AAEF,wBAAsB,6BAA6B,CAAC,EAClD,YAAY,EACZ,gBAAgB,EAChB,gBAAgB,EAChB,YAAY,EACZ,OAAO,EACP,yBAAyB,GAC1B,EAAE,iBAAiB,GAAG,OAAO,CAAC;IAC7B,OAAO,EAAE,OAAO,CAAC;IACjB,KAAK,EAAE,MAAM,CAAC;IACd,cAAc,CAAC,EAAE,MAAM,CAAC;CACzB,CAAC,CAsFD"}

package/dist/agent/browsing/run.js

@@ -14,14 +14,15 @@ async function generateTestsUsingMasterAgent({ testFilePath, filePathToUpdate, p
     if (!fs_extra_1.default.existsSync(testFilePath)) {
         throw new Error(`File for master agent to run not found: ${testFilePath}`);
     }
-    // detect available http port on the machine
-    const port = await (0, detect_port_1.default)(3030);
+    const pm = new exec_1.ProcessManager();
+    const availablePort = await (0, detect_port_1.default)(3030);
     // start a file service to handle file updates from agent
     // - also update the file path with updates when agent is done spitting out code
     const fileServer = new server_1.FileServiceServer({
-        port,
+        port: availablePort,
         repoDir,
         updateFile: editFileWithGeneratedCode,
+        onComplete: () => pm.terminate(),
     });
     await fileServer.startFileService();
     fileServer.setFilePath(filePathToUpdate);
@@ -44,9 +45,9 @@ async function generateTestsUsingMasterAgent({ testFilePath, filePathToUpdate, p
         if (!isTestRunTriggeredForTeardown) {
             removeListeners = await teardowns.skipAll();
         }
-        await (0, exec_1.cmd)(command.split(" "), {
+        await pm.execute(command.split(" "), {
             env: {
-                IPC_FILE_SERVICE_PORT: port.toString(),
+                IPC_FILE_SERVICE_PORT: availablePort.toString(),
                 PW_TEST_HTML_REPORT_OPEN: "never",
                 // pass the test gen token so that the agent has the same configuration as cli
                 TEST_GEN_TOKEN: testGenToken,

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

@@ -84,7 +84,7 @@ async function chatAgentLoop({ chatModel, selectedModel, reporter, trace, }) {
         }
         chatModel.pushMessage(response);
         const latest = chatModel.getHumanReadableLatestMessage();
-        await reporter((0, state_1.chatStateFromModel)(chatModel), latest);
+        await reporter((0, state_1.chatStateFromModel)(chatModel, selectedModel), latest);
     }
     (0, chat_1.cleanupBackupFiles)(process.cwd());
 }

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;AAgBhE,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;AAqBD,wBAAsB,wBAAwB,CAAC,EAC7C,aAAa,EACb,aAAa,GACd,EAAE;IACD,aAAa,EAAE,mBAAmB,CAAC;IACnC,aAAa,EAAE,MAAM,CAAC;CACvB,iBA+BA"}
+{"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;AAqBD,wBAAsB,wBAAwB,CAAC,EAC7C,aAAa,EACb,aAAa,GACd,EAAE;IACD,aAAa,EAAE,mBAAmB,CAAC;IACnC,aAAa,EAAE,MAAM,CAAC;CACvB,iBA+BA"}

package/dist/agent/chat/index.js

@@ -11,10 +11,10 @@ const state_1 = require("./state");
 function stopCriteria(userPrompt) {
     return userPrompt?.toLowerCase() === "stop";
 }
-function concludeAgent(chatModel, useDiskForChatState) {
+function concludeAgent(chatModel, useDiskForChatState, selectedModel) {
     console.log(`\n${(0, picocolors_1.gray)("Usage summary -> " + chatModel.getUsageSummary())}`);
     if (useDiskForChatState) {
-        (0, state_1.saveToDisk)(chatModel.messages);
+        (0, state_1.saveToDisk)(chatModel.messages, selectedModel);
     }
 }
 async function runChatAgentForCLI({ useDiskForChatState, selectedModel, initialPromptContent, }) {
@@ -39,7 +39,7 @@ async function runChatAgentForCLI({ useDiskForChatState, selectedModel, initialP
         }
     }
     const handleSigInt = () => {
-        concludeAgent(chatModel, useDiskForChatState);
+        concludeAgent(chatModel, useDiskForChatState, selectedModel);
         process.exit(0);
     };
     process.once("SIGINT", handleSigInt);
@@ -47,7 +47,7 @@ async function runChatAgentForCLI({ useDiskForChatState, selectedModel, initialP
     let userPrompt;
     let reporterFunc = async (chatState, latest) => {
         if (useDiskForChatState) {
-            (0, state_1.saveToDisk)(chatState.messages);
+            (0, state_1.saveToDisk)(chatState.messages, selectedModel);
         }
         if (latest) {
             console.log(`${(0, picocolors_1.blue)(latest.role)}: ${latest.textMessage}`);
@@ -72,7 +72,7 @@ async function runChatAgentForCLI({ useDiskForChatState, selectedModel, initialP
             catch (e) {
                 // https://github.com/SBoudrias/Inquirer.js/issues/1502#issuecomment-2275991680
                 if (e instanceof Error && e.name === "ExitPromptError") {
-                    concludeAgent(chatModel, useDiskForChatState);
+                    concludeAgent(chatModel, useDiskForChatState, selectedModel);
                     process.exit(0);
                 }
                 throw e;

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

@@ -4,11 +4,12 @@ export declare const CURRENT_CHAT_STATE_VERSION = "20250327.1";
 export declare const CHAT_STATE_PATH: string;
 export type ChatStateOnDisk<T> = {
     version: typeof CURRENT_CHAT_STATE_VERSION;
+    model: SupportedChatModels;
     messages: T[];
 };
 export declare function createChatState(userPrompt: string, existingState: ChatStateOnDisk<any>, selectedModel: SupportedChatModels): ChatStateOnDisk<unknown>;
-export declare function createChatStateForMessages<T>(messages: any): ChatStateOnDisk<T>;
-export declare function chatStateFromModel<T>(chatModel: IChatModel<T>): ChatStateOnDisk<unknown>;
+export declare function createChatStateForMessages<T>(messages: any, selectedModel: SupportedChatModels): ChatStateOnDisk<T>;
+export declare function chatStateFromModel<T>(chatModel: IChatModel<T>, selectedModel: SupportedChatModels): ChatStateOnDisk<unknown>;
 export declare function loadChatState<T>(): ChatStateOnDisk<T> | undefined;
-export declare function saveToDisk<T>(messages: Array<T>): void;
+export declare function saveToDisk<T>(messages: Array<T>, selectedModel: SupportedChatModels): void;
 //# sourceMappingURL=state.d.ts.map

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

@@ -1 +1 @@
-{"version":3,"file":"state.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/state.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAKpD,OAAO,EAAE,mBAAmB,EAAE,MAAM,SAAS,CAAC;AAE9C,eAAO,MAAM,0BAA0B,eAAe,CAAC;AAEvD,eAAO,MAAM,eAAe,QAI3B,CAAC;AAEF,MAAM,MAAM,eAAe,CAAC,CAAC,IAAI;IAC/B,OAAO,EAAE,OAAO,0BAA0B,CAAC;IAC3C,QAAQ,EAAE,CAAC,EAAE,CAAC;CACf,CAAC;AAEF,wBAAgB,eAAe,CAC7B,UAAU,EAAE,MAAM,EAClB,aAAa,EAAE,eAAe,CAAC,GAAG,CAAC,EACnC,aAAa,EAAE,mBAAmB,4BAMnC;AAED,wBAAgB,0BAA0B,CAAC,CAAC,EAC1C,QAAQ,EAAE,GAAG,GACZ,eAAe,CAAC,CAAC,CAAC,CAMpB;AAED,wBAAgB,kBAAkB,CAAC,CAAC,EAAE,SAAS,EAAE,UAAU,CAAC,CAAC,CAAC,4BAE7D;AAED,wBAAgB,aAAa,CAAC,CAAC,KAAK,eAAe,CAAC,CAAC,CAAC,GAAG,SAAS,CAajE;AAED,wBAAgB,UAAU,CAAC,CAAC,EAAE,QAAQ,EAAE,KAAK,CAAC,CAAC,CAAC,QAmB/C"}
+{"version":3,"file":"state.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/state.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAKpD,OAAO,EAAE,mBAAmB,EAAE,MAAM,SAAS,CAAC;AAE9C,eAAO,MAAM,0BAA0B,eAAe,CAAC;AAEvD,eAAO,MAAM,eAAe,QAI3B,CAAC;AAEF,MAAM,MAAM,eAAe,CAAC,CAAC,IAAI;IAC/B,OAAO,EAAE,OAAO,0BAA0B,CAAC;IAC3C,KAAK,EAAE,mBAAmB,CAAC;IAC3B,QAAQ,EAAE,CAAC,EAAE,CAAC;CACf,CAAC;AAEF,wBAAgB,eAAe,CAC7B,UAAU,EAAE,MAAM,EAClB,aAAa,EAAE,eAAe,CAAC,GAAG,CAAC,EACnC,aAAa,EAAE,mBAAmB,4BAMnC;AAED,wBAAgB,0BAA0B,CAAC,CAAC,EAC1C,QAAQ,EAAE,GAAG,EACb,aAAa,EAAE,mBAAmB,GACjC,eAAe,CAAC,CAAC,CAAC,CAOpB;AAED,wBAAgB,kBAAkB,CAAC,CAAC,EAClC,SAAS,EAAE,UAAU,CAAC,CAAC,CAAC,EACxB,aAAa,EAAE,mBAAmB,4BAGnC;AAED,wBAAgB,aAAa,CAAC,CAAC,KAAK,eAAe,CAAC,CAAC,CAAC,GAAG,SAAS,CAajE;AAED,wBAAgB,UAAU,CAAC,CAAC,EAC1B,QAAQ,EAAE,KAAK,CAAC,CAAC,CAAC,EAClB,aAAa,EAAE,mBAAmB,QAsBnC"}

package/dist/agent/chat/state.js

@@ -13,19 +13,20 @@ function createChatState(userPrompt, existingState, selectedModel) {
     const messages = existingState.messages || [];
     const chatModel = (0, model_1.createChatModel)(messages, selectedModel);
     chatModel.pushUserMessage(userPrompt);
-    return createChatStateForMessages(chatModel.messages);
+    return createChatStateForMessages(chatModel.messages, selectedModel);
 }
 exports.createChatState = createChatState;
-function createChatStateForMessages(messages) {
+function createChatStateForMessages(messages, selectedModel) {
     // TODO: Add better types for messages
     return {
         version: exports.CURRENT_CHAT_STATE_VERSION,
+        model: selectedModel,
         messages: messages,
     };
 }
 exports.createChatStateForMessages = createChatStateForMessages;
-function chatStateFromModel(chatModel) {
-    return createChatStateForMessages(chatModel.messages);
+function chatStateFromModel(chatModel, selectedModel) {
+    return createChatStateForMessages(chatModel.messages, selectedModel);
 }
 exports.chatStateFromModel = chatStateFromModel;
 function loadChatState() {
@@ -40,10 +41,11 @@ function loadChatState() {
     return state;
 }
 exports.loadChatState = loadChatState;
-function saveToDisk(messages) {
+function saveToDisk(messages, selectedModel) {
     const statePath = exports.CHAT_STATE_PATH;
     let existingState = {
         version: exports.CURRENT_CHAT_STATE_VERSION,
+        model: selectedModel,
         messages: [],
     };
     // Ensure directory exists before trying to read/write
@@ -57,6 +59,7 @@ function saveToDisk(messages) {
     const newState = {
         ...existingState,
         messages: messages,
+        model: selectedModel,
     };
     fs_1.default.writeFileSync(statePath, JSON.stringify(newState, null, 2));
 }

package/dist/file/server.d.ts

@@ -5,10 +5,12 @@ export declare class FileServiceServer {
     private server;
     private actionsSummary;
     private updateFile;
-    constructor({ port, repoDir, updateFile, }: {
+    private onComplete?;
+    constructor({ port, repoDir, updateFile, onComplete, }: {
         port: number;
         repoDir: string;
         updateFile: boolean;
+        onComplete?: () => void;
     });
     getActionsSummary(): string | undefined;
     setFilePath(filePath: string): void;

package/dist/file/server.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"server.d.ts","sourceRoot":"","sources":["../../src/file/server.ts"],"names":[],"mappings":"AAWA,qBAAa,iBAAiB;IAC5B,OAAO,CAAC,IAAI,CAAa;IACzB,OAAO,CAAC,QAAQ,CAAc;IAC9B,OAAO,CAAC,OAAO,CAAc;IAC7B,OAAO,CAAC,MAAM,CAA4C;IAC1D,OAAO,CAAC,cAAc,CAAqB;IAC3C,OAAO,CAAC,UAAU,CAAkB;gBAExB,EACV,IAAI,EACJ,OAAO,EACP,UAAU,GACX,EAAE;QACD,IAAI,EAAE,MAAM,CAAC;QACb,OAAO,EAAE,MAAM,CAAC;QAChB,UAAU,EAAE,OAAO,CAAC;KACrB;IAMD,iBAAiB;IAIjB,WAAW,CAAC,QAAQ,EAAE,MAAM;IAItB,gBAAgB,IAAI,OAAO,CAAC,MAAM,CAAC;IA+CnC,IAAI,IAAI,OAAO,CAAC,IAAI,CAAC;CAgB5B;AAED,wBAAsB,gBAAgB,kBAAK"}
+{"version":3,"file":"server.d.ts","sourceRoot":"","sources":["../../src/file/server.ts"],"names":[],"mappings":"AAWA,qBAAa,iBAAiB;IAC5B,OAAO,CAAC,IAAI,CAAa;IACzB,OAAO,CAAC,QAAQ,CAAc;IAC9B,OAAO,CAAC,OAAO,CAAc;IAC7B,OAAO,CAAC,MAAM,CAA4C;IAC1D,OAAO,CAAC,cAAc,CAAqB;IAC3C,OAAO,CAAC,UAAU,CAAkB;IACpC,OAAO,CAAC,UAAU,CAAC,CAAa;gBAEpB,EACV,IAAI,EACJ,OAAO,EACP,UAAU,EACV,UAAU,GACX,EAAE;QACD,IAAI,EAAE,MAAM,CAAC;QACb,OAAO,EAAE,MAAM,CAAC;QAChB,UAAU,EAAE,OAAO,CAAC;QACpB,UAAU,CAAC,EAAE,MAAM,IAAI,CAAC;KACzB;IAOD,iBAAiB;IAIjB,WAAW,CAAC,QAAQ,EAAE,MAAM;IAItB,gBAAgB,IAAI,OAAO,CAAC,MAAM,CAAC;IAmDnC,IAAI,IAAI,OAAO,CAAC,IAAI,CAAC;CAgB5B;AAED,wBAAsB,gBAAgB,kBAAK"}

package/dist/file/server.js

@@ -16,10 +16,12 @@ class FileServiceServer {
     server;
     actionsSummary;
     updateFile = false;
-    constructor({ port, repoDir, updateFile, }) {
+    onComplete;
+    constructor({ port, repoDir, updateFile, onComplete, }) {
         this.port = port;
         this.repoDir = repoDir;
         this.updateFile = updateFile;
+        this.onComplete = onComplete;
     }
     getActionsSummary() {
         return this.actionsSummary;
@@ -35,6 +37,9 @@ class FileServiceServer {
             const { generatedCode, importPaths, actionsSummary } = req.body;
             this.actionsSummary = actionsSummary;
             if (!this.updateFile) {
+                // Not updating the file in this scenario
+                if (this.onComplete)
+                    this.onComplete();
                 return res.send({ success: true });
             }
             try {
@@ -45,6 +50,8 @@ class FileServiceServer {
                     const importStatements = await (0, web_1.importAllExportsStmtFromFilePaths)(this.repoDir, importPaths, this.filePath);
                     fs_1.default.writeFileSync(testFilePath, `${importStatements.join("\n")}\n${newContents}`, "utf-8");
                     await (0, web_1.lintErrors)(testFilePath);
+                    if (this.onComplete)
+                        this.onComplete();
                     return res.send({ success: true });
                 }
             }

package/dist/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAQlC,OAAO,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AAqBpC,wBAAsB,UAAU,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,EAAE,KAAK,CAAC,EAAE,SAAS,iBAyD3E"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAQlC,OAAO,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AAoBpC,wBAAsB,UAAU,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,EAAE,KAAK,CAAC,EAAE,SAAS,iBAsD3E"}

package/dist/index.js

@@ -11,7 +11,6 @@ const scenarios_1 = require("./bin/utils/scenarios");
 const client_1 = __importDefault(require("./file/client"));
 const reporter_1 = require("./reporter");
 const session_1 = require("./session");
-const pw_test_1 = require("./utils/pw-test");
 const flushEvents = async () => {
     await (0, llm_1.flushAllTraces)();
 };
@@ -68,8 +67,6 @@ async function createTest(task, page, scope) {
             importPaths,
             actionsSummary,
         });
-        // skip the rest of the test once generation is over
-        await (0, pw_test_1.skipTest)();
     }
     finally {
         // Ensure listeners are removed even if an error occurs

package/dist/tools/commit-and-create-pr.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"commit-and-create-pr.d.ts","sourceRoot":"","sources":["../../src/tools/commit-and-create-pr.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAQnD,UAAU,cAAc;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;CACd;AAED,wBAAgB,cAAc,CAAC,GAAG,EAAE,MAAM,GAAG,cAAc,CAiB1D;AAiBD,eAAO,MAAM,wBAAwB,EAAE,IAkFtC,CAAC"}
+{"version":3,"file":"commit-and-create-pr.d.ts","sourceRoot":"","sources":["../../src/tools/commit-and-create-pr.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,wBAAwB,CAAC;AAQnD,UAAU,cAAc;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;CACd;AAED,wBAAgB,cAAc,CAAC,GAAG,EAAE,MAAM,GAAG,cAAc,CAiB1D;AA2CD,eAAO,MAAM,wBAAwB,EAAE,IA8FtC,CAAC"}

package/dist/tools/commit-and-create-pr.js

@@ -28,7 +28,24 @@ const CommitAndPushChangesSchema = zod_1.z.object({
     commitMessage: zod_1.z
         .string()
         .describe("A short message to use for the commit. Should not be more than 8 words. Should follow conventional commit format."),
+    description: zod_1.z.string().describe(`A longer description of the changes you made. This will be used as the description of a pull request on GitHub, and so you should follow markdown formatting.
+Your code will be reviewed by a human, and you should include everything that will provide context and improve the reviewer's confidence in the changes.
+
+For example, if you used the test run tool, you should include the results (and the report URL if available). Report URL is especially important, because it contains
+videos and other artifacts that help the reviewer gain more context and confidence in the changes. If tests pass, reviewer will see the video and merge the PR.
+If tests fail, reviewer will see the video and the test artifacts, and will be able to help you debug the issue.`),
 });
+function formatDescriptionWithTimestamp(description, existingBody, type = "create") {
+    const timestamp = new Date()
+        .toISOString()
+        .replace("T", " ")
+        .replace("Z", " UTC");
+    const timestampText = `<sup>${type === "create" ? "Created" : "Updated"} at ${timestamp}</sup>`;
+    if (existingBody) {
+        return `${existingBody}\n\n---\n${description}\n\n${timestampText}`;
+    }
+    return `${description}\n\n${timestampText}`;
+}
 exports.commitAndPushChangesTool = {
     schema: {
         name: "commitAndPushChanges",
@@ -41,7 +58,7 @@ Returns the URL of the created or updated pull request.`,
     },
     execute: async (input) => {
         try {
-            const { commitMessage } = input;
+            const { commitMessage, description } = input;
             const currentBranch = (0, child_process_1.execSync)("git rev-parse --abbrev-ref HEAD")
                 .toString()
                 .trim();
@@ -82,11 +99,21 @@ Returns the URL of the created or updated pull request.`,
             (0, child_process_1.execSync)(`git push origin ${branchName} --set-upstream`);
             const existingPR = existingPRs?.find((pr) => pr.head.ref === branchName);
             if (existingPR) {
+                // Append the new description to the existing PR description
+                const updatedDescription = formatDescriptionWithTimestamp(description, existingPR.body, "update");
+                await (0, utils_1.callGitHubProxy)({
+                    method: "PATCH",
+                    url: `https://api.github.com/repos/${owner}/${repo}/pulls/${existingPR.number}`,
+                    body: {
+                        body: updatedDescription,
+                    },
+                });
                 return {
                     isError: false,
                     result: `Committed and pushed changes to existing PR: ${existingPR.html_url}`,
                 };
             }
+            const initialDescription = formatDescriptionWithTimestamp(description);
             const pr = (await (0, utils_1.callGitHubProxy)({
                 method: "POST",
                 url: `https://api.github.com/repos/${owner}/${repo}/pulls`,
@@ -94,7 +121,7 @@ Returns the URL of the created or updated pull request.`,
                     title: commitMessage,
                     head: branchName,
                     base: "main",
-                    body: "Created via CommitAndPushChanges tool",
+                    body: initialDescription,
                 },
             }));
             return {

package/dist/tools/grep.js

@@ -41,15 +41,15 @@ exports.grepTool = {
                 cmd = `find ${dir} ${excludeFind} -name "${input.filePattern}" -exec grep -rin "${input.pattern}" {} \\;`;
             }
             const { stdout, stderr } = await execAsync(cmd);
-            if (stderr) {
+            if (stdout) {
                 return {
-                    isError: true,
-                    result: stderr,
+                    isError: false,
+                    result: stdout,
                 };
             }
             return {
-                isError: false,
-                result: stdout,
+                isError: true,
+                result: stderr,
             };
         }
         catch (error) {

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

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

package/dist/tools/test-run.js

@@ -12,11 +12,25 @@ const RunTestSchema = zod_1.z.object({
         .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(),
 });
+function hasCloudflareCredentials() {
+    return (process.env.R2_ACCOUNT_ID &&
+        process.env.R2_ACCESS_KEY_ID &&
+        process.env.R2_SECRET_ACCESS_KEY);
+}
+function buildReportUrl(projectName, testRunId) {
+    return `https://reports.empirical.run/${projectName}/${testRunId}/index.html`;
+}
+function buildResult({ hasTestPassed, summaryJson, reportUrl, }) {
+    return `
+Test run is complete. Result: ${hasTestPassed ? "Passed" : "Failed"}
+
+${reportUrl ? `Report URL: ${reportUrl}` : ""}
+
+# Raw result (in Playwright JSON format)
+${JSON.stringify(summaryJson)}
+`;
+}
 exports.runTestTool = {
     schema: {
         name: "runTest",
@@ -24,7 +38,18 @@ exports.runTestTool = {
         parameters: RunTestSchema,
     },
     execute: async (input) => {
-        const { testName, suites, fileName, project, headed } = input;
+        let reportUrl = undefined;
+        let projectName = undefined;
+        let testRunId = undefined;
+        if (hasCloudflareCredentials()) {
+            projectName = "test-gen-chat-agent";
+            testRunId = Date.now().toString();
+            reportUrl = buildReportUrl(projectName, testRunId);
+        }
+        else {
+            console.warn("R2 credentials not found: report artifacts will not be uploaded");
+        }
+        const { testName, suites, fileName, project } = input;
         try {
             // {"project":"chromium","suites":[],"fileName":"tests/quizizz-for-work/group.spec.ts","testName":"Create a group"}
             // This runs all tests - TODO: Debug this, should only run the testName
@@ -33,10 +58,20 @@ exports.runTestTool = {
                 suites,
                 fileName,
                 projects: [project],
-                headed,
+                // Adding these to enforce report artifacts are uploaded
+                envOverrides: projectName && testRunId
+                    ? {
+                        PROJECT_NAME: projectName,
+                        TEST_RUN_GITHUB_ACTION_ID: testRunId,
+                    }
+                    : undefined,
             });
             return {
-                result: JSON.stringify(result),
+                result: buildResult({
+                    hasTestPassed: result.hasTestPassed,
+                    summaryJson: result.summaryJson,
+                    reportUrl: reportUrl,
+                }),
                 isError: false,
             };
         }

package/dist/utils/exec.d.ts

@@ -1,3 +1,11 @@
+export declare class ProcessManager {
+    private childProcess;
+    execute(command: string[], options: {
+        env?: Record<string, string>;
+    }): Promise<number>;
+    terminate(): void;
+    isRunning(): boolean;
+}
 export declare function cmd(command: string[], options: {
     env?: Record<string, string>;
 }): Promise<number>;

package/dist/utils/exec.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"exec.d.ts","sourceRoot":"","sources":["../../src/utils/exec.ts"],"names":[],"mappings":"AAGA,wBAAsB,GAAG,CACvB,OAAO,EAAE,MAAM,EAAE,EACjB,OAAO,EAAE;IAAE,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;CAAE,GACxC,OAAO,CAAC,MAAM,CAAC,CAuCjB"}
+{"version":3,"file":"exec.d.ts","sourceRoot":"","sources":["../../src/utils/exec.ts"],"names":[],"mappings":"AAQA,qBAAa,cAAc;IACzB,OAAO,CAAC,YAAY,CAA6B;IAE3C,OAAO,CACX,OAAO,EAAE,MAAM,EAAE,EACjB,OAAO,EAAE;QAAE,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;KAAE,GACxC,OAAO,CAAC,MAAM,CAAC;IAkDlB,SAAS,IAAI,IAAI;IASjB,SAAS,IAAI,OAAO;CAGrB;AAED,wBAAsB,GAAG,CACvB,OAAO,EAAE,MAAM,EAAE,EACjB,OAAO,EAAE;IAAE,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;CAAE,GACxC,OAAO,CAAC,MAAM,CAAC,CAGjB"}

package/dist/utils/exec.js

@@ -3,43 +3,73 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
     return (mod && mod.__esModule) ? mod : { "default": mod };
 };
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.cmd = void 0;
+exports.cmd = exports.ProcessManager = void 0;
 const child_process_1 = require("child_process");
 const process_1 = __importDefault(require("process"));
-async function cmd(command, options) {
-    let errorLogs = [];
-    return new Promise((resolveFunc, rejectFunc) => {
-        if (!command[0]) {
-            rejectFunc(new Error("Command cannot be empty"));
-            return;
+const acceptableExitCodes = [
+    0, // Implies successful execution (no errors)
+    130, // Implies user interrupted the process (for SIGINT)
+];
+class ProcessManager {
+    childProcess = null;
+    async execute(command, options) {
+        if (this.childProcess) {
+            throw new Error("Process is already running");
         }
-        let p = (0, child_process_1.spawn)(command[0], command.slice(1), {
-            env: { ...process_1.default.env, ...options.env },
-        });
-        p.stdout.on("data", (x) => {
-            const log = x.toString();
-            if (log.includes("Error")) {
-                errorLogs.push(log);
-            }
-            process_1.default.stdout.write(log);
-        });
-        p.stderr.on("data", (x) => {
-            const log = x.toString();
-            process_1.default.stderr.write(log);
-            errorLogs.push(log);
-        });
-        p.on("error", async (err) => {
-            rejectFunc(err);
-        });
-        p.on("exit", async (code) => {
-            if (code !== 0) {
-                const errorMessage = errorLogs.slice(-3).join("\n");
-                rejectFunc(new Error(errorMessage));
-            }
-            else {
-                resolveFunc(code ?? 1);
+        let errorLogs = [];
+        return new Promise((resolveFunc, rejectFunc) => {
+            if (!command[0]) {
+                rejectFunc(new Error("Command cannot be empty"));
+                return;
             }
+            const p = (0, child_process_1.spawn)(command[0], command.slice(1), {
+                env: { ...process_1.default.env, ...options.env },
+                detached: true, // Create process group so we can terminate all child processes
+            });
+            this.childProcess = p;
+            p.stdout.on("data", (x) => {
+                const log = x.toString();
+                if (log.includes("Error")) {
+                    errorLogs.push(log);
+                }
+                process_1.default.stdout.write(log);
+            });
+            p.stderr.on("data", (x) => {
+                const log = x.toString();
+                process_1.default.stderr.write(log);
+                errorLogs.push(log);
+            });
+            p.on("error", async (err) => {
+                this.childProcess = null;
+                rejectFunc(err);
+            });
+            p.on("exit", async (code) => {
+                this.childProcess = null;
+                if (!acceptableExitCodes.includes(code ?? 1)) {
+                    const errorMessage = errorLogs.slice(-3).join("\n");
+                    rejectFunc(new Error(errorMessage));
+                }
+                else {
+                    resolveFunc(code ?? 1);
+                }
+            });
         });
-    });
+    }
+    terminate() {
+        if (!this.childProcess) {
+            throw new Error("No process is currently running");
+        }
+        if (this.childProcess.pid) {
+            process_1.default.kill(-this.childProcess.pid, "SIGINT");
+        }
+    }
+    isRunning() {
+        return this.childProcess !== null;
+    }
+}
+exports.ProcessManager = ProcessManager;
+async function cmd(command, options) {
+    const manager = new ProcessManager();
+    return manager.execute(command, options);
 }
 exports.cmd = cmd;

package/dist/utils/pw-test.d.ts

@@ -1,3 +1,2 @@
 export declare function getTestFixtureModuleFromRepo(repoDir: string): Promise<any>;
-export declare function skipTest(): Promise<void>;
 //# sourceMappingURL=pw-test.d.ts.map

package/dist/utils/pw-test.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"pw-test.d.ts","sourceRoot":"","sources":["../../src/utils/pw-test.ts"],"names":[],"mappings":"AAEA,wBAAsB,4BAA4B,CAChD,OAAO,EAAE,MAAM,GACd,OAAO,CAAC,GAAG,CAAC,CAKd;AAED,wBAAsB,QAAQ,kBAS7B"}
+{"version":3,"file":"pw-test.d.ts","sourceRoot":"","sources":["../../src/utils/pw-test.ts"],"names":[],"mappings":"AAEA,wBAAsB,4BAA4B,CAChD,OAAO,EAAE,MAAM,GACd,OAAO,CAAC,GAAG,CAAC,CAKd"}

package/dist/utils/pw-test.js

@@ -3,7 +3,7 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
     return (mod && mod.__esModule) ? mod : { "default": mod };
 };
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.skipTest = exports.getTestFixtureModuleFromRepo = void 0;
+exports.getTestFixtureModuleFromRepo = void 0;
 const api_1 = __importDefault(require("tsx/cjs/api"));
 async function getTestFixtureModuleFromRepo(repoDir) {
     const [lastDir] = repoDir.split("/").reverse();
@@ -12,15 +12,3 @@ async function getTestFixtureModuleFromRepo(repoDir) {
     return test;
 }
 exports.getTestFixtureModuleFromRepo = getTestFixtureModuleFromRepo;
-async function skipTest() {
-    let test;
-    let repoDir = process.cwd();
-    try {
-        test = await getTestFixtureModuleFromRepo(repoDir);
-    }
-    catch (e) {
-        console.error("Error while importing fixture module to extract test:", e);
-    }
-    test.skip();
-}
-exports.skipTest = skipTest;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@empiricalrun/test-gen",
-  "version": "0.53.8",
+  "version": "0.53.10",
   "publishConfig": {
     "registry": "https://registry.npmjs.org/",
     "access": "public"
@@ -56,9 +56,9 @@
     "tsx": "^4.16.2",
     "typescript": "^5.3.3",
     "zod": "^3.23.8",
+    "@empiricalrun/llm": "^0.14.5",
     "@empiricalrun/r2-uploader": "^0.3.8",
-    "@empiricalrun/test-run": "^0.7.6",
-    "@empiricalrun/llm": "^0.14.5"
+    "@empiricalrun/test-run": "^0.7.7"
   },
   "devDependencies": {
     "@playwright/test": "1.47.1",