@empiricalrun/test-gen 0.69.5 → 0.69.7

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md CHANGED
@@ -1,5 +1,24 @@
1
1
  # @empiricalrun/test-gen
2
2
 
3
+ ## 0.69.7
4
+
5
+ ### Patch Changes
6
+
7
+ - bca8c8b: fix: codegen timing issue
8
+ - 49a3382: feat: add commands for environments and download build
9
+ - Updated dependencies [16f3927]
10
+ - @empiricalrun/llm@0.19.3
11
+
12
+ ## 0.69.6
13
+
14
+ ### Patch Changes
15
+
16
+ - c0c1a9a: fix: use gemini stable instead of preview
17
+ - d67ca35: feat: add fetch-chat-session command, handle artifacts with same name
18
+ - 9a923e2: chore: preview env resources are in ap-south-1
19
+ - Updated dependencies [c0c1a9a]
20
+ - @empiricalrun/llm@0.19.2
21
+
3
22
  ## 0.69.5
4
23
 
5
24
  ### Patch Changes
@@ -1 +1 @@
1
- {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/index.ts"],"names":[],"mappings":"AAEA,OAAO,EAGL,mBAAmB,EACpB,MAAM,4BAA4B,CAAC;AAyCpC,wBAAsB,yBAAyB,IAAI,OAAO,CACxD,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CACvB,CAwBA;AAED,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,iBA+HA;AAuBD,wBAAsB,wBAAwB,CAAC,EAC7C,aAAa,EACb,aAAa,GACd,EAAE;IACD,aAAa,EAAE,mBAAmB,CAAC;IACnC,aAAa,EAAE,MAAM,CAAC;CACvB,iBA6DA"}
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/agent/chat/index.ts"],"names":[],"mappings":"AAEA,OAAO,EAGL,mBAAmB,EACpB,MAAM,4BAA4B,CAAC;AAyCpC,wBAAsB,yBAAyB,IAAI,OAAO,CACxD,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CACvB,CAwBA;AAED,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,iBAkIA;AAuBD,wBAAsB,wBAAwB,CAAC,EAC7C,aAAa,EACb,aAAa,GACd,EAAE;IACD,aAAa,EAAE,mBAAmB,CAAC;IACnC,aAAa,EAAE,MAAM,CAAC;CACvB,iBA6DA"}
@@ -118,7 +118,9 @@ async function runChatAgentForCLI({ useDiskForChatState, selectedModel, initialP
118
118
  }
119
119
  if (!stopCriteria(userPrompt)) {
120
120
  const { text, attachments } = (0, utils_1.extractAttachments)(userPrompt);
121
- const processed = await chatModel.processAttachments(attachments);
121
+ const processed = attachments.length > 0
122
+ ? await chatModel.processAttachments(attachments)
123
+ : [];
122
124
  chatModel.pushUserMessage(text, processed);
123
125
  }
124
126
  }
@@ -1 +1 @@
1
- {"version":3,"file":"for-recorder.d.ts","sourceRoot":"","sources":["../../../../../src/agent/cua/pw-codegen/pw-pause/for-recorder.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAIvC,OAAO,EAAE,cAAc,EAAE,MAAM,SAAS,CAAC;AAEzC,qBAAa,iCAAiC;IAO1C,OAAO,CAAC,eAAe;IANzB,OAAO,CAAC,IAAI,CAAa;IACzB,OAAO,CAAC,IAAI,CAAmB;IAC/B,OAAO,CAAC,MAAM,CAA4C;IAC1D,OAAO,CAAC,iBAAiB,CAAqB;gBAGpC,eAAe,EAAE,CAAC,IAAI,EAAE,cAAc,EAAE,KAAK,OAAO,CAAC,IAAI,CAAC;YAKtD,QAAQ;IAWhB,UAAU,CAAC,IAAI,EAAE,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC;IAkBrC,sBAAsB,CAAC,IAAI,EAAE,IAAI;CA4CxC"}
1
+ {"version":3,"file":"for-recorder.d.ts","sourceRoot":"","sources":["../../../../../src/agent/cua/pw-codegen/pw-pause/for-recorder.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAIvC,OAAO,EAAE,cAAc,EAAE,MAAM,SAAS,CAAC;AAGzC,qBAAa,iCAAiC;IAO1C,OAAO,CAAC,eAAe;IANzB,OAAO,CAAC,IAAI,CAAa;IACzB,OAAO,CAAC,IAAI,CAAmB;IAC/B,OAAO,CAAC,MAAM,CAA4C;IAC1D,OAAO,CAAC,iBAAiB,CAAqB;gBAGpC,eAAe,EAAE,CAAC,IAAI,EAAE,cAAc,EAAE,KAAK,OAAO,CAAC,IAAI,CAAC;YAKtD,QAAQ;IAWhB,UAAU,CAAC,IAAI,EAAE,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC;IAkBrC,sBAAsB,CAAC,IAAI,EAAE,IAAI;CAyCxC"}
@@ -7,6 +7,7 @@ exports.PlaywrightPauseCodegenForRecorder = void 0;
7
7
  const express_1 = __importDefault(require("express"));
8
8
  const logger_1 = require("../../../../logger");
9
9
  const ipc_1 = require("./ipc");
10
+ const utils_1 = require("./utils");
10
11
  class PlaywrightPauseCodegenForRecorder {
11
12
  sourcesCallback;
12
13
  port = 0;
@@ -45,29 +46,27 @@ class PlaywrightPauseCodegenForRecorder {
45
46
  await this.startPlaywrightCodegen(page);
46
47
  }
47
48
  async startPlaywrightCodegen(page) {
48
- // Similar to the same name method in the main codegen
49
- /**
50
- * x-pw-glass
51
- * # inside shadow DOM (closed)
52
- * |----> x-pw-overlay (controls)
53
- * |----> x-pw-highlight (red highlight)
54
- * |----> x-pw-tooltip
55
- */
56
- const timerPromise = new Promise((resolve) => {
57
- setTimeout(resolve, 1000);
58
- });
49
+ // TODO: Merge this with the other page.pause oriented codegen
59
50
  const pausePromise = page.pause();
60
- await timerPromise;
51
+ await (0, utils_1.sleep)(5_000);
61
52
  const evaluatePromise = page.evaluate(() => {
62
53
  // @ts-ignore
63
54
  console.log(window["__pw_recorderSetMode"]("recording"));
64
55
  // Remove playwright's glass pane
56
+ /**
57
+ * x-pw-glass
58
+ * # inside shadow DOM (closed)
59
+ * |----> x-pw-overlay (controls)
60
+ * |----> x-pw-highlight (red highlight)
61
+ * |----> x-pw-tooltip
62
+ */
65
63
  const glassPane = document.querySelector("x-pw-glass");
66
64
  if (glassPane) {
67
65
  const styles = glassPane.getAttribute("style") || "";
68
66
  glassPane.setAttribute("style", styles + "; display: none;");
69
67
  }
70
68
  });
69
+ await Promise.all([pausePromise, evaluatePromise]);
71
70
  await page.addInitScript(() => {
72
71
  let displayNoneSet = false;
73
72
  function setDisplayNone() {
@@ -86,7 +85,6 @@ class PlaywrightPauseCodegenForRecorder {
86
85
  setDisplayNone();
87
86
  }, 100);
88
87
  });
89
- await Promise.all([pausePromise, evaluatePromise]);
90
88
  }
91
89
  }
92
90
  exports.PlaywrightPauseCodegenForRecorder = PlaywrightPauseCodegenForRecorder;
@@ -1 +1 @@
1
- {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../../src/agent/cua/pw-codegen/pw-pause/index.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAGvC,OAAO,EAAE,qBAAqB,EAAE,MAAM,UAAU,CAAC;AAKjD,OAAO,EAAE,2BAA2B,EAAE,sBAAsB,EAAE,MAAM,SAAS,CAAC;AAE9E,wBAAsB,kBAAkB,CAAC,OAAO,EAAE,MAAM,oBAyCvD;AAED,qBAAa,sBAAuB,YAAW,qBAAqB;IAClE,OAAO,CAAC,IAAI,CAAa;IACzB,OAAO,CAAC,IAAI,CAAmB;IAC/B,OAAO,CAAC,MAAM,CAA4C;IAC1D,OAAO,CAAC,iBAAiB,CAAqB;;YAMhC,QAAQ;IAUhB,UAAU,CAAC,IAAI,EAAE,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC;IAkBrC,sBAAsB,CAAC,IAAI,EAAE,IAAI;IAyBjC,YAAY,IAAI,OAAO,CAAC,IAAI,CAAC;IAI7B,oBAAoB,IAAI,OAAO,CAAC,MAAM,CAAC;CAU9C"}
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../../src/agent/cua/pw-codegen/pw-pause/index.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAGvC,OAAO,EAAE,qBAAqB,EAAE,MAAM,UAAU,CAAC;AAKjD,OAAO,EAAE,2BAA2B,EAAE,sBAAsB,EAAE,MAAM,SAAS,CAAC;AAE9E,wBAAsB,kBAAkB,CAAC,OAAO,EAAE,MAAM,oBAyCvD;AAED,qBAAa,sBAAuB,YAAW,qBAAqB;IAClE,OAAO,CAAC,IAAI,CAAa;IACzB,OAAO,CAAC,IAAI,CAAmB;IAC/B,OAAO,CAAC,MAAM,CAA4C;IAC1D,OAAO,CAAC,iBAAiB,CAAqB;;YAMhC,QAAQ;IAUhB,UAAU,CAAC,IAAI,EAAE,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC;IAkBrC,sBAAsB,CAAC,IAAI,EAAE,IAAI;IAqBjC,YAAY,IAAI,OAAO,CAAC,IAAI,CAAC;IAI7B,oBAAoB,IAAI,OAAO,CAAC,MAAM,CAAC;CAU9C"}
@@ -9,6 +9,7 @@ const express_1 = __importDefault(require("express"));
9
9
  const logger_1 = require("../../../../logger");
10
10
  const ipc_1 = require("./ipc");
11
11
  const patch_1 = require("./patch");
12
+ const utils_1 = require("./utils");
12
13
  var patch_2 = require("./patch");
13
14
  Object.defineProperty(exports, "preparePlaywrightForCodegen", { enumerable: true, get: function () { return patch_2.preparePlaywrightForCodegen; } });
14
15
  Object.defineProperty(exports, "revertToOriginalPwCode", { enumerable: true, get: function () { return patch_2.revertToOriginalPwCode; } });
@@ -77,14 +78,10 @@ class PlaywrightPauseCodegen {
77
78
  await this.startPlaywrightCodegen(page);
78
79
  }
79
80
  async startPlaywrightCodegen(page) {
80
- // TODO: Glass pane needs to be removed on every page load
81
81
  // We use bindings that Playwright exposes to the page
82
82
  // Ref: https://github.com/microsoft/playwright/blob/e1c8e0f6b33923c95cc4b9416aefa6977b1d3c55/packages/playwright-core/src/server/recorder.ts#L191
83
- const timerPromise = new Promise((resolve) => {
84
- setTimeout(resolve, 1000);
85
- });
86
83
  const pausePromise = page.pause();
87
- await timerPromise;
84
+ await (0, utils_1.sleep)(5_000);
88
85
  const evaluatePromise = page.evaluate(() => {
89
86
  // First, we start recording
90
87
  // @ts-ignore
@@ -0,0 +1,2 @@
1
+ export declare function sleep(ms: number): Promise<void>;
2
+ //# sourceMappingURL=utils.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../../../../src/agent/cua/pw-codegen/pw-pause/utils.ts"],"names":[],"mappings":"AAAA,wBAAsB,KAAK,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAErD"}
@@ -0,0 +1,6 @@
1
+ "use strict";
2
+ Object.defineProperty(exports, "__esModule", { value: true });
3
+ exports.sleep = sleep;
4
+ async function sleep(ms) {
5
+ return new Promise((resolve) => setTimeout(resolve, ms));
6
+ }
@@ -1 +1 @@
1
- {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/artifacts/index.ts"],"names":[],"mappings":"AAIA,OAAO,EACL,QAAQ,EACR,aAAa,EAGd,MAAM,4BAA4B,CAAC;AAkBpC,wBAAgB,2BAA2B,uBAM1C;AA6HD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAsCG;AACH,wBAAsB,gBAAgB,CACpC,MAAM,EAAE,aAAa,EAAE,EACvB,OAAO,EAAE,MAAM,EACf,UAAU,EAAE,MAAM,GACjB,OAAO,CAAC,QAAQ,EAAE,CAAC,CA6DrB;AAED,qBAAa,oBAAoB;IAC/B,OAAO,CAAC,UAAU,CAAS;IAC3B,OAAO,CAAC,YAAY,CAAS;IAC7B,OAAO,CAAC,eAAe,CAAkB;IACzC,OAAO,CAAC,aAAa,CAA8B;gBAEvC,YAAY,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM;IAKvC,OAAO,CAAC,SAAS,EAAE,aAAa,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC;IAelD,iBAAiB,IAAI,OAAO,CAAC,QAAQ,EAAE,CAAC;CAMtD"}
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/artifacts/index.ts"],"names":[],"mappings":"AAIA,OAAO,EACL,QAAQ,EACR,aAAa,EAGd,MAAM,4BAA4B,CAAC;AAkBpC,wBAAgB,2BAA2B,uBAM1C;AA6HD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAsCG;AACH,wBAAsB,gBAAgB,CACpC,MAAM,EAAE,aAAa,EAAE,EACvB,OAAO,EAAE,MAAM,EACf,UAAU,EAAE,MAAM,GACjB,OAAO,CAAC,QAAQ,EAAE,CAAC,CA8DrB;AAED,qBAAa,oBAAoB;IAC/B,OAAO,CAAC,UAAU,CAAS;IAC3B,OAAO,CAAC,YAAY,CAAS;IAC7B,OAAO,CAAC,eAAe,CAAkB;IACzC,OAAO,CAAC,aAAa,CAA8B;gBAEvC,YAAY,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM;IAKvC,OAAO,CAAC,SAAS,EAAE,aAAa,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC;IAelD,iBAAiB,IAAI,OAAO,CAAC,QAAQ,EAAE,CAAC;CAMtD"}
@@ -167,8 +167,9 @@ async function collectArtifacts(inputs, repoDir, toolCallId) {
167
167
  try {
168
168
  const dirs = { destinationDir, sourceDir: repoDir };
169
169
  const dataArtifactsWithFileName = dataArtifacts.map((a) => {
170
+ const randomSuffix = Math.random().toString(36).substring(2, 15);
170
171
  const fileExt = mime_1.default.getExtension(a.contentType);
171
- const fileName = `${(0, slug_1.slugify)(a.name)}-${Date.now().toString()}.${fileExt}`;
172
+ const fileName = `${(0, slug_1.slugify)(a.name)}-${randomSuffix}.${fileExt}`;
172
173
  return { ...a, fileName };
173
174
  });
174
175
  await Promise.all([
package/dist/bin/index.js CHANGED
@@ -7,6 +7,7 @@ Object.defineProperty(exports, "__esModule", { value: true });
7
7
  const llm_1 = require("@empiricalrun/llm");
8
8
  const commander_1 = require("commander");
9
9
  const dotenv_1 = __importDefault(require("dotenv"));
10
+ const fs_1 = __importDefault(require("fs"));
10
11
  const run_1 = require("../agent/browsing/run");
11
12
  const utils_1 = require("../agent/browsing/utils");
12
13
  const chat_1 = require("../agent/chat");
@@ -19,6 +20,7 @@ const run_3 = require("../agent/planner/run");
19
20
  const auth_1 = require("../auth");
20
21
  const api_client_1 = require("../auth/api-client");
21
22
  const recorder_1 = require("../recorder");
23
+ const validation_1 = require("../recorder/validation");
22
24
  const reporter_1 = require("../reporter");
23
25
  const session_1 = require("../session");
24
26
  const test_build_1 = require("../test-build");
@@ -285,6 +287,63 @@ async function main() {
285
287
  }
286
288
  process.exit(0);
287
289
  });
290
+ program
291
+ .command("environments")
292
+ .description("List environments and their latest builds")
293
+ .option("--repo-name <name>", "Repository name to fetch environments for")
294
+ .action(async (opts) => {
295
+ try {
296
+ let repoName;
297
+ if (opts.repoName) {
298
+ repoName = opts.repoName;
299
+ }
300
+ else {
301
+ // Auto-detect repo name from current directory
302
+ repoName = await (0, validation_1.validatePackageJson)(process.cwd());
303
+ }
304
+ const response = await api_client_1.apiClient.request(`/api/environments/list?project_repo_name=${encodeURIComponent(repoName)}`);
305
+ if (!response.ok) {
306
+ console.log(await response.text());
307
+ console.error(`❌ Failed to fetch environments for repo ${repoName}:`, response.statusText);
308
+ process.exit(1);
309
+ }
310
+ const result = await response.json();
311
+ if (result.data.environments && result.data.environments.length > 0) {
312
+ result.data.environments.forEach((env) => {
313
+ console.log(` Environment: ${env.name}`);
314
+ if (env.latest_build) {
315
+ console.log(` Latest build: ${env.latest_build.url}`);
316
+ }
317
+ });
318
+ }
319
+ else {
320
+ console.log(`No environments found for repository: ${repoName}`);
321
+ }
322
+ }
323
+ catch (error) {
324
+ console.error("❌ Error fetching environments:", error.message);
325
+ process.exit(1);
326
+ }
327
+ process.exit(0);
328
+ });
329
+ program
330
+ .command("download-build")
331
+ .description("Download a build from a URL")
332
+ .requiredOption("--url <url>", "Build URL to download")
333
+ .action(async (opts) => {
334
+ try {
335
+ await (0, test_build_1.downloadBuild)({
336
+ buildUrl: opts.url,
337
+ repoPath: process.cwd(),
338
+ });
339
+ console.log("✅ Build downloaded successfully");
340
+ }
341
+ catch (error) {
342
+ console.error("❌ Error downloading build:", error.message);
343
+ process.exit(1);
344
+ }
345
+ process.exit(0);
346
+ });
288
347
  program
289
348
  .command("chat-agent")
290
349
  .description("Run the chat agent")
@@ -310,6 +369,29 @@ async function main() {
310
369
  await (0, setup_1.runSetup)({ repoName: options.repoName });
311
370
  process.exit(0);
312
371
  });
372
+ program
373
+ .command("fetch-chat-state")
374
+ .description("Fetch chat session state from the dashboard")
375
+ .requiredOption("--id <id>", "Chat session ID")
376
+ .action(async (opts) => {
377
+ try {
378
+ const response = await api_client_1.apiClient.request(`/api/chat-sessions/${opts.id}`);
379
+ if (!response.ok) {
380
+ console.error("❌ Failed to fetch chat session:", response.statusText);
381
+ process.exit(1);
382
+ }
383
+ const responseData = await response.json();
384
+ const chatState = responseData.data.chat_session.chat_state;
385
+ const filename = `chat-session-${opts.id}.json`;
386
+ fs_1.default.writeFileSync(filename, JSON.stringify(chatState, null, 2));
387
+ console.log(`✅ Chat state saved to ${filename}`);
388
+ }
389
+ catch (error) {
390
+ console.error("❌ Error fetching chat session:", error.message);
391
+ process.exit(1);
392
+ }
393
+ process.exit(0);
394
+ });
313
395
  program
314
396
  .command("legacy")
315
397
  .description("Run the legacy workflows")
@@ -338,7 +420,6 @@ async function main() {
338
420
  await (0, test_build_1.downloadBuild)({
339
421
  buildUrl: testGenConfig.build.url,
340
422
  repoPath: process.cwd(),
341
- apiKey: process.env.EMPIRICALRUN_API_KEY,
342
423
  });
343
424
  }
344
425
  let agentUsed;
@@ -14,9 +14,9 @@ exports.ARGS_TO_MODEL_MAP = {
14
14
  "claude-4": "claude-sonnet-4-20250514",
15
15
  "claude-sonnet-4": "claude-sonnet-4-20250514",
16
16
  "claude-opus-4": "claude-opus-4-20250514",
17
- "gemini-2.5": "gemini-2.5-pro-preview-06-05",
18
- "gemini-2.5-pro": "gemini-2.5-pro-preview-06-05",
19
- "gemini-2.5-pro-preview-03-25": "gemini-2.5-pro-preview-06-05",
17
+ gemini: "gemini-2.5-pro",
18
+ "gemini-2.5": "gemini-2.5-pro",
19
+ "gemini-2.5-pro": "gemini-2.5-pro",
20
20
  o3: "o3-2025-04-16",
21
21
  "o3-2025-04-16": "o3-2025-04-16",
22
22
  "o4-mini": "o4-mini-2025-04-16",
@@ -1,7 +1,6 @@
1
1
  export declare function hasDownloadScript(repoPath: string): Promise<boolean>;
2
- export declare function downloadBuild({ buildUrl, repoPath, apiKey, }: {
2
+ export declare function downloadBuild({ buildUrl, repoPath, }: {
3
3
  buildUrl: string;
4
4
  repoPath: string;
5
- apiKey: string | undefined;
6
5
  }): Promise<void>;
7
6
  //# sourceMappingURL=index.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/test-build/index.ts"],"names":[],"mappings":"AAgBA,wBAAsB,iBAAiB,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC,CAG1E;AAED,wBAAsB,aAAa,CAAC,EAClC,QAAQ,EACR,QAAQ,EACR,MAAM,GACP,EAAE;IACD,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,GAAG,SAAS,CAAC;CAC5B,GAAG,OAAO,CAAC,IAAI,CAAC,CAWhB"}
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/test-build/index.ts"],"names":[],"mappings":"AAgBA,wBAAsB,iBAAiB,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC,CAG1E;AAED,wBAAsB,aAAa,CAAC,EAClC,QAAQ,EACR,QAAQ,GACT,EAAE;IACD,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;CAClB,GAAG,OAAO,CAAC,IAAI,CAAC,CAWhB"}
@@ -18,7 +18,7 @@ async function hasDownloadScript(repoPath) {
18
18
  const packageJSON = await getPackageJSON(repoPath);
19
19
  return !!packageJSON.scripts["download"];
20
20
  }
21
- async function downloadBuild({ buildUrl, repoPath, apiKey, }) {
21
+ async function downloadBuild({ buildUrl, repoPath, }) {
22
22
  const logger = new logger_1.CustomLogger({ useReporter: false });
23
23
  const packageJSON = await getPackageJSON(repoPath);
24
24
  const buildDownloadScript = packageJSON.scripts["download"];
@@ -26,7 +26,7 @@ async function downloadBuild({ buildUrl, repoPath, apiKey, }) {
26
26
  logger.log(`Downloading build from ${buildUrl}`);
27
27
  await (0, exec_1.cmd)(`npm`, ["run", "download", buildUrl], {
28
28
  cwd: repoPath,
29
- envOverrides: { ...Object(process.env), EMPIRICALRUN_API_KEY: apiKey },
29
+ envOverrides: { ...Object(process.env) },
30
30
  });
31
31
  }
32
32
  }
@@ -1,5 +1,6 @@
1
1
  import { PendingToolCall, ServicePayload, Tool } from "@empiricalrun/shared-types";
2
2
  export declare function appendBranchNameToQueueUrl(queueUrl: string, branchName: string): string;
3
+ export declare function replaceRegionInUrl(url: string, newRegion: string): string;
3
4
  export declare function getQueueUrl(toolCalls: PendingToolCall[], tools: Tool[]): string;
4
5
  export declare function sendToolRequestToRemoteQueue(queueUrl: string, payload: ServicePayload): Promise<void>;
5
6
  //# sourceMappingURL=utils.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../src/tool-call-service/utils.ts"],"names":[],"mappings":"AACA,OAAO,EACL,eAAe,EACf,cAAc,EACd,IAAI,EACL,MAAM,4BAA4B,CAAC;AAEpC,wBAAgB,0BAA0B,CACxC,QAAQ,EAAE,MAAM,EAChB,UAAU,EAAE,MAAM,GACjB,MAAM,CAOR;AAYD,wBAAgB,WAAW,CACzB,SAAS,EAAE,eAAe,EAAE,EAC5B,KAAK,EAAE,IAAI,EAAE,GACZ,MAAM,CAeR;AAED,wBAAsB,4BAA4B,CAChD,QAAQ,EAAE,MAAM,EAChB,OAAO,EAAE,cAAc,iBAiBxB"}
1
+ {"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../src/tool-call-service/utils.ts"],"names":[],"mappings":"AACA,OAAO,EACL,eAAe,EACf,cAAc,EACd,IAAI,EACL,MAAM,4BAA4B,CAAC;AAEpC,wBAAgB,0BAA0B,CACxC,QAAQ,EAAE,MAAM,EAChB,UAAU,EAAE,MAAM,GACjB,MAAM,CAOR;AAED,wBAAgB,kBAAkB,CAAC,GAAG,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,GAAG,MAAM,CAOzE;AAgBD,wBAAgB,WAAW,CACzB,SAAS,EAAE,eAAe,EAAE,EAC5B,KAAK,EAAE,IAAI,EAAE,GACZ,MAAM,CAeR;AAED,wBAAsB,4BAA4B,CAChD,QAAQ,EAAE,MAAM,EAChB,OAAO,EAAE,cAAc,iBAiBxB"}
@@ -1,6 +1,7 @@
1
1
  "use strict";
2
2
  Object.defineProperty(exports, "__esModule", { value: true });
3
3
  exports.appendBranchNameToQueueUrl = appendBranchNameToQueueUrl;
4
+ exports.replaceRegionInUrl = replaceRegionInUrl;
4
5
  exports.getQueueUrl = getQueueUrl;
5
6
  exports.sendToolRequestToRemoteQueue = sendToolRequestToRemoteQueue;
6
7
  const client_sqs_1 = require("@aws-sdk/client-sqs");
@@ -12,12 +13,18 @@ function appendBranchNameToQueueUrl(queueUrl, branchName) {
12
13
  const queueNameWithoutFifo = queueName?.replace(/\.fifo$/, "");
13
14
  return `${parts.join("/")}/${queueNameWithoutFifo}-${suffix}.fifo`;
14
15
  }
16
+ function replaceRegionInUrl(url, newRegion) {
17
+ const parsedUrl = new URL(url);
18
+ parsedUrl.hostname = parsedUrl.hostname.replace(/\.([a-z0-9-]+)\.amazonaws\.com$/, `.${newRegion}.amazonaws.com`);
19
+ return parsedUrl.toString();
20
+ }
15
21
  function buildQueueUrlsForPreview(baseQueueUrl) {
16
22
  const branch = process.env.VERCEL_GIT_COMMIT_REF;
17
23
  if (!branch) {
18
24
  throw new Error("VERCEL_GIT_COMMIT_REF is not set");
19
25
  }
20
- const queueUrl = appendBranchNameToQueueUrl(baseQueueUrl, branch);
26
+ const updatedRegionForQueue = replaceRegionInUrl(baseQueueUrl, process.env.AWS_REGION);
27
+ const queueUrl = appendBranchNameToQueueUrl(updatedRegionForQueue, branch);
21
28
  console.log("queueUrl for preview", queueUrl);
22
29
  return queueUrl;
23
30
  }
@@ -17,7 +17,7 @@ Environment details will include the build URL.`,
17
17
  needsBrowser: false,
18
18
  execute: async (params) => {
19
19
  const { input } = params;
20
- const { repoPath, apiKey } = params;
20
+ const { repoPath } = params;
21
21
  if (!(await (0, test_build_1.hasDownloadScript)(repoPath))) {
22
22
  return {
23
23
  isError: true,
@@ -27,7 +27,7 @@ You probably don't need to worry about this, since it means this repo does not h
27
27
  }
28
28
  const { buildUrl } = input;
29
29
  try {
30
- await (0, test_build_1.downloadBuild)({ buildUrl, repoPath, apiKey });
30
+ await (0, test_build_1.downloadBuild)({ buildUrl, repoPath });
31
31
  return {
32
32
  isError: false,
33
33
  result: "Build downloaded successfully",
@@ -1 +1 @@
1
- {"version":3,"file":"test-gen-browser.d.ts","sourceRoot":"","sources":["../../src/tools/test-gen-browser.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EACV,IAAI,EAGL,MAAM,4BAA4B,CAAC;AAoGpC,eAAO,MAAM,4BAA4B,EAAE,IAoM1C,CAAC"}
1
+ {"version":3,"file":"test-gen-browser.d.ts","sourceRoot":"","sources":["../../src/tools/test-gen-browser.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EACV,IAAI,EAGL,MAAM,4BAA4B,CAAC;AAoGpC,eAAO,MAAM,4BAA4B,EAAE,IA+L1C,CAAC"}
@@ -153,21 +153,17 @@ exports.generateTestWithBrowserAgent = {
153
153
  catch (err) {
154
154
  console.warn("[generateTestWithBrowserAgent] Error preparing playwright for codegen", err);
155
155
  }
156
- let envOverrides = { ...environmentOverrides };
156
+ let envOverrides = {
157
+ ...environmentOverrides,
158
+ // Running browser agent in headed implies using page.pause method for codegen
159
+ RUN_PLAYWRIGHT_HEADED: "true",
160
+ };
157
161
  if (input.buildUrl) {
158
162
  envOverrides = {
159
163
  ...envOverrides,
160
164
  BUILD_URL: input.buildUrl,
161
165
  };
162
166
  }
163
- if (featureFlags.includes("headed_codegen_in_browser_agent")) {
164
- // Running browser agent in headed implies using page.pause method for codegen
165
- // which improves codegen quality when script injection fails (iframes, etc.)
166
- envOverrides = {
167
- ...envOverrides,
168
- RUN_PLAYWRIGHT_HEADED: "true",
169
- };
170
- }
171
167
  const testGenToken = (0, scenarios_1.buildTokenFromOptions)({ name: testName, file: filePath, prompt: input.changeToMake }, { useComputerUseAgent: true });
172
168
  console.log("[generateTestWithBrowserAgent] Validations passed, starting agent");
173
169
  const agentResult = await (0, run_1.runBrowsingAgent)({
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@empiricalrun/test-gen",
3
- "version": "0.69.5",
3
+ "version": "0.69.7",
4
4
  "publishConfig": {
5
5
  "registry": "https://registry.npmjs.org/",
6
6
  "access": "public"
@@ -64,7 +64,7 @@
64
64
  "tsx": "^4.16.2",
65
65
  "typescript": "^5.3.3",
66
66
  "zod": "^3.23.8",
67
- "@empiricalrun/llm": "^0.19.1",
67
+ "@empiricalrun/llm": "^0.19.3",
68
68
  "@empiricalrun/r2-uploader": "^0.3.9",
69
69
  "@empiricalrun/test-run": "^0.10.6"
70
70
  },
@@ -82,7 +82,7 @@
82
82
  "playwright": "1.53.0",
83
83
  "serve-handler": "^6.1.6",
84
84
  "ts-patch": "^3.3.0",
85
- "@empiricalrun/shared-types": "0.6.0"
85
+ "@empiricalrun/shared-types": "0.6.1"
86
86
  },
87
87
  "scripts": {
88
88
  "dev": "tspc --build --watch",
@@ -1 +1 @@
1
- {"root":["./src/index.ts","./src/logger.ts","./src/actions/assert.ts","./src/actions/click.ts","./src/actions/done.ts","./src/actions/fill.ts","./src/actions/goto.ts","./src/actions/hover.ts","./src/actions/index.ts","./src/actions/next-task.ts","./src/actions/press.ts","./src/actions/skill.ts","./src/actions/text-content.ts","./src/actions/constants/index.ts","./src/actions/utils/index.ts","./src/agent/browsing/index.ts","./src/agent/browsing/run.ts","./src/agent/browsing/utils.ts","./src/agent/chat/agent-loop.ts","./src/agent/chat/exports.ts","./src/agent/chat/index.ts","./src/agent/chat/models.ts","./src/agent/chat/state.ts","./src/agent/chat/types.ts","./src/agent/chat/utils.ts","./src/agent/chat/prompt/index.ts","./src/agent/chat/prompt/pw-utils-docs.ts","./src/agent/chat/prompt/repo.ts","./src/agent/codegen/create-test-block.ts","./src/agent/codegen/fix-ts-errors.ts","./src/agent/codegen/generate-code-apply-changes.ts","./src/agent/codegen/lexical-scoped-vars.ts","./src/agent/codegen/repo-edit.ts","./src/agent/codegen/run.ts","./src/agent/codegen/skills-retriever.ts","./src/agent/codegen/test-update-feedback.ts","./src/agent/codegen/types.ts","./src/agent/codegen/update-flow.ts","./src/agent/codegen/use-skill.ts","./src/agent/codegen/utils.ts","./src/agent/cua/computer.ts","./src/agent/cua/index.ts","./src/agent/cua/model.ts","./src/agent/cua/pw-codegen/element-from-point.ts","./src/agent/cua/pw-codegen/types.ts","./src/agent/cua/pw-codegen/pw-pause/for-recorder.ts","./src/agent/cua/pw-codegen/pw-pause/index.ts","./src/agent/cua/pw-codegen/pw-pause/ipc.ts","./src/agent/cua/pw-codegen/pw-pause/patch.ts","./src/agent/cua/pw-codegen/pw-pause/types.ts","./src/agent/diagnosis-agent/index.ts","./src/agent/diagnosis-agent/strict-mode-violation.ts","./src/agent/enrich-prompt/index.ts","./src/agent/enrich-prompt/utils.ts","./src/agent/infer-agent/index.ts","./src/agent/master/action-tool-calls.ts","./src/agent/master/element-annotation.ts","./src/agent/master/execute-browser-action.ts","./src/agent/master/execute-skill-action.ts","./src/agent/master/next-action.ts","./src/agent/master/planner.ts","./src/agent/master/run.ts","./src/agent/master/scroller.ts","./src/agent/master/with-hints.ts","./src/agent/master/browser-tests/cua.spec.ts","./src/agent/master/browser-tests/fixtures.ts","./src/agent/master/browser-tests/index.spec.ts","./src/agent/master/browser-tests/skills.spec.ts","./src/agent/master/icon-descriptor/index.ts","./src/agent/master/icon-descriptor/normalize-svg.ts","./src/agent/planner/run-time-planner.ts","./src/agent/planner/run.ts","./src/artifacts/index.ts","./src/artifacts/utils.ts","./src/auth/api-client.ts","./src/auth/cli-auth.ts","./src/auth/index.ts","./src/auth/token-store.ts","./src/bin/index.ts","./src/bin/setup.ts","./src/bin/logger/index.ts","./src/bin/utils/context.ts","./src/bin/utils/index.ts","./src/bin/utils/fs/index.ts","./src/bin/utils/platform/web/index.ts","./src/bin/utils/platform/web/test-files/ts-path-import-validate.ts","./src/bin/utils/scenarios/index.ts","./src/browser-injected-scripts/annotate-elements.spec.ts","./src/constants/index.ts","./src/errors/index.ts","./src/evals/add-scenario-agent.evals.ts","./src/evals/append-create-test-agent.evals.ts","./src/evals/fetch-pom-skills-agent.evals.ts","./src/evals/infer-master-or-code-agent.evals.ts","./src/evals/master-agent.evals.ts","./src/evals/type.ts","./src/evals/update-scenario-agent.evals.ts","./src/file/client.ts","./src/file/server.ts","./src/human-in-the-loop/cli.ts","./src/human-in-the-loop/index.ts","./src/human-in-the-loop/ipc.ts","./src/page/index.ts","./src/prompts/lib/ts-transformer.ts","./src/recorder/env-variables.ts","./src/recorder/glob.ts","./src/recorder/index.ts","./src/recorder/request.ts","./src/recorder/temp-files.ts","./src/recorder/upload.ts","./src/recorder/validation.ts","./src/reporter/index.ts","./src/reporter/lib.ts","./src/session/index.ts","./src/test-build/index.ts","./src/tool-call-service/index.ts","./src/tool-call-service/utils.ts","./src/tools/commit-and-create-pr.ts","./src/tools/diagnosis-fetcher.ts","./src/tools/download-build.ts","./src/tools/list-environments.ts","./src/tools/str_replace_editor.ts","./src/tools/test-gen-browser.ts","./src/tools/test-run.ts","./src/tools/grep/index.ts","./src/tools/grep/ripgrep/index.ts","./src/tools/grep/ripgrep/types.ts","./src/tools/test-run-fetcher/index.ts","./src/tools/test-run-fetcher/types.ts","./src/tools/upgrade-packages/index.ts","./src/tools/upgrade-packages/utils.ts","./src/tools/utils/index.ts","./src/types/handlebars.d.ts","./src/types/index.ts","./src/uploader/index.ts","./src/uploader/utils.ts","./src/utils/checkpoint.ts","./src/utils/env.ts","./src/utils/exec.ts","./src/utils/file-tree.ts","./src/utils/file.ts","./src/utils/git.ts","./src/utils/html.ts","./src/utils/index.ts","./src/utils/json.ts","./src/utils/repo-tree.ts","./src/utils/slug.ts","./src/utils/string.ts","./src/utils/stripAnsi.ts"],"version":"5.8.3"}
1
+ {"root":["./src/index.ts","./src/logger.ts","./src/actions/assert.ts","./src/actions/click.ts","./src/actions/done.ts","./src/actions/fill.ts","./src/actions/goto.ts","./src/actions/hover.ts","./src/actions/index.ts","./src/actions/next-task.ts","./src/actions/press.ts","./src/actions/skill.ts","./src/actions/text-content.ts","./src/actions/constants/index.ts","./src/actions/utils/index.ts","./src/agent/browsing/index.ts","./src/agent/browsing/run.ts","./src/agent/browsing/utils.ts","./src/agent/chat/agent-loop.ts","./src/agent/chat/exports.ts","./src/agent/chat/index.ts","./src/agent/chat/models.ts","./src/agent/chat/state.ts","./src/agent/chat/types.ts","./src/agent/chat/utils.ts","./src/agent/chat/prompt/index.ts","./src/agent/chat/prompt/pw-utils-docs.ts","./src/agent/chat/prompt/repo.ts","./src/agent/codegen/create-test-block.ts","./src/agent/codegen/fix-ts-errors.ts","./src/agent/codegen/generate-code-apply-changes.ts","./src/agent/codegen/lexical-scoped-vars.ts","./src/agent/codegen/repo-edit.ts","./src/agent/codegen/run.ts","./src/agent/codegen/skills-retriever.ts","./src/agent/codegen/test-update-feedback.ts","./src/agent/codegen/types.ts","./src/agent/codegen/update-flow.ts","./src/agent/codegen/use-skill.ts","./src/agent/codegen/utils.ts","./src/agent/cua/computer.ts","./src/agent/cua/index.ts","./src/agent/cua/model.ts","./src/agent/cua/pw-codegen/element-from-point.ts","./src/agent/cua/pw-codegen/types.ts","./src/agent/cua/pw-codegen/pw-pause/for-recorder.ts","./src/agent/cua/pw-codegen/pw-pause/index.ts","./src/agent/cua/pw-codegen/pw-pause/ipc.ts","./src/agent/cua/pw-codegen/pw-pause/patch.ts","./src/agent/cua/pw-codegen/pw-pause/types.ts","./src/agent/cua/pw-codegen/pw-pause/utils.ts","./src/agent/diagnosis-agent/index.ts","./src/agent/diagnosis-agent/strict-mode-violation.ts","./src/agent/enrich-prompt/index.ts","./src/agent/enrich-prompt/utils.ts","./src/agent/infer-agent/index.ts","./src/agent/master/action-tool-calls.ts","./src/agent/master/element-annotation.ts","./src/agent/master/execute-browser-action.ts","./src/agent/master/execute-skill-action.ts","./src/agent/master/next-action.ts","./src/agent/master/planner.ts","./src/agent/master/run.ts","./src/agent/master/scroller.ts","./src/agent/master/with-hints.ts","./src/agent/master/browser-tests/cua.spec.ts","./src/agent/master/browser-tests/fixtures.ts","./src/agent/master/browser-tests/index.spec.ts","./src/agent/master/browser-tests/skills.spec.ts","./src/agent/master/icon-descriptor/index.ts","./src/agent/master/icon-descriptor/normalize-svg.ts","./src/agent/planner/run-time-planner.ts","./src/agent/planner/run.ts","./src/artifacts/index.ts","./src/artifacts/utils.ts","./src/auth/api-client.ts","./src/auth/cli-auth.ts","./src/auth/index.ts","./src/auth/token-store.ts","./src/bin/index.ts","./src/bin/setup.ts","./src/bin/logger/index.ts","./src/bin/utils/context.ts","./src/bin/utils/index.ts","./src/bin/utils/fs/index.ts","./src/bin/utils/platform/web/index.ts","./src/bin/utils/platform/web/test-files/ts-path-import-validate.ts","./src/bin/utils/scenarios/index.ts","./src/browser-injected-scripts/annotate-elements.spec.ts","./src/constants/index.ts","./src/errors/index.ts","./src/evals/add-scenario-agent.evals.ts","./src/evals/append-create-test-agent.evals.ts","./src/evals/fetch-pom-skills-agent.evals.ts","./src/evals/infer-master-or-code-agent.evals.ts","./src/evals/master-agent.evals.ts","./src/evals/type.ts","./src/evals/update-scenario-agent.evals.ts","./src/file/client.ts","./src/file/server.ts","./src/human-in-the-loop/cli.ts","./src/human-in-the-loop/index.ts","./src/human-in-the-loop/ipc.ts","./src/page/index.ts","./src/prompts/lib/ts-transformer.ts","./src/recorder/env-variables.ts","./src/recorder/glob.ts","./src/recorder/index.ts","./src/recorder/request.ts","./src/recorder/temp-files.ts","./src/recorder/upload.ts","./src/recorder/validation.ts","./src/reporter/index.ts","./src/reporter/lib.ts","./src/session/index.ts","./src/test-build/index.ts","./src/tool-call-service/index.ts","./src/tool-call-service/utils.ts","./src/tools/commit-and-create-pr.ts","./src/tools/diagnosis-fetcher.ts","./src/tools/download-build.ts","./src/tools/list-environments.ts","./src/tools/str_replace_editor.ts","./src/tools/test-gen-browser.ts","./src/tools/test-run.ts","./src/tools/grep/index.ts","./src/tools/grep/ripgrep/index.ts","./src/tools/grep/ripgrep/types.ts","./src/tools/test-run-fetcher/index.ts","./src/tools/test-run-fetcher/types.ts","./src/tools/upgrade-packages/index.ts","./src/tools/upgrade-packages/utils.ts","./src/tools/utils/index.ts","./src/types/handlebars.d.ts","./src/types/index.ts","./src/uploader/index.ts","./src/uploader/utils.ts","./src/utils/checkpoint.ts","./src/utils/env.ts","./src/utils/exec.ts","./src/utils/file-tree.ts","./src/utils/file.ts","./src/utils/git.ts","./src/utils/html.ts","./src/utils/index.ts","./src/utils/json.ts","./src/utils/repo-tree.ts","./src/utils/slug.ts","./src/utils/string.ts","./src/utils/stripAnsi.ts"],"version":"5.8.3"}