@agentium/browser 2.0.8 → 2.1.1

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/dist/index.js CHANGED
@@ -1,5 +1,5 @@
1
1
  // src/browser-agent.ts
2
- import { EventBus, Logger, MemoryManager } from "@agentium/core";
2
+ import { EventBus, Logger, MemoryManager, RunContext } from "@agentium/core";
3
3
  import { z } from "zod";
4
4
 
5
5
  // src/stealth.ts
@@ -80,29 +80,39 @@ function getStealthScript() {
80
80
  };
81
81
  }
82
82
 
83
- // \u2500\u2500 Prevent iframe detection of automation \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500
84
- Object.defineProperty(HTMLIFrameElement.prototype, 'contentWindow', {
85
- get: function() {
86
- return window;
87
- },
88
- });
89
-
90
- // \u2500\u2500 Remove "cdc_" Playwright/ChromeDriver markers from DOM \u2500\u2500\u2500\u2500\u2500\u2500\u2500
91
- const observer = new MutationObserver((mutations) => {
92
- for (const mutation of mutations) {
93
- for (const node of mutation.addedNodes) {
94
- if (node.nodeType === 1) {
95
- const el = node;
96
- for (const attr of [...el.attributes]) {
97
- if (attr.name.startsWith('cdc_') || attr.name.startsWith('__playwright')) {
98
- el.removeAttribute(attr.name);
83
+ // \u2500\u2500 Remove "cdc_" Playwright/ChromeDriver markers from new elements \u2500
84
+ // Only watches childList additions, NOT every attribute mutation \u2014
85
+ // attributes:true subtree:true is O(every DOM write) on heavy SPAs
86
+ // like FreightOS and noticeably slows the page.
87
+ try {
88
+ const observer = new MutationObserver((mutations) => {
89
+ for (const mutation of mutations) {
90
+ for (const node of mutation.addedNodes) {
91
+ if (node && node.nodeType === 1) {
92
+ const el = node;
93
+ if (!el.attributes) continue;
94
+ for (const attr of [...el.attributes]) {
95
+ if (attr.name.startsWith('cdc_') || attr.name.startsWith('__playwright')) {
96
+ el.removeAttribute(attr.name);
97
+ }
99
98
  }
100
99
  }
101
100
  }
102
101
  }
102
+ });
103
+ if (document.documentElement) {
104
+ observer.observe(document.documentElement, { childList: true, subtree: true });
103
105
  }
104
- });
105
- observer.observe(document.documentElement, { attributes: true, childList: true, subtree: true });
106
+ } catch (_e) { /* never break the page over a stealth patch */ }
107
+
108
+ // NOTE on iframe.contentWindow:
109
+ // We deliberately do NOT override HTMLIFrameElement.prototype.contentWindow.
110
+ // An earlier version returned the parent window, which breaks legitimate
111
+ // page JS on any site that uses iframes (checkout widgets, embedded
112
+ // videos, A/B test scripts, analytics). The "anti-detection" value was
113
+ // minimal \u2014 modern detectors don't rely on that property anyway \u2014 and
114
+ // the breakage caused pages to fail to boot, leaving font-test residue
115
+ // on screen.
106
116
  `;
107
117
  }
108
118
  function buildStealthContextOpts(config, viewport) {
@@ -150,6 +160,13 @@ var BrowserProvider = class {
150
160
  _viewport;
151
161
  _videoDir;
152
162
  _humanize;
163
+ /**
164
+ * Most recent DOM snapshot (one per `extractDOM` call). Indexed actions
165
+ * (`clickByIndex`, `inputByIndex`, …) resolve their `index` against this.
166
+ */
167
+ _lastDom = [];
168
+ /** True if we connected over CDP (don't tear down the browser on close). */
169
+ _attached = false;
153
170
  constructor() {
154
171
  this._viewport = { width: 1280, height: 720 };
155
172
  }
@@ -158,8 +175,6 @@ var BrowserProvider = class {
158
175
  const pw = await import("playwright");
159
176
  const chromium = pw.chromium;
160
177
  this._viewport = opts?.viewport ?? { width: 1280, height: 720 };
161
- const stealthEnabled = !!opts?.stealth;
162
- const stealthCfg = typeof opts?.stealth === "object" ? opts.stealth : {};
163
178
  if (opts?.humanize) {
164
179
  const h = typeof opts.humanize === "object" ? opts.humanize : {};
165
180
  this._humanize = {
@@ -169,14 +184,34 @@ var BrowserProvider = class {
169
184
  mouseMovement: h.mouseMovement ?? true
170
185
  };
171
186
  }
172
- const launchOpts = {
173
- headless: opts?.headless ?? true
174
- };
187
+ if (opts?.cdpUrl) {
188
+ this.browser = await chromium.connectOverCDP(opts.cdpUrl);
189
+ this._attached = true;
190
+ const contexts = this.browser.contexts();
191
+ this.context = contexts.length > 0 ? contexts[0] : await this.browser.newContext({ viewport: this._viewport });
192
+ const existingPages = this.context.pages();
193
+ this.page = existingPages.length > 0 ? existingPages[0] : await this.context.newPage();
194
+ this.tabCounter = 0;
195
+ this.activeTabId = "tab-0";
196
+ this.pages.set("tab-0", this.page);
197
+ return;
198
+ }
199
+ const stealthEnabled = !!opts?.stealth;
200
+ const stealthCfg = typeof opts?.stealth === "object" ? opts.stealth : {};
201
+ const headless = opts?.headless ?? true;
202
+ const launchOpts = { headless };
203
+ const args = [];
175
204
  if (stealthEnabled) {
176
- const { args, proxy } = buildStealthLaunchArgs(stealthCfg);
177
- launchOpts.args = args;
178
- if (proxy) launchOpts.proxy = proxy;
205
+ const stealth = buildStealthLaunchArgs(stealthCfg);
206
+ args.push(...stealth.args);
207
+ if (stealth.proxy) launchOpts.proxy = stealth.proxy;
208
+ }
209
+ if (!headless) {
210
+ const CHROME_VERTICAL_CHROME_PX = 140;
211
+ args.push(`--window-size=${this._viewport.width},${this._viewport.height + CHROME_VERTICAL_CHROME_PX}`);
212
+ args.push("--window-position=0,0");
179
213
  }
214
+ if (args.length) launchOpts.args = args;
180
215
  this.browser = await chromium.launch(launchOpts);
181
216
  let contextOpts;
182
217
  if (stealthEnabled) {
@@ -235,7 +270,11 @@ var BrowserProvider = class {
235
270
  get viewport() {
236
271
  return this._viewport;
237
272
  }
238
- // ── Interaction (with optional humanize) ─────────────────────────────
273
+ /** Most recent DOM snapshot. Each entry has a stable `index`. */
274
+ get lastDom() {
275
+ return this._lastDom;
276
+ }
277
+ // ── Coordinate-based Interaction ─────────────────────────────────────
239
278
  async click(x, y) {
240
279
  this.ensurePage();
241
280
  const [cx, cy] = this.clampToViewport(x, y);
@@ -265,6 +304,20 @@ var BrowserProvider = class {
265
304
  this.ensurePage();
266
305
  await this.page.keyboard.press(key);
267
306
  }
307
+ /**
308
+ * Send arbitrary keyboard keys / shortcuts. Accepts a single
309
+ * Playwright key spec (`"Enter"`, `"Control+l"`, `"Shift+ArrowDown"`)
310
+ * or a space-separated sequence (`"Tab Tab Enter"`).
311
+ */
312
+ async sendKeys(keys) {
313
+ this.ensurePage();
314
+ const tokens = keys.trim().split(/\s+/).filter(Boolean);
315
+ for (const token of tokens) {
316
+ await this.page.keyboard.press(token);
317
+ await this.sleep(this._humanize ? this.randInt(40, 120) : 30);
318
+ }
319
+ await this.humanPause();
320
+ }
268
321
  async scroll(direction, amount) {
269
322
  this.ensurePage();
270
323
  const base = amount ?? 400;
@@ -282,13 +335,262 @@ var BrowserProvider = class {
282
335
  }
283
336
  await this.humanPause();
284
337
  }
285
- // ── DOM Extraction ───────────────────────────────────────────────────
338
+ // ── Indexed Interaction (preferred path) ─────────────────────────────
339
+ /**
340
+ * Build a Playwright locator for a DOM-snapshot index. Each `extractDOM`
341
+ * call tags surviving elements with `data-bua-idx="<n>"`; we resolve by
342
+ * that attribute. Returns null if the index is unknown.
343
+ */
344
+ locatorForIndex(index) {
345
+ if (!this.page) return null;
346
+ if (!this._lastDom.find((e) => e.index === index)) return null;
347
+ return this.page.locator(`[data-bua-idx="${index}"]`).first();
348
+ }
349
+ /**
350
+ * Click an element by its DOM-snapshot index. The most reliable click
351
+ * path on dynamic pages — survives layout shifts and DPR oddities.
352
+ */
353
+ async clickByIndex(index, opts) {
354
+ this.ensurePage();
355
+ const loc = this.locatorForIndex(index);
356
+ if (!loc) return false;
357
+ try {
358
+ await loc.scrollIntoViewIfNeeded({ timeout: 1500 }).catch(() => {
359
+ });
360
+ await loc.click({ timeout: opts?.timeout ?? 5e3 });
361
+ await this.humanPause();
362
+ return true;
363
+ } catch {
364
+ return false;
365
+ }
366
+ }
367
+ /**
368
+ * Focus an indexed input, optionally clear it, and type. Returns false
369
+ * if the index couldn't be resolved or the input couldn't be focused.
370
+ */
371
+ async inputByIndex(index, text, opts) {
372
+ this.ensurePage();
373
+ const loc = this.locatorForIndex(index);
374
+ if (!loc) return false;
375
+ try {
376
+ const timeout = opts?.timeout ?? 5e3;
377
+ await loc.scrollIntoViewIfNeeded({ timeout: 1500 }).catch(() => {
378
+ });
379
+ if (opts?.clear !== false) {
380
+ await loc.fill("", { timeout });
381
+ }
382
+ const delay = this._humanize ? this.randInt(this._humanize.typingDelay[0], this._humanize.typingDelay[1]) : 30;
383
+ await loc.click({ timeout });
384
+ await loc.pressSequentially(text, { delay });
385
+ if (opts?.submit) {
386
+ await this.page.keyboard.press("Enter");
387
+ }
388
+ await this.humanPause();
389
+ return true;
390
+ } catch {
391
+ return false;
392
+ }
393
+ }
394
+ async uploadFileByIndex(index, path) {
395
+ this.ensurePage();
396
+ const loc = this.locatorForIndex(index);
397
+ if (!loc) return false;
398
+ try {
399
+ await loc.setInputFiles(path, { timeout: 5e3 });
400
+ await this.humanPause();
401
+ return true;
402
+ } catch {
403
+ return false;
404
+ }
405
+ }
406
+ /** Scroll the indexed element into view (no click). */
407
+ async scrollIntoViewByIndex(index) {
408
+ this.ensurePage();
409
+ const loc = this.locatorForIndex(index);
410
+ if (!loc) return false;
411
+ try {
412
+ await loc.scrollIntoViewIfNeeded({ timeout: 3e3 });
413
+ return true;
414
+ } catch {
415
+ return false;
416
+ }
417
+ }
418
+ // ── Text-based Interaction ───────────────────────────────────────────
419
+ /**
420
+ * Deterministic, DOM-based click using Playwright's text locator.
421
+ *
422
+ * Returns `true` if a matching, visible, clickable element was found and
423
+ * clicked within `timeout` ms; `false` otherwise (so the caller can fall
424
+ * back to coordinate clicking). Substring-matches by default — e.g.
425
+ * `clickByText("Cheapest")` matches "Cheapest · 23-28 days · $2,550".
426
+ */
427
+ async clickByText(keyword, opts) {
428
+ this.ensurePage();
429
+ const timeout = opts?.timeout ?? 3e3;
430
+ try {
431
+ const locator = this.page.locator(`text=${keyword}`).first();
432
+ await locator.click({ timeout });
433
+ await this.humanPause();
434
+ return true;
435
+ } catch {
436
+ return false;
437
+ }
438
+ }
439
+ /**
440
+ * Scroll the first occurrence of `text` into view. Returns false if no
441
+ * match was found within the timeout.
442
+ */
443
+ async findText(text, opts) {
444
+ this.ensurePage();
445
+ const timeout = opts?.timeout ?? 3e3;
446
+ try {
447
+ const locator = this.page.getByText(text).first();
448
+ await locator.scrollIntoViewIfNeeded({ timeout });
449
+ return true;
450
+ } catch {
451
+ return false;
452
+ }
453
+ }
454
+ // ── Dropdowns ────────────────────────────────────────────────────────
455
+ /**
456
+ * Read the options of a native `<select>` at the given DOM-snapshot
457
+ * index. Returns `[]` if the element is not a `<select>`.
458
+ */
459
+ async dropdownOptions(index) {
460
+ this.ensurePage();
461
+ const loc = this.locatorForIndex(index);
462
+ if (!loc) return [];
463
+ try {
464
+ return await loc.evaluate((el) => {
465
+ if (!el || el.tagName !== "SELECT") return [];
466
+ return Array.from(el.options).map((opt) => ({
467
+ value: opt.value,
468
+ label: (opt.label || opt.textContent || "").trim(),
469
+ selected: !!opt.selected
470
+ }));
471
+ });
472
+ } catch {
473
+ return [];
474
+ }
475
+ }
476
+ /**
477
+ * Select an option in a native `<select>` by its visible text or value.
478
+ * Returns false if the element isn't a `<select>` or no option matched.
479
+ */
480
+ async selectDropdown(index, text) {
481
+ this.ensurePage();
482
+ const loc = this.locatorForIndex(index);
483
+ if (!loc) return false;
484
+ try {
485
+ await loc.selectOption({ label: text }, { timeout: 3e3 });
486
+ await this.humanPause();
487
+ return true;
488
+ } catch {
489
+ try {
490
+ await loc.selectOption({ value: text }, { timeout: 3e3 });
491
+ await this.humanPause();
492
+ return true;
493
+ } catch {
494
+ return false;
495
+ }
496
+ }
497
+ }
498
+ // ── JS Evaluation ────────────────────────────────────────────────────
499
+ /**
500
+ * Run arbitrary JS in the page context. The caller is responsible for
501
+ * gating this behind a config flag — the BrowserAgent only routes the
502
+ * `evaluate` action here when `allowEvaluate: true`.
503
+ *
504
+ * The code is wrapped in `(async () => { ... })()` and the return value
505
+ * is coerced to a string for the model.
506
+ */
507
+ async evaluate(code) {
508
+ this.ensurePage();
509
+ try {
510
+ const result = await this.page.evaluate(
511
+ // eslint-disable-next-line @typescript-eslint/no-implied-eval
512
+ new Function("return (async () => { " + code + " })()")
513
+ );
514
+ if (result === void 0) return "undefined";
515
+ if (result === null) return "null";
516
+ if (typeof result === "string") return result;
517
+ try {
518
+ return JSON.stringify(result);
519
+ } catch {
520
+ return String(result);
521
+ }
522
+ } catch (e) {
523
+ throw new Error(`evaluate failed: ${e?.message ?? e}`);
524
+ }
525
+ }
526
+ // ── Page text (for `extract`) ────────────────────────────────────────
527
+ /**
528
+ * Returns a clean text representation of the visible page body, with
529
+ * optional link extraction. Used by the BrowserAgent's `extract` action
530
+ * — the text is passed to a (usually cheap) LLM with the user's query.
531
+ */
532
+ async pageText(opts) {
533
+ this.ensurePage();
534
+ const maxChars = opts?.maxChars ?? 2e4;
535
+ const extractLinks = !!opts?.extractLinks;
536
+ const raw = await this.page.evaluate((withLinks) => {
537
+ const doc = globalThis.document;
538
+ const SKIP = /* @__PURE__ */ new Set(["SCRIPT", "STYLE", "NOSCRIPT", "TEMPLATE", "SVG"]);
539
+ const win = globalThis.window;
540
+ const lines = [];
541
+ function visit(node) {
542
+ if (!node) return;
543
+ if (node.nodeType === 3) {
544
+ const t = (node.nodeValue || "").trim();
545
+ if (t) lines.push(t);
546
+ return;
547
+ }
548
+ if (node.nodeType !== 1) return;
549
+ const tag = node.tagName.toUpperCase();
550
+ if (SKIP.has(tag)) return;
551
+ const style = win.getComputedStyle(node);
552
+ if (style && (style.visibility === "hidden" || style.display === "none")) return;
553
+ if (tag === "A" && withLinks) {
554
+ const href = node.getAttribute("href") || "";
555
+ const txt = (node.innerText || node.textContent || "").trim();
556
+ if (txt && href) {
557
+ lines.push(`[${txt}](${href})`);
558
+ return;
559
+ }
560
+ }
561
+ for (const child of node.childNodes) visit(child);
562
+ }
563
+ visit(doc.body);
564
+ return lines.join("\n");
565
+ }, extractLinks);
566
+ const collapsed = raw.replace(/\n{3,}/g, "\n\n").trim();
567
+ return collapsed.length > maxChars ? `${collapsed.slice(0, maxChars)}
568
+ \u2026[truncated]` : collapsed;
569
+ }
570
+ // ── DOM Extraction (with index tagging) ──────────────────────────────
571
+ /**
572
+ * Snapshot the interactive elements visible in the viewport, tag each
573
+ * with a `data-bua-idx="<n>"` attribute (used by indexed actions), and
574
+ * return BOTH a human-readable string (for the model) and the
575
+ * structured list (for the runtime).
576
+ *
577
+ * Three properties matter for accuracy:
578
+ * - Hit-tested: each listed coordinate / index actually reaches the
579
+ * labeled element (overlays / occlusion skip the entry).
580
+ * - Visibility-filtered: invisible, zero-size, `pointer-events: none`,
581
+ * and near-zero-opacity elements are excluded.
582
+ * - `cursor: pointer` fallback pass: a second scan catches custom React
583
+ * widgets that have no semantic role/href/onclick but are clickable.
584
+ */
286
585
  async extractDOM(opts) {
287
586
  this.ensurePage();
288
587
  const max = opts?.maxElements ?? 120;
289
- const elements = await this.page.evaluate((limit) => {
588
+ const raw = await this.page.evaluate((limit) => {
290
589
  const doc = globalThis.document;
291
590
  const win = globalThis.window;
591
+ for (const stale of doc.querySelectorAll("[data-bua-idx]")) {
592
+ stale.removeAttribute("data-bua-idx");
593
+ }
292
594
  const selectors = [
293
595
  "a[href]",
294
596
  "button",
@@ -351,6 +653,13 @@ var BrowserProvider = class {
351
653
  } catch {
352
654
  }
353
655
  seen.add(el);
656
+ entries.push({ el, cx, cy });
657
+ }
658
+ entries.sort((a, b) => a.cy === b.cy ? a.cx - b.cx : a.cy - b.cy);
659
+ const out = [];
660
+ let idx = 1;
661
+ for (const { el, cx, cy } of entries) {
662
+ if (out.length >= limit) break;
354
663
  const tag = el.tagName.toLowerCase();
355
664
  const role = el.getAttribute("role") || tag;
356
665
  const type = el.getAttribute("type") || "";
@@ -365,39 +674,29 @@ var BrowserProvider = class {
365
674
  let label = ariaLabel || text || placeholder || title || value || name;
366
675
  if (!label && href) label = href.slice(0, 60);
367
676
  if (!label) label = `(${tag}${type ? ` type=${type}` : ""})`;
368
- entries.push({
677
+ el.setAttribute("data-bua-idx", String(idx));
678
+ out.push({
679
+ index: idx,
369
680
  cx,
370
681
  cy,
371
- line: `[${cx},${cy}] ${role}${type ? `(${type})` : ""}: "${label}"`
682
+ role,
683
+ type: type || void 0,
684
+ label,
685
+ tag,
686
+ isInput: tag === "input" || tag === "textarea" || el.isContentEditable === true,
687
+ isSelect: tag === "select",
688
+ isFile: tag === "input" && type === "file"
372
689
  });
690
+ idx++;
373
691
  }
374
- entries.sort((a, b) => a.cy === b.cy ? a.cx - b.cx : a.cy - b.cy);
375
- return entries.slice(0, limit).map((e) => e.line).join("\n");
692
+ return out;
376
693
  }, max);
377
- return elements;
378
- }
379
- /**
380
- * Deterministic, DOM-based click using Playwright's text locator.
381
- *
382
- * Returns `true` if a matching, visible, clickable element was found and
383
- * clicked within `timeout` ms; `false` otherwise (so the caller can fall
384
- * back to coordinate clicking). Substring-matches by default — e.g.
385
- * `clickByText("Cheapest")` matches "Cheapest · 23-28 days · $2,550".
386
- *
387
- * This is the same pattern browser-use uses in DOM-only mode and is far
388
- * more reliable than coordinate clicks for text-bearing targets.
389
- */
390
- async clickByText(keyword, opts) {
391
- this.ensurePage();
392
- const timeout = opts?.timeout ?? 3e3;
393
- try {
394
- const locator = this.page.locator(`text=${keyword}`).first();
395
- await locator.click({ timeout });
396
- await this.humanPause();
397
- return true;
398
- } catch {
399
- return false;
400
- }
694
+ this._lastDom = raw;
695
+ const lines = raw.map((e) => {
696
+ const typeSuffix = e.type ? `(${e.type})` : "";
697
+ return `[${e.index}] [${e.cx},${e.cy}] ${e.role}${typeSuffix}: "${e.label}"`;
698
+ });
699
+ return { text: lines.join("\n"), elements: raw };
401
700
  }
402
701
  // ── Page Info ────────────────────────────────────────────────────────
403
702
  async getPageInfo() {
@@ -482,12 +781,18 @@ var BrowserProvider = class {
482
781
  // ── Cleanup ──────────────────────────────────────────────────────────
483
782
  async close() {
484
783
  try {
485
- if (this.context) await this.context.close();
784
+ if (this.context && !this._attached) await this.context.close();
486
785
  } catch (err) {
487
786
  console.warn("[agentium/browser] Error closing context:", err instanceof Error ? err.message : err);
488
787
  }
489
788
  try {
490
- if (this.browser) await this.browser.close();
789
+ if (this.browser && !this._attached) await this.browser.close();
790
+ else if (this.browser && this._attached) {
791
+ try {
792
+ await this.browser.close();
793
+ } catch {
794
+ }
795
+ }
491
796
  } catch (err) {
492
797
  console.warn("[agentium/browser] Error closing browser:", err instanceof Error ? err.message : err);
493
798
  }
@@ -495,6 +800,7 @@ var BrowserProvider = class {
495
800
  this.context = null;
496
801
  this.browser = null;
497
802
  this.pages.clear();
803
+ this._lastDom = [];
498
804
  }
499
805
  // ── Private: Humanize helpers ────────────────────────────────────────
500
806
  /** Add small random offset to coordinates to avoid pixel-perfect bot patterns. */
@@ -516,8 +822,7 @@ var BrowserProvider = class {
516
822
  return [cx, cy];
517
823
  }
518
824
  /**
519
- * Simulate human mouse movement using Bézier-like interpolation.
520
- * Moves from the current mouse position to the target in small steps.
825
+ * Simulate human mouse movement using smoothstep interpolation.
521
826
  */
522
827
  async humanMouseMove(targetX, targetY) {
523
828
  const steps = this.randInt(5, 12);
@@ -554,90 +859,144 @@ var BrowserProvider = class {
554
859
  };
555
860
 
556
861
  // src/prompts.ts
557
- function buildSystemPrompt(viewport, extraInstructions, credentialKeys) {
558
- const lines = [
559
- `You are a browser automation agent. You receive a screenshot of a web browser and decide what action to take next to complete the user's task.`,
862
+ function buildSystemPrompt(viewport, extraInstructions, credentialKeys, options) {
863
+ const maxActions = options?.maxActionsPerStep ?? 3;
864
+ const allowEvaluate = !!options?.allowEvaluate;
865
+ const tools = options?.tools ?? [];
866
+ const useVision = options?.useVision ?? "auto";
867
+ if (options?.overrideSystemMessage) {
868
+ const lines2 = [options.overrideSystemMessage];
869
+ appendCredentials(lines2, credentialKeys);
870
+ lines2.push("", "## Response Format", responseFormatLines(maxActions).join("\n"));
871
+ return lines2.join("\n");
872
+ }
873
+ const lines = [];
874
+ lines.push(
875
+ `You are a browser automation agent. Your job is to complete the user's task by interacting with a real web page.`,
560
876
  ``,
561
- `## Viewport & Coordinate System`,
562
- `- The browser viewport is ${viewport.width}\xD7${viewport.height} CSS pixels.`,
563
- `- The screenshot you are given has EXACTLY the same dimensions: ${viewport.width}\xD7${viewport.height} pixels (top-left = 0,0).`,
564
- `- ALL coordinates you return MUST be within 0..${viewport.width - 1} for x and 0..${viewport.height - 1} for y.`,
565
- `- Coordinates outside this range will be rejected \u2014 your click will go nowhere.`,
566
- `- If a "Interactive elements on page" list is provided below, PREFER its coordinates over your own visual estimation. They are exact element centers in the same coordinate system.`,
877
+ `## What you receive each step`,
878
+ `- The current URL and page title`,
879
+ `- A numbered list of interactive elements visible in the viewport (the "DOM snapshot"). Each entry looks like \`[idx] [cx,cy] role(type): "label"\` where \`idx\` is a stable per-step index, \`cx,cy\` are the CSS-pixel center coordinates, and \`label\` is the visible text / aria-label / placeholder. **Use the index to act on elements** \u2014 it is the most reliable handle.`
880
+ );
881
+ if (useVision !== false) {
882
+ lines.push(
883
+ `- A PNG screenshot of the same viewport at exactly ${viewport.width}\xD7${viewport.height} pixels. Use it for visual context (layout, icons, charts) when the DOM snapshot is ambiguous.`
884
+ );
885
+ } else {
886
+ lines.push(
887
+ `- (Vision is disabled \u2014 no screenshot. Rely entirely on the DOM snapshot and page text.)`
888
+ );
889
+ }
890
+ lines.push(
891
+ `- A short list of your previous actions (so you don't repeat yourself).`,
892
+ ``,
893
+ `## Coordinate System`,
894
+ `The browser viewport is ${viewport.width}\xD7${viewport.height} CSS pixels, top-left origin. Coordinates outside that range are rejected.`,
567
895
  ``,
568
896
  `## Available Actions`,
569
- `Respond with a JSON object containing one of these actions:`,
897
+ `Respond with a JSON object \u2014 or, when several independent actions can be safely batched (e.g. filling multiple form fields), an array of up to ${maxActions} JSON objects. The runtime executes them in order and stops early if the page navigates.`,
570
898
  ``,
571
- `### click`,
572
- `Click at a specific coordinate. Use for buttons, links, inputs, checkboxes, etc.`,
573
- `\`{ "action": "click", "x": <number>, "y": <number>, "description": "<what you are clicking>" }\``,
574
- `IMPORTANT: when the target has visible text, ALWAYS put that text in single or double quotes inside the description, e.g. \`"description": "Click on 'Cheapest' tab"\` or \`"description": "Press the \\"Sign in\\" button"\`. The runtime uses the quoted phrase to do a deterministic DOM text-locator click and only falls back to (x,y) if no match is found, which dramatically improves accuracy on dynamic pages.`,
899
+ `### click \u2014 by index (preferred)`,
900
+ `\`{ "action": "click", "index": <n>, "description": "<what you are clicking, ideally with the visible text in quotes>" }\``,
901
+ `If you cannot find a matching index (e.g. an element is partly off-screen and not in the snapshot), you may fall back to coordinates: \`{ "action": "click", "x": <n>, "y": <n>, "description": "Click on 'Cheapest' tab" }\`. The runtime will additionally try a Playwright text locator using the quoted phrase.`,
575
902
  ``,
576
- `### type`,
577
- `Type text. If x/y are provided, click that position first (to focus the input), then type. If omitted, types into the currently focused element. To press Enter after typing (e.g., to submit a search), append "\\n" to the text.`,
578
- `\`{ "action": "type", "text": "<text to type>", "x": <number|optional>, "y": <number|optional> }\``,
579
- `Example: \`{ "action": "type", "text": "search query\\n", "x": 640, "y": 300 }\` \u2014 clicks the search box, types, and presses Enter.`,
903
+ `### type \u2014 by index (preferred)`,
904
+ `Type into the element at \`index\`. The field is cleared first by default. Set \`"submit": true\` to press Enter after typing (e.g. submit a search).`,
905
+ `\`{ "action": "type", "index": <n>, "text": "<text>", "clear": true, "submit": false }\``,
906
+ `Fallback: \`{ "action": "type", "text": "<text>", "x": <n>, "y": <n> }\` clicks the coordinates first, then types.`,
580
907
  ``,
581
908
  `### scroll`,
582
- `Scroll the page up or down. Use when content is below or above the visible area.`,
583
- `\`{ "action": "scroll", "direction": "up"|"down", "amount": <pixels, optional, default 400> }\``,
909
+ `\`{ "action": "scroll", "direction": "down"|"up", "amount": 400 }\` or \`{ "action": "scroll", "index": <n> }\` to scroll a specific element into view.`,
584
910
  ``,
585
- `### navigate`,
586
- `Navigate to a specific URL. Use when you know the exact URL to visit.`,
587
- `\`{ "action": "navigate", "url": "<full URL>" }\``,
911
+ `### find_text`,
912
+ `Scroll the first occurrence of a phrase into the viewport. Use this instead of repeated \`scroll\` actions when you know what you're hunting for.`,
913
+ `\`{ "action": "find_text", "text": "Annual report 2024" }\``,
588
914
  ``,
589
- `### back`,
590
- `Go back to the previous page.`,
591
- `\`{ "action": "back" }\``,
915
+ `### send_keys`,
916
+ `Send arbitrary keys / shortcuts. Single key, combo (with \`+\`), or space-separated sequence.`,
917
+ `\`{ "action": "send_keys", "keys": "Tab Tab Enter" }\` \xB7 \`{ "action": "send_keys", "keys": "Control+l" }\` \xB7 \`{ "action": "send_keys", "keys": "Escape" }\``,
592
918
  ``,
593
- `### wait`,
594
- `Wait for the page to load or for a timed event. Use sparingly.`,
595
- `\`{ "action": "wait", "ms": <milliseconds> }\``,
919
+ `### dropdown_options / select_dropdown`,
920
+ `Inspect or set a native \`<select>\` by index. **Do not click into native selects** \u2014 the OS overlay is not part of the DOM.`,
921
+ `\`{ "action": "dropdown_options", "index": <n> }\` returns the option list in the next observation.`,
922
+ `\`{ "action": "select_dropdown", "index": <n>, "text": "United States" }\` picks the matching option.`,
596
923
  ``,
597
- `### done`,
598
- `The task is complete. Provide a summary of what was accomplished.`,
599
- `\`{ "action": "done", "result": "<summary of what was accomplished>" }\``,
924
+ `### upload_file`,
925
+ `Set a file on an \`<input type="file">\` by index. \`path\` must be a path the runtime can read; the agent does NOT have a filesystem of its own.`,
926
+ `\`{ "action": "upload_file", "index": <n>, "path": "/abs/path/to/file.pdf" }\``,
600
927
  ``,
601
- `### fail`,
602
- `The task cannot be completed. Explain why.`,
603
- `\`{ "action": "fail", "reason": "<why the task failed>" }\``,
928
+ `### navigate / back`,
929
+ `\`{ "action": "navigate", "url": "<full URL>" }\` \xB7 \`{ "action": "back" }\``,
604
930
  ``,
605
- `## Rules`,
606
- `1. ALWAYS look at the screenshot carefully before deciding your action.`,
607
- `2. Provide coordinates that target the CENTER of the element you want to interact with.`,
608
- `3. When an "Interactive elements on page" list is provided, identify the target element in that list by its visible label (button text, aria-label, placeholder, link href) and use ITS coordinates \u2014 they are the ground truth. Only fall back to visual estimation when no matching element is listed.`,
609
- `4. For text inputs: click the input field first (using "type" with x/y), then the text will be typed.`,
610
- `5. After typing in a search box, you often need to press Enter \u2014 use type with text "\\n" or click the search/submit button.`,
611
- `6. If your previous click landed on the wrong element (e.g. page didn't change or a different element became focused), the target was probably misaligned \u2014 re-read the elements list and pick a different coordinate; do NOT repeat the same click.`,
612
- `7. If a page is loading or blank, use "wait" with a short delay and try again.`,
613
- `8. If you see a cookie banner, consent dialog, or popup, dismiss it FIRST before proceeding with the task \u2014 these often intercept clicks on elements behind them.`,
614
- `9. NEVER hallucinate content. Only report what you can actually see on the screen.`,
615
- `10. When the task is fully complete, use "done" immediately with a comprehensive result.`,
616
- `11. If after several attempts you cannot complete the task, use "fail" with a clear reason.`,
931
+ `### wait / screenshot`,
932
+ `\`{ "action": "wait", "ms": <ms \u2264 10000> }\` \xB7 \`{ "action": "screenshot" }\` (request a fresh image on the next step \u2014 only useful when vision mode is "auto").`,
617
933
  ``,
618
- `## Response Format`,
619
- `Respond with ONLY a valid JSON object. No markdown, no explanation, just the JSON action.`
620
- ];
621
- if (credentialKeys && credentialKeys.length > 0) {
934
+ `### extract`,
935
+ `Extract structured information from the current page using a secondary LLM. Cheaper than reasoning over multiple screenshots yourself when you just need facts/text. The extracted result is returned in the next observation as \`Last extract result:\`.`,
936
+ `\`{ "action": "extract", "query": "List the top 5 result titles and their prices", "extractLinks": false }\``
937
+ );
938
+ if (allowEvaluate) {
622
939
  lines.push(
623
940
  ``,
624
- `## Secure Credentials`,
625
- `The following credential placeholders are available for use in "type" actions:`,
626
- ...credentialKeys.map((k) => `- \`{{${k}}}\``),
627
- ``,
628
- `When you need to fill in a login form or any field requiring these credentials,`,
629
- `use the EXACT placeholder (e.g. \`{{email}}\`) as the "text" value in a type action.`,
630
- `The system will securely replace them with real values at execution time.`,
631
- `NEVER guess, invent, or ask the user for the actual credential values.`,
632
- `NEVER include real credential values in "done" or "fail" results.`
941
+ `### evaluate (JS escape hatch)`,
942
+ `Run arbitrary JavaScript in the page context. Use this ONLY when no other action fits \u2014 e.g. shadow DOM access, complex selectors, custom widget state. The return value is stringified and returned in the next observation.`,
943
+ `\`{ "action": "evaluate", "code": "return document.title" }\``
633
944
  );
634
945
  }
946
+ if (tools.length > 0) {
947
+ lines.push(``, `### tool \u2014 invoke a custom tool`);
948
+ lines.push(
949
+ `In addition to browser actions, you have access to the following custom tools. Invoke them with \`{ "action": "tool", "name": "<tool>", "args": { ... } }\`. The result is returned in the next observation.`
950
+ );
951
+ for (const t of tools) {
952
+ lines.push(`- **${t.name}** \u2014 ${t.description}`);
953
+ }
954
+ }
955
+ lines.push(
956
+ ``,
957
+ `### done / fail`,
958
+ `\`{ "action": "done", "result": "<comprehensive summary of what was accomplished and any data the user asked for>" }\` \u2014 use when the task is complete.`,
959
+ `\`{ "action": "fail", "reason": "<why>" }\` \u2014 use when the task cannot be completed even after several attempts.`,
960
+ ``,
961
+ `## Rules`,
962
+ `1. ALWAYS check the DOM snapshot first. If the target element is listed, use its \`index\`. Coordinates are a fallback.`,
963
+ `2. For text-bearing targets, ALSO include the visible label in quotes in \`description\` \u2014 the runtime will use it as a third-tier fallback if the locator fails.`,
964
+ `3. Dismiss cookie banners, consent dialogs, and modal popups FIRST \u2014 they intercept clicks on elements behind them.`,
965
+ `4. If a previous action clearly hit the wrong thing, do NOT repeat it. Pick a different element or a different action.`,
966
+ `5. Batch independent actions when safe (e.g. filling 3 form fields). Don't batch when an action navigates or substantially changes the DOM.`,
967
+ `6. NEVER hallucinate. Only report data you can actually see in the snapshot, screenshot, or an extract result.`,
968
+ `7. When the task is fully complete, return \`done\` IMMEDIATELY with a thorough result \u2014 don't add extra confirmation steps.`,
969
+ `8. If after several attempts you cannot make progress, return \`fail\` with a clear reason.`
970
+ );
971
+ appendCredentials(lines, credentialKeys);
635
972
  if (extraInstructions) {
636
973
  lines.push(``, `## Additional Instructions`, extraInstructions);
637
974
  }
975
+ lines.push(``, `## Response Format`, responseFormatLines(maxActions).join("\n"));
638
976
  return lines.join("\n");
639
977
  }
640
- function buildUserMessage(task, pageUrl, pageTitle, stepIndex, actionHistory, domSnapshot) {
978
+ function appendCredentials(lines, credentialKeys) {
979
+ if (!credentialKeys || credentialKeys.length === 0) return;
980
+ lines.push(
981
+ ``,
982
+ `## Secure Credentials`,
983
+ `The following credential placeholders are available for use in "type" actions:`,
984
+ ...credentialKeys.map((k) => `- \`{{${k}}}\``),
985
+ ``,
986
+ `When you need to fill in a login form or any field requiring these credentials,`,
987
+ `use the EXACT placeholder (e.g. \`{{email}}\`) as the "text" value in a type action.`,
988
+ `The system will securely replace them with real values at execution time.`,
989
+ `NEVER guess, invent, or ask the user for the actual credential values.`,
990
+ `NEVER include real credential values in "done" or "fail" results.`
991
+ );
992
+ }
993
+ function responseFormatLines(maxActions) {
994
+ return [
995
+ `Respond with ONLY valid JSON. No markdown, no commentary.`,
996
+ `Either a single action object, or an array of up to ${maxActions} action objects to execute in order.`
997
+ ];
998
+ }
999
+ function buildUserMessage(task, pageUrl, pageTitle, stepIndex, actionHistory, domSnapshot, lastExtract) {
641
1000
  const lines = [];
642
1001
  lines.push(`**Task:** ${task}`);
643
1002
  lines.push(`**Current URL:** ${pageUrl}`);
@@ -645,9 +1004,14 @@ function buildUserMessage(task, pageUrl, pageTitle, stepIndex, actionHistory, do
645
1004
  lines.push(`**Step:** ${stepIndex + 1}`);
646
1005
  if (domSnapshot) {
647
1006
  lines.push(``);
648
- lines.push(`**Interactive elements on page (format: [centerX,centerY] role: "label"):**`);
1007
+ lines.push(`**Interactive elements (format: [idx] [cx,cy] role: "label"):**`);
649
1008
  lines.push(domSnapshot);
650
1009
  }
1010
+ if (lastExtract) {
1011
+ lines.push(``);
1012
+ lines.push(`**Last extract result:**`);
1013
+ lines.push(lastExtract);
1014
+ }
651
1015
  if (actionHistory.length > 0) {
652
1016
  lines.push(``);
653
1017
  lines.push(`**Previous actions:**`);
@@ -656,16 +1020,19 @@ function buildUserMessage(task, pageUrl, pageTitle, stepIndex, actionHistory, do
656
1020
  }
657
1021
  }
658
1022
  lines.push(``);
659
- lines.push(`Look at the screenshot and decide the next action to complete the task.`);
1023
+ lines.push(`Decide the next action(s) to complete the task.`);
660
1024
  return lines.join("\n");
661
1025
  }
662
1026
  function summarizeAction(action) {
663
1027
  switch (action.action) {
664
1028
  case "click":
665
- return `Clicked at (${action.x}, ${action.y}): ${action.description}`;
1029
+ if (typeof action.index === "number") return `Clicked [${action.index}]: ${action.description ?? ""}`.trim();
1030
+ return `Clicked at (${action.x}, ${action.y}): ${action.description ?? ""}`.trim();
666
1031
  case "type":
1032
+ if (typeof action.index === "number") return `Typed into [${action.index}]: "${action.text}"`;
667
1033
  return action.x != null ? `Clicked (${action.x}, ${action.y}) and typed "${action.text}"` : `Typed "${action.text}"`;
668
1034
  case "scroll":
1035
+ if (typeof action.index === "number") return `Scrolled [${action.index}] into view`;
669
1036
  return `Scrolled ${action.direction}${action.amount ? ` ${action.amount}px` : ""}`;
670
1037
  case "navigate":
671
1038
  return `Navigated to ${action.url}`;
@@ -674,7 +1041,23 @@ function summarizeAction(action) {
674
1041
  case "wait":
675
1042
  return `Waited ${action.ms}ms`;
676
1043
  case "screenshot":
677
- return `Took an extra screenshot`;
1044
+ return `Requested a fresh screenshot`;
1045
+ case "send_keys":
1046
+ return `Sent keys: ${action.keys}`;
1047
+ case "find_text":
1048
+ return `Scrolled to text: "${action.text}"`;
1049
+ case "evaluate":
1050
+ return `Evaluated JS`;
1051
+ case "dropdown_options":
1052
+ return `Read dropdown options at [${action.index}]`;
1053
+ case "select_dropdown":
1054
+ return `Selected "${action.text}" in dropdown [${action.index}]`;
1055
+ case "upload_file":
1056
+ return `Uploaded file to [${action.index}]: ${action.path}`;
1057
+ case "extract":
1058
+ return `Extracted: "${action.query}"`;
1059
+ case "tool":
1060
+ return `Called tool "${action.name}"`;
678
1061
  case "done":
679
1062
  return `Done: ${action.result}`;
680
1063
  case "fail":
@@ -689,19 +1072,32 @@ var BrowserAgent = class {
689
1072
  name;
690
1073
  eventBus;
691
1074
  model;
1075
+ pageExtractionLLM;
692
1076
  instructions;
1077
+ extendSystemMessage;
1078
+ overrideSystemMessage;
693
1079
  maxSteps;
1080
+ maxFailures;
1081
+ maxActionsPerStep;
1082
+ initialActions;
1083
+ useVision;
1084
+ directlyOpenUrl;
694
1085
  headless;
695
1086
  viewport;
696
1087
  defaultStartUrl;
697
1088
  waitAfterAction;
698
1089
  maxRepeats;
699
1090
  useDOM;
1091
+ allowEvaluate;
1092
+ allowedDomains;
1093
+ prohibitedDomains;
700
1094
  storageState;
1095
+ cdpUrl;
701
1096
  recordVideo;
702
1097
  credentials;
703
1098
  stealth;
704
1099
  humanize;
1100
+ tools;
705
1101
  costTracker;
706
1102
  memoryManager = null;
707
1103
  logger;
@@ -712,19 +1108,32 @@ var BrowserAgent = class {
712
1108
  constructor(config) {
713
1109
  this.name = config.name;
714
1110
  this.model = config.model;
1111
+ this.pageExtractionLLM = config.pageExtractionLLM ?? null;
715
1112
  this.instructions = config.instructions;
1113
+ this.extendSystemMessage = config.extendSystemMessage;
1114
+ this.overrideSystemMessage = config.overrideSystemMessage;
716
1115
  this.maxSteps = config.maxSteps ?? 30;
1116
+ this.maxFailures = config.maxFailures ?? 3;
1117
+ this.maxActionsPerStep = Math.max(1, config.maxActionsPerStep ?? 3);
1118
+ this.initialActions = config.initialActions ?? [];
1119
+ this.useVision = config.useVision ?? "auto";
1120
+ this.directlyOpenUrl = config.directlyOpenUrl ?? true;
717
1121
  this.headless = config.headless ?? true;
718
1122
  this.viewport = config.viewport ?? { width: 1280, height: 720 };
719
1123
  this.defaultStartUrl = config.startUrl;
720
1124
  this.waitAfterAction = config.waitAfterAction ?? 1500;
721
1125
  this.maxRepeats = config.maxRepeats ?? 3;
722
1126
  this.useDOM = config.useDOM ?? true;
1127
+ this.allowEvaluate = config.allowEvaluate ?? false;
1128
+ this.allowedDomains = config.allowedDomains;
1129
+ this.prohibitedDomains = config.prohibitedDomains;
723
1130
  this.storageState = config.storageState;
1131
+ this.cdpUrl = config.cdpUrl;
724
1132
  this.recordVideo = config.recordVideo;
725
1133
  this.credentials = config.credentials;
726
1134
  this.stealth = config.stealth;
727
1135
  this.humanize = config.humanize;
1136
+ this.tools = config.tools ?? [];
728
1137
  this.costTracker = config.costTracker ?? null;
729
1138
  this.eventBus = config.eventBus ?? new EventBus();
730
1139
  this.logger = new Logger({
@@ -737,13 +1146,20 @@ var BrowserAgent = class {
737
1146
  }
738
1147
  async run(task, opts) {
739
1148
  const startTime = Date.now();
740
- const startUrl = opts?.startUrl ?? this.defaultStartUrl;
1149
+ const maxSteps = opts?.maxSteps ?? this.maxSteps;
741
1150
  const sessionId = opts?.sessionId ?? `browser_${Date.now()}`;
742
1151
  const userId = opts?.userId;
743
1152
  const browser = new BrowserProvider();
744
1153
  const steps = [];
745
1154
  const actionHistory = [];
746
- let extraInstructions = this.instructions ?? "";
1155
+ const extractedContent = [];
1156
+ let lastExtractResult;
1157
+ let failureCount = 0;
1158
+ let lastActionKey = "";
1159
+ let repeatCount = 0;
1160
+ let lastActionWasScreenshot = false;
1161
+ const baseExtra = [this.instructions, this.extendSystemMessage].filter(Boolean).join("\n\n");
1162
+ let extraInstructions = baseExtra;
747
1163
  if (this.memoryManager) {
748
1164
  await this.memoryManager.ensureReady();
749
1165
  const memoryContext = await this.memoryManager.buildContext(sessionId, userId, task, this.name);
@@ -754,14 +1170,21 @@ ${memoryContext}` : memoryContext;
754
1170
  }
755
1171
  }
756
1172
  const credentialKeys = this.credentials?.keys();
757
- const systemPrompt = buildSystemPrompt(this.viewport, extraInstructions || void 0, credentialKeys);
758
- let lastActionKey = "";
759
- let repeatCount = 0;
1173
+ const systemPrompt = buildSystemPrompt(this.viewport, extraInstructions || void 0, credentialKeys, {
1174
+ overrideSystemMessage: this.overrideSystemMessage,
1175
+ maxActionsPerStep: this.maxActionsPerStep,
1176
+ allowEvaluate: this.allowEvaluate,
1177
+ tools: this.tools,
1178
+ useVision: this.useVision,
1179
+ useDOM: this.useDOM
1180
+ });
760
1181
  try {
761
1182
  this.logger.info("Launching browser", {
762
1183
  headless: this.headless,
763
1184
  viewport: this.viewport,
764
1185
  useDOM: this.useDOM,
1186
+ useVision: this.useVision,
1187
+ cdpUrl: this.cdpUrl ?? void 0,
765
1188
  recordVideo: !!this.recordVideo,
766
1189
  stealth: !!this.stealth,
767
1190
  humanize: !!this.humanize
@@ -772,42 +1195,72 @@ ${memoryContext}` : memoryContext;
772
1195
  storageState: this.storageState,
773
1196
  recordVideo: this.recordVideo,
774
1197
  stealth: this.stealth,
775
- humanize: this.humanize
1198
+ humanize: this.humanize,
1199
+ cdpUrl: this.cdpUrl
776
1200
  });
1201
+ const startUrl = opts?.startUrl ?? this.defaultStartUrl ?? this.detectUrlInTask(task);
777
1202
  if (startUrl) {
778
1203
  this.logger.info("Navigating to start URL", { url: startUrl });
1204
+ this.assertDomainAllowed(startUrl);
779
1205
  await browser.navigate(startUrl);
780
1206
  }
781
- for (let step = 0; step < this.maxSteps; step++) {
782
- const screenshot = await browser.screenshot();
1207
+ for (const ia of this.initialActions) {
1208
+ try {
1209
+ await this.executeAction(browser, ia, actionHistory, extractedContent);
1210
+ } catch (e) {
1211
+ this.logger.warn("initialAction failed", { action: ia, error: e?.message });
1212
+ }
1213
+ await this.sleep(this.waitAfterAction);
1214
+ }
1215
+ for (let step = 0; step < maxSteps; step++) {
783
1216
  const pageInfo = await browser.getPageInfo();
784
1217
  let domSnapshot;
785
1218
  if (this.useDOM) {
786
- domSnapshot = await browser.extractDOM();
1219
+ const dom = await browser.extractDOM();
1220
+ domSnapshot = dom.text;
1221
+ }
1222
+ const wantVision = this.shouldCaptureVision(step, lastActionWasScreenshot);
1223
+ const screenshot = wantVision ? await browser.screenshot() : Buffer.alloc(0);
1224
+ if (wantVision) this.eventBus.emit("browser.screenshot", { data: screenshot });
1225
+ const userText = buildUserMessage(
1226
+ task,
1227
+ pageInfo.url,
1228
+ pageInfo.title,
1229
+ step,
1230
+ actionHistory,
1231
+ domSnapshot,
1232
+ lastExtractResult
1233
+ );
1234
+ const content = [];
1235
+ if (wantVision) {
1236
+ content.push({ type: "image", data: screenshot.toString("base64"), mimeType: "image/png" });
787
1237
  }
788
- this.eventBus.emit("browser.screenshot", { data: screenshot });
789
- const userText = buildUserMessage(task, pageInfo.url, pageInfo.title, step, actionHistory, domSnapshot);
1238
+ content.push({ type: "text", text: userText });
790
1239
  const messages = [
791
1240
  { role: "system", content: systemPrompt },
792
- {
793
- role: "user",
794
- content: [
795
- {
796
- type: "image",
797
- data: screenshot.toString("base64"),
798
- mimeType: "image/png"
799
- },
800
- { type: "text", text: userText }
801
- ]
802
- }
1241
+ { role: "user", content }
803
1242
  ];
804
- this.logger.debug("Sending screenshot to vision model", { step, url: pageInfo.url });
805
- const response = await this.model.generate(messages, {
806
- temperature: 0.1,
807
- maxTokens: 1024,
808
- apiKey: opts?.apiKey,
809
- responseFormat: "json"
810
- });
1243
+ this.logger.debug("Calling model", { step, url: pageInfo.url, vision: wantVision });
1244
+ let response;
1245
+ try {
1246
+ response = await this.model.generate(messages, {
1247
+ temperature: 0.1,
1248
+ maxTokens: 1024,
1249
+ apiKey: opts?.apiKey,
1250
+ responseFormat: "json"
1251
+ });
1252
+ } catch (e) {
1253
+ failureCount++;
1254
+ this.logger.warn("Model call failed", { error: e?.message, failureCount });
1255
+ if (failureCount > this.maxFailures) {
1256
+ return await this.finalize(browser, steps, startTime, opts, extractedContent, {
1257
+ result: `Model call failed ${failureCount} times: ${e?.message}`,
1258
+ success: false
1259
+ });
1260
+ }
1261
+ actionHistory.push(`(model call failed: ${e?.message ?? "unknown"} \u2014 retrying)`);
1262
+ continue;
1263
+ }
811
1264
  if (this.costTracker && response.usage) {
812
1265
  this.costTracker.track({
813
1266
  runId: sessionId,
@@ -819,78 +1272,119 @@ ${memoryContext}` : memoryContext;
819
1272
  });
820
1273
  }
821
1274
  const raw = typeof response.message.content === "string" ? response.message.content : "";
822
- let action;
1275
+ let parsed;
823
1276
  try {
824
- action = JSON.parse(raw);
1277
+ parsed = JSON.parse(raw);
825
1278
  } catch {
826
- this.logger.warn("Failed to parse model response as JSON, retrying", { raw });
1279
+ failureCount++;
1280
+ this.logger.warn("Failed to parse model response as JSON", { raw, failureCount });
1281
+ if (failureCount > this.maxFailures) {
1282
+ return await this.finalize(browser, steps, startTime, opts, extractedContent, {
1283
+ result: `Model returned invalid JSON ${failureCount} times. Last response: ${raw.slice(0, 200)}`,
1284
+ success: false
1285
+ });
1286
+ }
827
1287
  actionHistory.push("(invalid JSON response \u2014 retrying)");
828
1288
  continue;
829
1289
  }
830
- const step_ = {
831
- index: step,
832
- action,
833
- screenshot,
834
- pageUrl: pageInfo.url,
835
- pageTitle: pageInfo.title,
836
- timestamp: /* @__PURE__ */ new Date(),
837
- dom: domSnapshot
838
- };
839
- steps.push(step_);
840
- let summary = summarizeAction(action);
841
- if (this.credentials) {
842
- summary = this.credentials.mask(summary);
843
- }
844
- const actionKey = JSON.stringify(action);
845
- if (actionKey === lastActionKey) {
846
- repeatCount++;
847
- } else {
848
- lastActionKey = actionKey;
849
- repeatCount = 1;
850
- }
851
- if (repeatCount > this.maxRepeats && action.action !== "done" && action.action !== "fail") {
852
- this.logger.warn("Stuck in a loop \u2014 same action repeated", {
853
- action: action.action,
854
- repeats: repeatCount,
855
- maxRepeats: this.maxRepeats
856
- });
857
- actionHistory.push(
858
- `\u26A0 LOOP DETECTED: "${summary}" repeated ${repeatCount} times. The agent was stuck and auto-stopped. Try a different approach or a different startUrl.`
859
- );
860
- return await this.finalize(browser, steps, startTime, opts, {
861
- result: `Stuck in a loop: "${summary}" was repeated ${repeatCount} times. The page may have a popup, consent banner, or unexpected state blocking progress.`,
862
- success: false
1290
+ const actions = (Array.isArray(parsed) ? parsed : [parsed]).slice(
1291
+ 0,
1292
+ this.maxActionsPerStep
1293
+ );
1294
+ let didTerminate = null;
1295
+ let didNavigate = false;
1296
+ lastActionWasScreenshot = false;
1297
+ for (let ai = 0; ai < actions.length; ai++) {
1298
+ const action = actions[ai];
1299
+ let summary = summarizeAction(action);
1300
+ if (this.credentials) summary = this.credentials.mask(summary);
1301
+ const actionKey = JSON.stringify(action);
1302
+ if (actionKey === lastActionKey) {
1303
+ repeatCount++;
1304
+ } else {
1305
+ lastActionKey = actionKey;
1306
+ repeatCount = 1;
1307
+ }
1308
+ if (repeatCount > this.maxRepeats && action.action !== "done" && action.action !== "fail") {
1309
+ this.logger.warn("Stuck in a loop \u2014 same action repeated", {
1310
+ action: action.action,
1311
+ repeats: repeatCount,
1312
+ maxRepeats: this.maxRepeats
1313
+ });
1314
+ actionHistory.push(
1315
+ `\u26A0 LOOP DETECTED: "${summary}" repeated ${repeatCount} times. Auto-stopping.`
1316
+ );
1317
+ didTerminate = {
1318
+ result: `Stuck in a loop: "${summary}" was repeated ${repeatCount} times. The page may have a popup, consent banner, or unexpected state blocking progress.`,
1319
+ success: false
1320
+ };
1321
+ break;
1322
+ }
1323
+ actionHistory.push(summary);
1324
+ this.logger.info(`Step ${step + 1}.${ai + 1}: ${summary}`);
1325
+ this.eventBus.emit("browser.action", { action });
1326
+ if (action.action === "done") {
1327
+ const result = this.credentials ? this.credentials.mask(action.result) : action.result;
1328
+ didTerminate = { result, success: true };
1329
+ break;
1330
+ }
1331
+ if (action.action === "fail") {
1332
+ const result = this.credentials ? this.credentials.mask(action.reason) : action.reason;
1333
+ didTerminate = { result, success: false };
1334
+ break;
1335
+ }
1336
+ let stepOk = true;
1337
+ let stepOutput;
1338
+ try {
1339
+ const exec = await this.executeAction(browser, action, actionHistory, extractedContent);
1340
+ stepOutput = exec?.output;
1341
+ if (exec?.didNavigate) didNavigate = true;
1342
+ if (action.action === "extract" && exec?.output) {
1343
+ lastExtractResult = exec.output;
1344
+ }
1345
+ if (action.action === "screenshot") {
1346
+ lastActionWasScreenshot = true;
1347
+ }
1348
+ } catch (e) {
1349
+ stepOk = false;
1350
+ failureCount++;
1351
+ this.logger.warn("Action failed", { action: action.action, error: e?.message, failureCount });
1352
+ actionHistory.push(`(error: ${e?.message ?? "unknown"})`);
1353
+ if (failureCount > this.maxFailures) {
1354
+ didTerminate = {
1355
+ result: `Action ${action.action} failed ${failureCount} times. Last error: ${e?.message}`,
1356
+ success: false
1357
+ };
1358
+ break;
1359
+ }
1360
+ }
1361
+ steps.push({
1362
+ index: steps.length,
1363
+ action,
1364
+ screenshot,
1365
+ pageUrl: pageInfo.url,
1366
+ pageTitle: pageInfo.title,
1367
+ timestamp: /* @__PURE__ */ new Date(),
1368
+ dom: domSnapshot,
1369
+ output: stepOutput,
1370
+ ok: stepOk
863
1371
  });
864
- }
865
- actionHistory.push(summary);
866
- this.logger.info(`Step ${step + 1}: ${summary}`);
867
- this.eventBus.emit("browser.action", { action });
868
- this.eventBus.emit("browser.step", {
869
- index: step,
870
- action,
871
- pageUrl: pageInfo.url,
872
- screenshot
873
- });
874
- if (action.action === "done") {
875
- const result = this.credentials ? this.credentials.mask(action.result) : action.result;
876
- return await this.finalize(browser, steps, startTime, opts, {
877
- result,
878
- success: true
1372
+ this.eventBus.emit("browser.step", {
1373
+ index: steps.length - 1,
1374
+ action,
1375
+ pageUrl: pageInfo.url,
1376
+ screenshot
879
1377
  });
1378
+ await this.sleep(this.waitAfterAction);
1379
+ if (didNavigate) break;
880
1380
  }
881
- if (action.action === "fail") {
882
- const result = this.credentials ? this.credentials.mask(action.reason) : action.reason;
883
- return await this.finalize(browser, steps, startTime, opts, {
884
- result,
885
- success: false
886
- });
1381
+ if (didTerminate) {
1382
+ return await this.finalize(browser, steps, startTime, opts, extractedContent, didTerminate);
887
1383
  }
888
- await this.executeAction(browser, action);
889
- await this.sleep(this.waitAfterAction);
890
1384
  }
891
- this.logger.warn("Max steps reached without completing task", { maxSteps: this.maxSteps });
892
- return await this.finalize(browser, steps, startTime, opts, {
893
- result: `Task not completed within ${this.maxSteps} steps. Last actions: ${actionHistory.slice(-3).join("; ")}`,
1385
+ this.logger.warn("Max steps reached without completing task", { maxSteps });
1386
+ return await this.finalize(browser, steps, startTime, opts, extractedContent, {
1387
+ result: `Task not completed within ${maxSteps} steps. Last actions: ${actionHistory.slice(-3).join("; ")}`,
894
1388
  success: false
895
1389
  });
896
1390
  } catch (error) {
@@ -903,7 +1397,8 @@ ${memoryContext}` : memoryContext;
903
1397
  steps,
904
1398
  finalUrl: "",
905
1399
  finalScreenshot: Buffer.alloc(0),
906
- durationMs: Date.now() - startTime
1400
+ durationMs: Date.now() - startTime,
1401
+ extractedContent
907
1402
  };
908
1403
  }
909
1404
  }
@@ -926,9 +1421,43 @@ ${memoryContext}` : memoryContext;
926
1421
  };
927
1422
  }
928
1423
  // ── Private helpers ──────────────────────────────────────────────────
929
- async finalize(browser, steps, startTime, opts, outcome) {
930
- const finalScreenshot = await browser.screenshot();
931
- const finalInfo = await browser.getPageInfo();
1424
+ shouldCaptureVision(step, lastActionWasScreenshot) {
1425
+ if (this.useVision === false) return false;
1426
+ if (this.useVision === true) return true;
1427
+ if (step === 0) return true;
1428
+ if (!this.useDOM) return true;
1429
+ return lastActionWasScreenshot;
1430
+ }
1431
+ detectUrlInTask(task) {
1432
+ if (!this.directlyOpenUrl) return void 0;
1433
+ const match = task.match(/https?:\/\/[^\s)<>"']+/);
1434
+ return match ? match[0] : void 0;
1435
+ }
1436
+ assertDomainAllowed(url) {
1437
+ if (!this.allowedDomains?.length && !this.prohibitedDomains?.length) return;
1438
+ let host;
1439
+ try {
1440
+ host = new URL(url).hostname.toLowerCase();
1441
+ } catch {
1442
+ throw new Error(`Cannot parse URL for domain check: ${url}`);
1443
+ }
1444
+ if (this.allowedDomains?.length && !this.allowedDomains.some((p) => matchDomain(host, p))) {
1445
+ throw new Error(`Navigation blocked: ${host} is not in allowedDomains`);
1446
+ }
1447
+ if (this.prohibitedDomains?.length && this.prohibitedDomains.some((p) => matchDomain(host, p))) {
1448
+ throw new Error(`Navigation blocked: ${host} is in prohibitedDomains`);
1449
+ }
1450
+ }
1451
+ async finalize(browser, steps, startTime, opts, extractedContent, outcome) {
1452
+ let finalScreenshot;
1453
+ let finalInfo;
1454
+ try {
1455
+ finalScreenshot = await browser.screenshot();
1456
+ finalInfo = await browser.getPageInfo();
1457
+ } catch {
1458
+ finalScreenshot = Buffer.alloc(0);
1459
+ finalInfo = { url: "", title: "" };
1460
+ }
932
1461
  if (opts?.saveStorageState) {
933
1462
  try {
934
1463
  await browser.saveStorageState(opts.saveStorageState);
@@ -949,7 +1478,8 @@ ${memoryContext}` : memoryContext;
949
1478
  finalUrl: finalInfo.url,
950
1479
  finalScreenshot,
951
1480
  durationMs: Date.now() - startTime,
952
- videoPath
1481
+ videoPath,
1482
+ extractedContent
953
1483
  };
954
1484
  if (this.memoryManager) {
955
1485
  const sessionId = opts?.sessionId ?? `browser_${startTime}`;
@@ -973,54 +1503,137 @@ ${memoryContext}` : memoryContext;
973
1503
  });
974
1504
  return output;
975
1505
  }
976
- async executeAction(browser, action) {
977
- try {
978
- switch (action.action) {
979
- case "click": {
980
- const keyword = this.extractClickKeyword(action.description);
981
- let clicked = false;
982
- if (keyword) {
983
- clicked = await browser.clickByText(keyword);
984
- if (clicked) {
985
- this.logger.debug("Clicked by text", { keyword });
986
- }
987
- }
988
- if (!clicked) {
989
- await browser.click(action.x, action.y);
990
- }
991
- break;
1506
+ /**
1507
+ * Execute a single action. Returns `{ output?, didNavigate? }` for the
1508
+ * caller's state tracking. May throw — the loop handles failure-budget
1509
+ * accounting in that case.
1510
+ */
1511
+ async executeAction(browser, action, _actionHistory, extractedContent) {
1512
+ switch (action.action) {
1513
+ case "click": {
1514
+ if (typeof action.index === "number") {
1515
+ const ok = await browser.clickByIndex(action.index);
1516
+ if (ok) return {};
992
1517
  }
993
- case "type": {
994
- const resolvedText = this.credentials ? this.credentials.resolve(action.text) : action.text;
995
- if (action.x != null && action.y != null) {
996
- await browser.clickAndType(action.x, action.y, resolvedText);
997
- } else {
998
- await browser.type(resolvedText);
999
- }
1000
- if (resolvedText.includes("\n")) {
1001
- await browser.pressKey("Enter");
1002
- }
1003
- break;
1518
+ const keyword = this.extractClickKeyword(action.description);
1519
+ if (keyword && await browser.clickByText(keyword)) {
1520
+ this.logger.debug("Clicked by text", { keyword });
1521
+ return {};
1522
+ }
1523
+ if (typeof action.x === "number" && typeof action.y === "number") {
1524
+ await browser.click(action.x, action.y);
1525
+ return {};
1004
1526
  }
1005
- case "scroll":
1006
- await browser.scroll(action.direction, action.amount);
1007
- break;
1008
- case "navigate":
1009
- await browser.navigate(action.url);
1010
- break;
1011
- case "back":
1012
- await browser.back();
1013
- break;
1014
- case "wait":
1015
- await this.sleep(Math.min(action.ms, 1e4));
1016
- break;
1017
- case "screenshot":
1018
- break;
1019
- default:
1020
- this.logger.warn("Unknown action", { action });
1527
+ throw new Error(`click action has no resolvable target (index/text/coordinates)`);
1021
1528
  }
1022
- } catch (error) {
1023
- this.logger.warn("Action execution failed", { action: action.action, error: error.message });
1529
+ case "type": {
1530
+ const resolvedText = this.credentials ? this.credentials.resolve(action.text) : action.text;
1531
+ const submit = action.submit ?? resolvedText.includes("\n");
1532
+ const cleanText = submit ? resolvedText.replace(/\n+$/, "") : resolvedText;
1533
+ if (typeof action.index === "number") {
1534
+ const ok = await browser.inputByIndex(action.index, cleanText, {
1535
+ clear: action.clear,
1536
+ submit
1537
+ });
1538
+ if (ok) return {};
1539
+ }
1540
+ if (typeof action.x === "number" && typeof action.y === "number") {
1541
+ await browser.clickAndType(action.x, action.y, cleanText);
1542
+ if (submit) await browser.pressKey("Enter");
1543
+ return {};
1544
+ }
1545
+ await browser.type(cleanText);
1546
+ if (submit) await browser.pressKey("Enter");
1547
+ return {};
1548
+ }
1549
+ case "scroll": {
1550
+ if (typeof action.index === "number") {
1551
+ await browser.scrollIntoViewByIndex(action.index);
1552
+ return {};
1553
+ }
1554
+ await browser.scroll(action.direction, action.amount);
1555
+ return {};
1556
+ }
1557
+ case "navigate": {
1558
+ this.assertDomainAllowed(action.url);
1559
+ await browser.navigate(action.url);
1560
+ return { didNavigate: true };
1561
+ }
1562
+ case "back":
1563
+ await browser.back();
1564
+ return { didNavigate: true };
1565
+ case "wait":
1566
+ await this.sleep(Math.min(action.ms, 1e4));
1567
+ return {};
1568
+ case "screenshot":
1569
+ return {};
1570
+ case "send_keys":
1571
+ await browser.sendKeys(action.keys);
1572
+ return {};
1573
+ case "find_text":
1574
+ await browser.findText(action.text);
1575
+ return {};
1576
+ case "evaluate": {
1577
+ if (!this.allowEvaluate) {
1578
+ throw new Error("evaluate is disabled. Set allowEvaluate: true on BrowserAgent to enable.");
1579
+ }
1580
+ const out = await browser.evaluate(action.code);
1581
+ return { output: `evaluate \u2192 ${out}` };
1582
+ }
1583
+ case "dropdown_options": {
1584
+ const options = await browser.dropdownOptions(action.index);
1585
+ const formatted = options.map((o, i) => `${i + 1}. "${o.label}" (value="${o.value}")${o.selected ? " [selected]" : ""}`).join("\n");
1586
+ return { output: `Dropdown [${action.index}] options:
1587
+ ${formatted || "(no options found)"}` };
1588
+ }
1589
+ case "select_dropdown": {
1590
+ const ok = await browser.selectDropdown(action.index, action.text);
1591
+ if (!ok) throw new Error(`Could not select "${action.text}" on dropdown [${action.index}]`);
1592
+ return {};
1593
+ }
1594
+ case "upload_file": {
1595
+ const ok = await browser.uploadFileByIndex(action.index, action.path);
1596
+ if (!ok) throw new Error(`Could not upload "${action.path}" to [${action.index}]`);
1597
+ return {};
1598
+ }
1599
+ case "extract": {
1600
+ const pageText = await browser.pageText({ extractLinks: action.extractLinks });
1601
+ const model = this.pageExtractionLLM ?? this.model;
1602
+ const messages = [
1603
+ {
1604
+ role: "system",
1605
+ content: "You extract information from web pages. Use ONLY the page content provided. If the requested information is not present, say so explicitly. Be concise and structured (lists/tables) when appropriate."
1606
+ },
1607
+ {
1608
+ role: "user",
1609
+ content: `Query: ${action.query}
1610
+
1611
+ Page content:
1612
+ ${pageText}`
1613
+ }
1614
+ ];
1615
+ const response = await model.generate(messages, { temperature: 0, maxTokens: 2048 });
1616
+ const out = typeof response.message.content === "string" ? response.message.content : "";
1617
+ const masked = this.credentials ? this.credentials.mask(out) : out;
1618
+ extractedContent.push(masked);
1619
+ return { output: masked };
1620
+ }
1621
+ case "tool": {
1622
+ const tool = this.tools.find((t) => t.name === action.name);
1623
+ if (!tool) throw new Error(`Tool "${action.name}" is not registered on this BrowserAgent`);
1624
+ const ctx = new RunContext({
1625
+ sessionId: `browser_${this.name}_${Date.now()}`,
1626
+ eventBus: this.eventBus
1627
+ });
1628
+ const result = await tool.execute(action.args ?? {}, ctx);
1629
+ const out = typeof result === "string" ? result : JSON.stringify(result);
1630
+ return { output: `${action.name} \u2192 ${out}` };
1631
+ }
1632
+ // `done` and `fail` are intercepted by the run loop before we get
1633
+ // here. Listing them keeps the discriminated union exhaustive.
1634
+ case "done":
1635
+ case "fail":
1636
+ return {};
1024
1637
  }
1025
1638
  }
1026
1639
  sleep(ms) {
@@ -1028,16 +1641,9 @@ ${memoryContext}` : memoryContext;
1028
1641
  }
1029
1642
  /**
1030
1643
  * Parse a quoted target keyword from a click action's `description`.
1031
- *
1032
- * The system prompt asks the model to include the visible label of its
1033
- * click target in quotes (e.g. `"Click on 'Cheapest' tab"`). When a usable
1034
- * quoted phrase is present we return it so `executeAction` can try a
1035
- * deterministic Playwright text locator before falling back to
1036
- * coordinate clicking.
1037
- *
1038
1644
  * Returns `undefined` for generic / ambiguous labels (login buttons,
1039
- * close, OK, etc.) where a substring text match could trivially fire on
1040
- * the wrong element.
1645
+ * close, OK, etc.) where a substring text match could fire on the
1646
+ * wrong element.
1041
1647
  */
1042
1648
  extractClickKeyword(description) {
1043
1649
  if (!description) return void 0;
@@ -1073,6 +1679,18 @@ ${memoryContext}` : memoryContext;
1073
1679
  return keyword;
1074
1680
  }
1075
1681
  };
1682
+ function matchDomain(host, pattern) {
1683
+ const h = host.toLowerCase();
1684
+ let p = pattern.toLowerCase().replace(/^https?\*?:\/\//, "").replace(/^\/+/, "");
1685
+ if (p.includes("/")) p = p.split("/")[0];
1686
+ if (p === "*") return true;
1687
+ if (p === h) return true;
1688
+ if (p.startsWith("*.")) {
1689
+ const base = p.slice(2);
1690
+ return h === base || h.endsWith(`.${base}`);
1691
+ }
1692
+ return false;
1693
+ }
1076
1694
 
1077
1695
  // src/credential-vault.ts
1078
1696
  var CredentialVault = class {