@agentium/browser 2.0.7 → 2.1.0

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
@@ -112,7 +112,11 @@ function buildStealthContextOpts(config, viewport) {
112
112
  locale: config.locale ?? "en-US",
113
113
  timezoneId: config.timezone ?? "America/New_York",
114
114
  colorScheme: "light",
115
- deviceScaleFactor: 2,
115
+ // Default to 1 — matches the most common real-user setup and avoids
116
+ // visual zoom/stretch in headed mode on non-Retina displays (the host
117
+ // OS would downsample a 2× rendered surface). Users on Retina-only
118
+ // deployments can opt back into DPR=2 for sharper screenshots.
119
+ deviceScaleFactor: config.deviceScaleFactor ?? 1,
116
120
  hasTouch: false,
117
121
  javaScriptEnabled: true,
118
122
  ignoreHTTPSErrors: config.ignoreHTTPSErrors ?? false
@@ -146,6 +150,13 @@ var BrowserProvider = class {
146
150
  _viewport;
147
151
  _videoDir;
148
152
  _humanize;
153
+ /**
154
+ * Most recent DOM snapshot (one per `extractDOM` call). Indexed actions
155
+ * (`clickByIndex`, `inputByIndex`, …) resolve their `index` against this.
156
+ */
157
+ _lastDom = [];
158
+ /** True if we connected over CDP (don't tear down the browser on close). */
159
+ _attached = false;
149
160
  constructor() {
150
161
  this._viewport = { width: 1280, height: 720 };
151
162
  }
@@ -154,8 +165,6 @@ var BrowserProvider = class {
154
165
  const pw = await import("playwright");
155
166
  const chromium = pw.chromium;
156
167
  this._viewport = opts?.viewport ?? { width: 1280, height: 720 };
157
- const stealthEnabled = !!opts?.stealth;
158
- const stealthCfg = typeof opts?.stealth === "object" ? opts.stealth : {};
159
168
  if (opts?.humanize) {
160
169
  const h = typeof opts.humanize === "object" ? opts.humanize : {};
161
170
  this._humanize = {
@@ -165,6 +174,20 @@ var BrowserProvider = class {
165
174
  mouseMovement: h.mouseMovement ?? true
166
175
  };
167
176
  }
177
+ if (opts?.cdpUrl) {
178
+ this.browser = await chromium.connectOverCDP(opts.cdpUrl);
179
+ this._attached = true;
180
+ const contexts = this.browser.contexts();
181
+ this.context = contexts.length > 0 ? contexts[0] : await this.browser.newContext({ viewport: this._viewport });
182
+ const existingPages = this.context.pages();
183
+ this.page = existingPages.length > 0 ? existingPages[0] : await this.context.newPage();
184
+ this.tabCounter = 0;
185
+ this.activeTabId = "tab-0";
186
+ this.pages.set("tab-0", this.page);
187
+ return;
188
+ }
189
+ const stealthEnabled = !!opts?.stealth;
190
+ const stealthCfg = typeof opts?.stealth === "object" ? opts.stealth : {};
168
191
  const launchOpts = {
169
192
  headless: opts?.headless ?? true
170
193
  };
@@ -231,7 +254,11 @@ var BrowserProvider = class {
231
254
  get viewport() {
232
255
  return this._viewport;
233
256
  }
234
- // ── Interaction (with optional humanize) ─────────────────────────────
257
+ /** Most recent DOM snapshot. Each entry has a stable `index`. */
258
+ get lastDom() {
259
+ return this._lastDom;
260
+ }
261
+ // ── Coordinate-based Interaction ─────────────────────────────────────
235
262
  async click(x, y) {
236
263
  this.ensurePage();
237
264
  const [cx, cy] = this.clampToViewport(x, y);
@@ -261,6 +288,20 @@ var BrowserProvider = class {
261
288
  this.ensurePage();
262
289
  await this.page.keyboard.press(key);
263
290
  }
291
+ /**
292
+ * Send arbitrary keyboard keys / shortcuts. Accepts a single
293
+ * Playwright key spec (`"Enter"`, `"Control+l"`, `"Shift+ArrowDown"`)
294
+ * or a space-separated sequence (`"Tab Tab Enter"`).
295
+ */
296
+ async sendKeys(keys) {
297
+ this.ensurePage();
298
+ const tokens = keys.trim().split(/\s+/).filter(Boolean);
299
+ for (const token of tokens) {
300
+ await this.page.keyboard.press(token);
301
+ await this.sleep(this._humanize ? this.randInt(40, 120) : 30);
302
+ }
303
+ await this.humanPause();
304
+ }
264
305
  async scroll(direction, amount) {
265
306
  this.ensurePage();
266
307
  const base = amount ?? 400;
@@ -278,13 +319,262 @@ var BrowserProvider = class {
278
319
  }
279
320
  await this.humanPause();
280
321
  }
281
- // ── DOM Extraction ───────────────────────────────────────────────────
322
+ // ── Indexed Interaction (preferred path) ─────────────────────────────
323
+ /**
324
+ * Build a Playwright locator for a DOM-snapshot index. Each `extractDOM`
325
+ * call tags surviving elements with `data-bua-idx="<n>"`; we resolve by
326
+ * that attribute. Returns null if the index is unknown.
327
+ */
328
+ locatorForIndex(index) {
329
+ if (!this.page) return null;
330
+ if (!this._lastDom.find((e) => e.index === index)) return null;
331
+ return this.page.locator(`[data-bua-idx="${index}"]`).first();
332
+ }
333
+ /**
334
+ * Click an element by its DOM-snapshot index. The most reliable click
335
+ * path on dynamic pages — survives layout shifts and DPR oddities.
336
+ */
337
+ async clickByIndex(index, opts) {
338
+ this.ensurePage();
339
+ const loc = this.locatorForIndex(index);
340
+ if (!loc) return false;
341
+ try {
342
+ await loc.scrollIntoViewIfNeeded({ timeout: 1500 }).catch(() => {
343
+ });
344
+ await loc.click({ timeout: opts?.timeout ?? 5e3 });
345
+ await this.humanPause();
346
+ return true;
347
+ } catch {
348
+ return false;
349
+ }
350
+ }
351
+ /**
352
+ * Focus an indexed input, optionally clear it, and type. Returns false
353
+ * if the index couldn't be resolved or the input couldn't be focused.
354
+ */
355
+ async inputByIndex(index, text, opts) {
356
+ this.ensurePage();
357
+ const loc = this.locatorForIndex(index);
358
+ if (!loc) return false;
359
+ try {
360
+ const timeout = opts?.timeout ?? 5e3;
361
+ await loc.scrollIntoViewIfNeeded({ timeout: 1500 }).catch(() => {
362
+ });
363
+ if (opts?.clear !== false) {
364
+ await loc.fill("", { timeout });
365
+ }
366
+ const delay = this._humanize ? this.randInt(this._humanize.typingDelay[0], this._humanize.typingDelay[1]) : 30;
367
+ await loc.click({ timeout });
368
+ await loc.pressSequentially(text, { delay });
369
+ if (opts?.submit) {
370
+ await this.page.keyboard.press("Enter");
371
+ }
372
+ await this.humanPause();
373
+ return true;
374
+ } catch {
375
+ return false;
376
+ }
377
+ }
378
+ async uploadFileByIndex(index, path) {
379
+ this.ensurePage();
380
+ const loc = this.locatorForIndex(index);
381
+ if (!loc) return false;
382
+ try {
383
+ await loc.setInputFiles(path, { timeout: 5e3 });
384
+ await this.humanPause();
385
+ return true;
386
+ } catch {
387
+ return false;
388
+ }
389
+ }
390
+ /** Scroll the indexed element into view (no click). */
391
+ async scrollIntoViewByIndex(index) {
392
+ this.ensurePage();
393
+ const loc = this.locatorForIndex(index);
394
+ if (!loc) return false;
395
+ try {
396
+ await loc.scrollIntoViewIfNeeded({ timeout: 3e3 });
397
+ return true;
398
+ } catch {
399
+ return false;
400
+ }
401
+ }
402
+ // ── Text-based Interaction ───────────────────────────────────────────
403
+ /**
404
+ * Deterministic, DOM-based click using Playwright's text locator.
405
+ *
406
+ * Returns `true` if a matching, visible, clickable element was found and
407
+ * clicked within `timeout` ms; `false` otherwise (so the caller can fall
408
+ * back to coordinate clicking). Substring-matches by default — e.g.
409
+ * `clickByText("Cheapest")` matches "Cheapest · 23-28 days · $2,550".
410
+ */
411
+ async clickByText(keyword, opts) {
412
+ this.ensurePage();
413
+ const timeout = opts?.timeout ?? 3e3;
414
+ try {
415
+ const locator = this.page.locator(`text=${keyword}`).first();
416
+ await locator.click({ timeout });
417
+ await this.humanPause();
418
+ return true;
419
+ } catch {
420
+ return false;
421
+ }
422
+ }
423
+ /**
424
+ * Scroll the first occurrence of `text` into view. Returns false if no
425
+ * match was found within the timeout.
426
+ */
427
+ async findText(text, opts) {
428
+ this.ensurePage();
429
+ const timeout = opts?.timeout ?? 3e3;
430
+ try {
431
+ const locator = this.page.getByText(text).first();
432
+ await locator.scrollIntoViewIfNeeded({ timeout });
433
+ return true;
434
+ } catch {
435
+ return false;
436
+ }
437
+ }
438
+ // ── Dropdowns ────────────────────────────────────────────────────────
439
+ /**
440
+ * Read the options of a native `<select>` at the given DOM-snapshot
441
+ * index. Returns `[]` if the element is not a `<select>`.
442
+ */
443
+ async dropdownOptions(index) {
444
+ this.ensurePage();
445
+ const loc = this.locatorForIndex(index);
446
+ if (!loc) return [];
447
+ try {
448
+ return await loc.evaluate((el) => {
449
+ if (!el || el.tagName !== "SELECT") return [];
450
+ return Array.from(el.options).map((opt) => ({
451
+ value: opt.value,
452
+ label: (opt.label || opt.textContent || "").trim(),
453
+ selected: !!opt.selected
454
+ }));
455
+ });
456
+ } catch {
457
+ return [];
458
+ }
459
+ }
460
+ /**
461
+ * Select an option in a native `<select>` by its visible text or value.
462
+ * Returns false if the element isn't a `<select>` or no option matched.
463
+ */
464
+ async selectDropdown(index, text) {
465
+ this.ensurePage();
466
+ const loc = this.locatorForIndex(index);
467
+ if (!loc) return false;
468
+ try {
469
+ await loc.selectOption({ label: text }, { timeout: 3e3 });
470
+ await this.humanPause();
471
+ return true;
472
+ } catch {
473
+ try {
474
+ await loc.selectOption({ value: text }, { timeout: 3e3 });
475
+ await this.humanPause();
476
+ return true;
477
+ } catch {
478
+ return false;
479
+ }
480
+ }
481
+ }
482
+ // ── JS Evaluation ────────────────────────────────────────────────────
483
+ /**
484
+ * Run arbitrary JS in the page context. The caller is responsible for
485
+ * gating this behind a config flag — the BrowserAgent only routes the
486
+ * `evaluate` action here when `allowEvaluate: true`.
487
+ *
488
+ * The code is wrapped in `(async () => { ... })()` and the return value
489
+ * is coerced to a string for the model.
490
+ */
491
+ async evaluate(code) {
492
+ this.ensurePage();
493
+ try {
494
+ const result = await this.page.evaluate(
495
+ // eslint-disable-next-line @typescript-eslint/no-implied-eval
496
+ new Function("return (async () => { " + code + " })()")
497
+ );
498
+ if (result === void 0) return "undefined";
499
+ if (result === null) return "null";
500
+ if (typeof result === "string") return result;
501
+ try {
502
+ return JSON.stringify(result);
503
+ } catch {
504
+ return String(result);
505
+ }
506
+ } catch (e) {
507
+ throw new Error(`evaluate failed: ${e?.message ?? e}`);
508
+ }
509
+ }
510
+ // ── Page text (for `extract`) ────────────────────────────────────────
511
+ /**
512
+ * Returns a clean text representation of the visible page body, with
513
+ * optional link extraction. Used by the BrowserAgent's `extract` action
514
+ * — the text is passed to a (usually cheap) LLM with the user's query.
515
+ */
516
+ async pageText(opts) {
517
+ this.ensurePage();
518
+ const maxChars = opts?.maxChars ?? 2e4;
519
+ const extractLinks = !!opts?.extractLinks;
520
+ const raw = await this.page.evaluate((withLinks) => {
521
+ const doc = globalThis.document;
522
+ const SKIP = /* @__PURE__ */ new Set(["SCRIPT", "STYLE", "NOSCRIPT", "TEMPLATE", "SVG"]);
523
+ const win = globalThis.window;
524
+ const lines = [];
525
+ function visit(node) {
526
+ if (!node) return;
527
+ if (node.nodeType === 3) {
528
+ const t = (node.nodeValue || "").trim();
529
+ if (t) lines.push(t);
530
+ return;
531
+ }
532
+ if (node.nodeType !== 1) return;
533
+ const tag = node.tagName.toUpperCase();
534
+ if (SKIP.has(tag)) return;
535
+ const style = win.getComputedStyle(node);
536
+ if (style && (style.visibility === "hidden" || style.display === "none")) return;
537
+ if (tag === "A" && withLinks) {
538
+ const href = node.getAttribute("href") || "";
539
+ const txt = (node.innerText || node.textContent || "").trim();
540
+ if (txt && href) {
541
+ lines.push(`[${txt}](${href})`);
542
+ return;
543
+ }
544
+ }
545
+ for (const child of node.childNodes) visit(child);
546
+ }
547
+ visit(doc.body);
548
+ return lines.join("\n");
549
+ }, extractLinks);
550
+ const collapsed = raw.replace(/\n{3,}/g, "\n\n").trim();
551
+ return collapsed.length > maxChars ? `${collapsed.slice(0, maxChars)}
552
+ \u2026[truncated]` : collapsed;
553
+ }
554
+ // ── DOM Extraction (with index tagging) ──────────────────────────────
555
+ /**
556
+ * Snapshot the interactive elements visible in the viewport, tag each
557
+ * with a `data-bua-idx="<n>"` attribute (used by indexed actions), and
558
+ * return BOTH a human-readable string (for the model) and the
559
+ * structured list (for the runtime).
560
+ *
561
+ * Three properties matter for accuracy:
562
+ * - Hit-tested: each listed coordinate / index actually reaches the
563
+ * labeled element (overlays / occlusion skip the entry).
564
+ * - Visibility-filtered: invisible, zero-size, `pointer-events: none`,
565
+ * and near-zero-opacity elements are excluded.
566
+ * - `cursor: pointer` fallback pass: a second scan catches custom React
567
+ * widgets that have no semantic role/href/onclick but are clickable.
568
+ */
282
569
  async extractDOM(opts) {
283
570
  this.ensurePage();
284
571
  const max = opts?.maxElements ?? 120;
285
- const elements = await this.page.evaluate((limit) => {
572
+ const raw = await this.page.evaluate((limit) => {
286
573
  const doc = globalThis.document;
287
574
  const win = globalThis.window;
575
+ for (const stale of doc.querySelectorAll("[data-bua-idx]")) {
576
+ stale.removeAttribute("data-bua-idx");
577
+ }
288
578
  const selectors = [
289
579
  "a[href]",
290
580
  "button",
@@ -304,7 +594,24 @@ var BrowserProvider = class {
304
594
  "[contenteditable='true']",
305
595
  "[tabindex]:not([tabindex='-1'])"
306
596
  ];
307
- const all = Array.from(doc.querySelectorAll(selectors.join(",")));
597
+ const semantic = Array.from(doc.querySelectorAll(selectors.join(",")));
598
+ const POINTER_SCAN_CAP = 2e3;
599
+ const pointerCandidates = [];
600
+ const allEls = doc.querySelectorAll("*");
601
+ for (let i = 0; i < allEls.length && pointerCandidates.length < POINTER_SCAN_CAP; i++) {
602
+ const el = allEls[i];
603
+ if (!el) continue;
604
+ const tag = el.tagName.toLowerCase();
605
+ if (tag === "html" || tag === "body" || tag === "head" || tag === "script" || tag === "style") continue;
606
+ const rect = el.getBoundingClientRect();
607
+ if (rect.width < 4 || rect.height < 4) continue;
608
+ if (rect.bottom < 0 || rect.top > win.innerHeight) continue;
609
+ if (rect.right < 0 || rect.left > win.innerWidth) continue;
610
+ const style = win.getComputedStyle(el);
611
+ if (style.cursor !== "pointer") continue;
612
+ pointerCandidates.push(el);
613
+ }
614
+ const all = semantic.concat(pointerCandidates);
308
615
  const vh = win.innerHeight;
309
616
  const vw = win.innerWidth;
310
617
  const entries = [];
@@ -326,9 +633,17 @@ var BrowserProvider = class {
326
633
  if (hit && hit !== el && !el.contains(hit) && !hit.contains(el)) {
327
634
  continue;
328
635
  }
636
+ if (hit && el.contains(hit)) seen.add(hit);
329
637
  } catch {
330
638
  }
331
639
  seen.add(el);
640
+ entries.push({ el, cx, cy });
641
+ }
642
+ entries.sort((a, b) => a.cy === b.cy ? a.cx - b.cx : a.cy - b.cy);
643
+ const out = [];
644
+ let idx = 1;
645
+ for (const { el, cx, cy } of entries) {
646
+ if (out.length >= limit) break;
332
647
  const tag = el.tagName.toLowerCase();
333
648
  const role = el.getAttribute("role") || tag;
334
649
  const type = el.getAttribute("type") || "";
@@ -343,16 +658,29 @@ var BrowserProvider = class {
343
658
  let label = ariaLabel || text || placeholder || title || value || name;
344
659
  if (!label && href) label = href.slice(0, 60);
345
660
  if (!label) label = `(${tag}${type ? ` type=${type}` : ""})`;
346
- entries.push({
661
+ el.setAttribute("data-bua-idx", String(idx));
662
+ out.push({
663
+ index: idx,
347
664
  cx,
348
665
  cy,
349
- line: `[${cx},${cy}] ${role}${type ? `(${type})` : ""}: "${label}"`
666
+ role,
667
+ type: type || void 0,
668
+ label,
669
+ tag,
670
+ isInput: tag === "input" || tag === "textarea" || el.isContentEditable === true,
671
+ isSelect: tag === "select",
672
+ isFile: tag === "input" && type === "file"
350
673
  });
674
+ idx++;
351
675
  }
352
- entries.sort((a, b) => a.cy === b.cy ? a.cx - b.cx : a.cy - b.cy);
353
- return entries.slice(0, limit).map((e) => e.line).join("\n");
676
+ return out;
354
677
  }, max);
355
- return elements;
678
+ this._lastDom = raw;
679
+ const lines = raw.map((e) => {
680
+ const typeSuffix = e.type ? `(${e.type})` : "";
681
+ return `[${e.index}] [${e.cx},${e.cy}] ${e.role}${typeSuffix}: "${e.label}"`;
682
+ });
683
+ return { text: lines.join("\n"), elements: raw };
356
684
  }
357
685
  // ── Page Info ────────────────────────────────────────────────────────
358
686
  async getPageInfo() {
@@ -437,12 +765,18 @@ var BrowserProvider = class {
437
765
  // ── Cleanup ──────────────────────────────────────────────────────────
438
766
  async close() {
439
767
  try {
440
- if (this.context) await this.context.close();
768
+ if (this.context && !this._attached) await this.context.close();
441
769
  } catch (err) {
442
770
  console.warn("[agentium/browser] Error closing context:", err instanceof Error ? err.message : err);
443
771
  }
444
772
  try {
445
- if (this.browser) await this.browser.close();
773
+ if (this.browser && !this._attached) await this.browser.close();
774
+ else if (this.browser && this._attached) {
775
+ try {
776
+ await this.browser.close();
777
+ } catch {
778
+ }
779
+ }
446
780
  } catch (err) {
447
781
  console.warn("[agentium/browser] Error closing browser:", err instanceof Error ? err.message : err);
448
782
  }
@@ -450,6 +784,7 @@ var BrowserProvider = class {
450
784
  this.context = null;
451
785
  this.browser = null;
452
786
  this.pages.clear();
787
+ this._lastDom = [];
453
788
  }
454
789
  // ── Private: Humanize helpers ────────────────────────────────────────
455
790
  /** Add small random offset to coordinates to avoid pixel-perfect bot patterns. */
@@ -471,8 +806,7 @@ var BrowserProvider = class {
471
806
  return [cx, cy];
472
807
  }
473
808
  /**
474
- * Simulate human mouse movement using Bézier-like interpolation.
475
- * Moves from the current mouse position to the target in small steps.
809
+ * Simulate human mouse movement using smoothstep interpolation.
476
810
  */
477
811
  async humanMouseMove(targetX, targetY) {
478
812
  const steps = this.randInt(5, 12);
@@ -509,89 +843,144 @@ var BrowserProvider = class {
509
843
  };
510
844
 
511
845
  // src/prompts.ts
512
- function buildSystemPrompt(viewport, extraInstructions, credentialKeys) {
513
- const lines = [
514
- `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.`,
846
+ function buildSystemPrompt(viewport, extraInstructions, credentialKeys, options) {
847
+ const maxActions = options?.maxActionsPerStep ?? 3;
848
+ const allowEvaluate = !!options?.allowEvaluate;
849
+ const tools = options?.tools ?? [];
850
+ const useVision = options?.useVision ?? "auto";
851
+ if (options?.overrideSystemMessage) {
852
+ const lines2 = [options.overrideSystemMessage];
853
+ appendCredentials(lines2, credentialKeys);
854
+ lines2.push("", "## Response Format", responseFormatLines(maxActions).join("\n"));
855
+ return lines2.join("\n");
856
+ }
857
+ const lines = [];
858
+ lines.push(
859
+ `You are a browser automation agent. Your job is to complete the user's task by interacting with a real web page.`,
860
+ ``,
861
+ `## What you receive each step`,
862
+ `- The current URL and page title`,
863
+ `- 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.`
864
+ );
865
+ if (useVision !== false) {
866
+ lines.push(
867
+ `- 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.`
868
+ );
869
+ } else {
870
+ lines.push(
871
+ `- (Vision is disabled \u2014 no screenshot. Rely entirely on the DOM snapshot and page text.)`
872
+ );
873
+ }
874
+ lines.push(
875
+ `- A short list of your previous actions (so you don't repeat yourself).`,
515
876
  ``,
516
- `## Viewport & Coordinate System`,
517
- `- The browser viewport is ${viewport.width}\xD7${viewport.height} CSS pixels.`,
518
- `- The screenshot you are given has EXACTLY the same dimensions: ${viewport.width}\xD7${viewport.height} pixels (top-left = 0,0).`,
519
- `- ALL coordinates you return MUST be within 0..${viewport.width - 1} for x and 0..${viewport.height - 1} for y.`,
520
- `- Coordinates outside this range will be rejected \u2014 your click will go nowhere.`,
521
- `- 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
+ `## Coordinate System`,
878
+ `The browser viewport is ${viewport.width}\xD7${viewport.height} CSS pixels, top-left origin. Coordinates outside that range are rejected.`,
522
879
  ``,
523
880
  `## Available Actions`,
524
- `Respond with a JSON object containing one of these actions:`,
881
+ `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.`,
525
882
  ``,
526
- `### click`,
527
- `Click at a specific coordinate. Use for buttons, links, inputs, checkboxes, etc.`,
528
- `\`{ "action": "click", "x": <number>, "y": <number>, "description": "<what you are clicking>" }\``,
883
+ `### click \u2014 by index (preferred)`,
884
+ `\`{ "action": "click", "index": <n>, "description": "<what you are clicking, ideally with the visible text in quotes>" }\``,
885
+ `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.`,
529
886
  ``,
530
- `### type`,
531
- `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.`,
532
- `\`{ "action": "type", "text": "<text to type>", "x": <number|optional>, "y": <number|optional> }\``,
533
- `Example: \`{ "action": "type", "text": "search query\\n", "x": 640, "y": 300 }\` \u2014 clicks the search box, types, and presses Enter.`,
887
+ `### type \u2014 by index (preferred)`,
888
+ `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).`,
889
+ `\`{ "action": "type", "index": <n>, "text": "<text>", "clear": true, "submit": false }\``,
890
+ `Fallback: \`{ "action": "type", "text": "<text>", "x": <n>, "y": <n> }\` clicks the coordinates first, then types.`,
534
891
  ``,
535
892
  `### scroll`,
536
- `Scroll the page up or down. Use when content is below or above the visible area.`,
537
- `\`{ "action": "scroll", "direction": "up"|"down", "amount": <pixels, optional, default 400> }\``,
893
+ `\`{ "action": "scroll", "direction": "down"|"up", "amount": 400 }\` or \`{ "action": "scroll", "index": <n> }\` to scroll a specific element into view.`,
538
894
  ``,
539
- `### navigate`,
540
- `Navigate to a specific URL. Use when you know the exact URL to visit.`,
541
- `\`{ "action": "navigate", "url": "<full URL>" }\``,
895
+ `### find_text`,
896
+ `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.`,
897
+ `\`{ "action": "find_text", "text": "Annual report 2024" }\``,
542
898
  ``,
543
- `### back`,
544
- `Go back to the previous page.`,
545
- `\`{ "action": "back" }\``,
899
+ `### send_keys`,
900
+ `Send arbitrary keys / shortcuts. Single key, combo (with \`+\`), or space-separated sequence.`,
901
+ `\`{ "action": "send_keys", "keys": "Tab Tab Enter" }\` \xB7 \`{ "action": "send_keys", "keys": "Control+l" }\` \xB7 \`{ "action": "send_keys", "keys": "Escape" }\``,
546
902
  ``,
547
- `### wait`,
548
- `Wait for the page to load or for a timed event. Use sparingly.`,
549
- `\`{ "action": "wait", "ms": <milliseconds> }\``,
903
+ `### dropdown_options / select_dropdown`,
904
+ `Inspect or set a native \`<select>\` by index. **Do not click into native selects** \u2014 the OS overlay is not part of the DOM.`,
905
+ `\`{ "action": "dropdown_options", "index": <n> }\` returns the option list in the next observation.`,
906
+ `\`{ "action": "select_dropdown", "index": <n>, "text": "United States" }\` picks the matching option.`,
550
907
  ``,
551
- `### done`,
552
- `The task is complete. Provide a summary of what was accomplished.`,
553
- `\`{ "action": "done", "result": "<summary of what was accomplished>" }\``,
908
+ `### upload_file`,
909
+ `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.`,
910
+ `\`{ "action": "upload_file", "index": <n>, "path": "/abs/path/to/file.pdf" }\``,
554
911
  ``,
555
- `### fail`,
556
- `The task cannot be completed. Explain why.`,
557
- `\`{ "action": "fail", "reason": "<why the task failed>" }\``,
912
+ `### navigate / back`,
913
+ `\`{ "action": "navigate", "url": "<full URL>" }\` \xB7 \`{ "action": "back" }\``,
558
914
  ``,
559
- `## Rules`,
560
- `1. ALWAYS look at the screenshot carefully before deciding your action.`,
561
- `2. Provide coordinates that target the CENTER of the element you want to interact with.`,
562
- `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.`,
563
- `4. For text inputs: click the input field first (using "type" with x/y), then the text will be typed.`,
564
- `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.`,
565
- `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.`,
566
- `7. If a page is loading or blank, use "wait" with a short delay and try again.`,
567
- `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.`,
568
- `9. NEVER hallucinate content. Only report what you can actually see on the screen.`,
569
- `10. When the task is fully complete, use "done" immediately with a comprehensive result.`,
570
- `11. If after several attempts you cannot complete the task, use "fail" with a clear reason.`,
915
+ `### wait / screenshot`,
916
+ `\`{ "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").`,
571
917
  ``,
572
- `## Response Format`,
573
- `Respond with ONLY a valid JSON object. No markdown, no explanation, just the JSON action.`
574
- ];
575
- if (credentialKeys && credentialKeys.length > 0) {
918
+ `### extract`,
919
+ `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:\`.`,
920
+ `\`{ "action": "extract", "query": "List the top 5 result titles and their prices", "extractLinks": false }\``
921
+ );
922
+ if (allowEvaluate) {
576
923
  lines.push(
577
924
  ``,
578
- `## Secure Credentials`,
579
- `The following credential placeholders are available for use in "type" actions:`,
580
- ...credentialKeys.map((k) => `- \`{{${k}}}\``),
581
- ``,
582
- `When you need to fill in a login form or any field requiring these credentials,`,
583
- `use the EXACT placeholder (e.g. \`{{email}}\`) as the "text" value in a type action.`,
584
- `The system will securely replace them with real values at execution time.`,
585
- `NEVER guess, invent, or ask the user for the actual credential values.`,
586
- `NEVER include real credential values in "done" or "fail" results.`
925
+ `### evaluate (JS escape hatch)`,
926
+ `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.`,
927
+ `\`{ "action": "evaluate", "code": "return document.title" }\``
587
928
  );
588
929
  }
930
+ if (tools.length > 0) {
931
+ lines.push(``, `### tool \u2014 invoke a custom tool`);
932
+ lines.push(
933
+ `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.`
934
+ );
935
+ for (const t of tools) {
936
+ lines.push(`- **${t.name}** \u2014 ${t.description}`);
937
+ }
938
+ }
939
+ lines.push(
940
+ ``,
941
+ `### done / fail`,
942
+ `\`{ "action": "done", "result": "<comprehensive summary of what was accomplished and any data the user asked for>" }\` \u2014 use when the task is complete.`,
943
+ `\`{ "action": "fail", "reason": "<why>" }\` \u2014 use when the task cannot be completed even after several attempts.`,
944
+ ``,
945
+ `## Rules`,
946
+ `1. ALWAYS check the DOM snapshot first. If the target element is listed, use its \`index\`. Coordinates are a fallback.`,
947
+ `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.`,
948
+ `3. Dismiss cookie banners, consent dialogs, and modal popups FIRST \u2014 they intercept clicks on elements behind them.`,
949
+ `4. If a previous action clearly hit the wrong thing, do NOT repeat it. Pick a different element or a different action.`,
950
+ `5. Batch independent actions when safe (e.g. filling 3 form fields). Don't batch when an action navigates or substantially changes the DOM.`,
951
+ `6. NEVER hallucinate. Only report data you can actually see in the snapshot, screenshot, or an extract result.`,
952
+ `7. When the task is fully complete, return \`done\` IMMEDIATELY with a thorough result \u2014 don't add extra confirmation steps.`,
953
+ `8. If after several attempts you cannot make progress, return \`fail\` with a clear reason.`
954
+ );
955
+ appendCredentials(lines, credentialKeys);
589
956
  if (extraInstructions) {
590
957
  lines.push(``, `## Additional Instructions`, extraInstructions);
591
958
  }
959
+ lines.push(``, `## Response Format`, responseFormatLines(maxActions).join("\n"));
592
960
  return lines.join("\n");
593
961
  }
594
- function buildUserMessage(task, pageUrl, pageTitle, stepIndex, actionHistory, domSnapshot) {
962
+ function appendCredentials(lines, credentialKeys) {
963
+ if (!credentialKeys || credentialKeys.length === 0) return;
964
+ lines.push(
965
+ ``,
966
+ `## Secure Credentials`,
967
+ `The following credential placeholders are available for use in "type" actions:`,
968
+ ...credentialKeys.map((k) => `- \`{{${k}}}\``),
969
+ ``,
970
+ `When you need to fill in a login form or any field requiring these credentials,`,
971
+ `use the EXACT placeholder (e.g. \`{{email}}\`) as the "text" value in a type action.`,
972
+ `The system will securely replace them with real values at execution time.`,
973
+ `NEVER guess, invent, or ask the user for the actual credential values.`,
974
+ `NEVER include real credential values in "done" or "fail" results.`
975
+ );
976
+ }
977
+ function responseFormatLines(maxActions) {
978
+ return [
979
+ `Respond with ONLY valid JSON. No markdown, no commentary.`,
980
+ `Either a single action object, or an array of up to ${maxActions} action objects to execute in order.`
981
+ ];
982
+ }
983
+ function buildUserMessage(task, pageUrl, pageTitle, stepIndex, actionHistory, domSnapshot, lastExtract) {
595
984
  const lines = [];
596
985
  lines.push(`**Task:** ${task}`);
597
986
  lines.push(`**Current URL:** ${pageUrl}`);
@@ -599,9 +988,14 @@ function buildUserMessage(task, pageUrl, pageTitle, stepIndex, actionHistory, do
599
988
  lines.push(`**Step:** ${stepIndex + 1}`);
600
989
  if (domSnapshot) {
601
990
  lines.push(``);
602
- lines.push(`**Interactive elements on page (format: [centerX,centerY] role: "label"):**`);
991
+ lines.push(`**Interactive elements (format: [idx] [cx,cy] role: "label"):**`);
603
992
  lines.push(domSnapshot);
604
993
  }
994
+ if (lastExtract) {
995
+ lines.push(``);
996
+ lines.push(`**Last extract result:**`);
997
+ lines.push(lastExtract);
998
+ }
605
999
  if (actionHistory.length > 0) {
606
1000
  lines.push(``);
607
1001
  lines.push(`**Previous actions:**`);
@@ -610,16 +1004,19 @@ function buildUserMessage(task, pageUrl, pageTitle, stepIndex, actionHistory, do
610
1004
  }
611
1005
  }
612
1006
  lines.push(``);
613
- lines.push(`Look at the screenshot and decide the next action to complete the task.`);
1007
+ lines.push(`Decide the next action(s) to complete the task.`);
614
1008
  return lines.join("\n");
615
1009
  }
616
1010
  function summarizeAction(action) {
617
1011
  switch (action.action) {
618
1012
  case "click":
619
- return `Clicked at (${action.x}, ${action.y}): ${action.description}`;
1013
+ if (typeof action.index === "number") return `Clicked [${action.index}]: ${action.description ?? ""}`.trim();
1014
+ return `Clicked at (${action.x}, ${action.y}): ${action.description ?? ""}`.trim();
620
1015
  case "type":
1016
+ if (typeof action.index === "number") return `Typed into [${action.index}]: "${action.text}"`;
621
1017
  return action.x != null ? `Clicked (${action.x}, ${action.y}) and typed "${action.text}"` : `Typed "${action.text}"`;
622
1018
  case "scroll":
1019
+ if (typeof action.index === "number") return `Scrolled [${action.index}] into view`;
623
1020
  return `Scrolled ${action.direction}${action.amount ? ` ${action.amount}px` : ""}`;
624
1021
  case "navigate":
625
1022
  return `Navigated to ${action.url}`;
@@ -628,7 +1025,23 @@ function summarizeAction(action) {
628
1025
  case "wait":
629
1026
  return `Waited ${action.ms}ms`;
630
1027
  case "screenshot":
631
- return `Took an extra screenshot`;
1028
+ return `Requested a fresh screenshot`;
1029
+ case "send_keys":
1030
+ return `Sent keys: ${action.keys}`;
1031
+ case "find_text":
1032
+ return `Scrolled to text: "${action.text}"`;
1033
+ case "evaluate":
1034
+ return `Evaluated JS`;
1035
+ case "dropdown_options":
1036
+ return `Read dropdown options at [${action.index}]`;
1037
+ case "select_dropdown":
1038
+ return `Selected "${action.text}" in dropdown [${action.index}]`;
1039
+ case "upload_file":
1040
+ return `Uploaded file to [${action.index}]: ${action.path}`;
1041
+ case "extract":
1042
+ return `Extracted: "${action.query}"`;
1043
+ case "tool":
1044
+ return `Called tool "${action.name}"`;
632
1045
  case "done":
633
1046
  return `Done: ${action.result}`;
634
1047
  case "fail":
@@ -643,19 +1056,32 @@ var BrowserAgent = class {
643
1056
  name;
644
1057
  eventBus;
645
1058
  model;
1059
+ pageExtractionLLM;
646
1060
  instructions;
1061
+ extendSystemMessage;
1062
+ overrideSystemMessage;
647
1063
  maxSteps;
1064
+ maxFailures;
1065
+ maxActionsPerStep;
1066
+ initialActions;
1067
+ useVision;
1068
+ directlyOpenUrl;
648
1069
  headless;
649
1070
  viewport;
650
1071
  defaultStartUrl;
651
1072
  waitAfterAction;
652
1073
  maxRepeats;
653
1074
  useDOM;
1075
+ allowEvaluate;
1076
+ allowedDomains;
1077
+ prohibitedDomains;
654
1078
  storageState;
1079
+ cdpUrl;
655
1080
  recordVideo;
656
1081
  credentials;
657
1082
  stealth;
658
1083
  humanize;
1084
+ tools;
659
1085
  costTracker;
660
1086
  memoryManager = null;
661
1087
  logger;
@@ -666,19 +1092,32 @@ var BrowserAgent = class {
666
1092
  constructor(config) {
667
1093
  this.name = config.name;
668
1094
  this.model = config.model;
1095
+ this.pageExtractionLLM = config.pageExtractionLLM ?? null;
669
1096
  this.instructions = config.instructions;
1097
+ this.extendSystemMessage = config.extendSystemMessage;
1098
+ this.overrideSystemMessage = config.overrideSystemMessage;
670
1099
  this.maxSteps = config.maxSteps ?? 30;
1100
+ this.maxFailures = config.maxFailures ?? 3;
1101
+ this.maxActionsPerStep = Math.max(1, config.maxActionsPerStep ?? 3);
1102
+ this.initialActions = config.initialActions ?? [];
1103
+ this.useVision = config.useVision ?? "auto";
1104
+ this.directlyOpenUrl = config.directlyOpenUrl ?? true;
671
1105
  this.headless = config.headless ?? true;
672
1106
  this.viewport = config.viewport ?? { width: 1280, height: 720 };
673
1107
  this.defaultStartUrl = config.startUrl;
674
1108
  this.waitAfterAction = config.waitAfterAction ?? 1500;
675
1109
  this.maxRepeats = config.maxRepeats ?? 3;
676
1110
  this.useDOM = config.useDOM ?? true;
1111
+ this.allowEvaluate = config.allowEvaluate ?? false;
1112
+ this.allowedDomains = config.allowedDomains;
1113
+ this.prohibitedDomains = config.prohibitedDomains;
677
1114
  this.storageState = config.storageState;
1115
+ this.cdpUrl = config.cdpUrl;
678
1116
  this.recordVideo = config.recordVideo;
679
1117
  this.credentials = config.credentials;
680
1118
  this.stealth = config.stealth;
681
1119
  this.humanize = config.humanize;
1120
+ this.tools = config.tools ?? [];
682
1121
  this.costTracker = config.costTracker ?? null;
683
1122
  this.eventBus = config.eventBus ?? new EventBus();
684
1123
  this.logger = new Logger({
@@ -691,13 +1130,20 @@ var BrowserAgent = class {
691
1130
  }
692
1131
  async run(task, opts) {
693
1132
  const startTime = Date.now();
694
- const startUrl = opts?.startUrl ?? this.defaultStartUrl;
1133
+ const maxSteps = opts?.maxSteps ?? this.maxSteps;
695
1134
  const sessionId = opts?.sessionId ?? `browser_${Date.now()}`;
696
1135
  const userId = opts?.userId;
697
1136
  const browser = new BrowserProvider();
698
1137
  const steps = [];
699
1138
  const actionHistory = [];
700
- let extraInstructions = this.instructions ?? "";
1139
+ const extractedContent = [];
1140
+ let lastExtractResult;
1141
+ let failureCount = 0;
1142
+ let lastActionKey = "";
1143
+ let repeatCount = 0;
1144
+ let lastActionWasScreenshot = false;
1145
+ const baseExtra = [this.instructions, this.extendSystemMessage].filter(Boolean).join("\n\n");
1146
+ let extraInstructions = baseExtra;
701
1147
  if (this.memoryManager) {
702
1148
  await this.memoryManager.ensureReady();
703
1149
  const memoryContext = await this.memoryManager.buildContext(sessionId, userId, task, this.name);
@@ -708,14 +1154,21 @@ ${memoryContext}` : memoryContext;
708
1154
  }
709
1155
  }
710
1156
  const credentialKeys = this.credentials?.keys();
711
- const systemPrompt = buildSystemPrompt(this.viewport, extraInstructions || void 0, credentialKeys);
712
- let lastActionKey = "";
713
- let repeatCount = 0;
1157
+ const systemPrompt = buildSystemPrompt(this.viewport, extraInstructions || void 0, credentialKeys, {
1158
+ overrideSystemMessage: this.overrideSystemMessage,
1159
+ maxActionsPerStep: this.maxActionsPerStep,
1160
+ allowEvaluate: this.allowEvaluate,
1161
+ tools: this.tools,
1162
+ useVision: this.useVision,
1163
+ useDOM: this.useDOM
1164
+ });
714
1165
  try {
715
1166
  this.logger.info("Launching browser", {
716
1167
  headless: this.headless,
717
1168
  viewport: this.viewport,
718
1169
  useDOM: this.useDOM,
1170
+ useVision: this.useVision,
1171
+ cdpUrl: this.cdpUrl ?? void 0,
719
1172
  recordVideo: !!this.recordVideo,
720
1173
  stealth: !!this.stealth,
721
1174
  humanize: !!this.humanize
@@ -726,42 +1179,72 @@ ${memoryContext}` : memoryContext;
726
1179
  storageState: this.storageState,
727
1180
  recordVideo: this.recordVideo,
728
1181
  stealth: this.stealth,
729
- humanize: this.humanize
1182
+ humanize: this.humanize,
1183
+ cdpUrl: this.cdpUrl
730
1184
  });
1185
+ const startUrl = opts?.startUrl ?? this.defaultStartUrl ?? this.detectUrlInTask(task);
731
1186
  if (startUrl) {
732
1187
  this.logger.info("Navigating to start URL", { url: startUrl });
1188
+ this.assertDomainAllowed(startUrl);
733
1189
  await browser.navigate(startUrl);
734
1190
  }
735
- for (let step = 0; step < this.maxSteps; step++) {
736
- const screenshot = await browser.screenshot();
1191
+ for (const ia of this.initialActions) {
1192
+ try {
1193
+ await this.executeAction(browser, ia, actionHistory, extractedContent);
1194
+ } catch (e) {
1195
+ this.logger.warn("initialAction failed", { action: ia, error: e?.message });
1196
+ }
1197
+ await this.sleep(this.waitAfterAction);
1198
+ }
1199
+ for (let step = 0; step < maxSteps; step++) {
737
1200
  const pageInfo = await browser.getPageInfo();
738
1201
  let domSnapshot;
739
1202
  if (this.useDOM) {
740
- domSnapshot = await browser.extractDOM();
1203
+ const dom = await browser.extractDOM();
1204
+ domSnapshot = dom.text;
1205
+ }
1206
+ const wantVision = this.shouldCaptureVision(step, lastActionWasScreenshot);
1207
+ const screenshot = wantVision ? await browser.screenshot() : Buffer.alloc(0);
1208
+ if (wantVision) this.eventBus.emit("browser.screenshot", { data: screenshot });
1209
+ const userText = buildUserMessage(
1210
+ task,
1211
+ pageInfo.url,
1212
+ pageInfo.title,
1213
+ step,
1214
+ actionHistory,
1215
+ domSnapshot,
1216
+ lastExtractResult
1217
+ );
1218
+ const content = [];
1219
+ if (wantVision) {
1220
+ content.push({ type: "image", data: screenshot.toString("base64"), mimeType: "image/png" });
741
1221
  }
742
- this.eventBus.emit("browser.screenshot", { data: screenshot });
743
- const userText = buildUserMessage(task, pageInfo.url, pageInfo.title, step, actionHistory, domSnapshot);
1222
+ content.push({ type: "text", text: userText });
744
1223
  const messages = [
745
1224
  { role: "system", content: systemPrompt },
746
- {
747
- role: "user",
748
- content: [
749
- {
750
- type: "image",
751
- data: screenshot.toString("base64"),
752
- mimeType: "image/png"
753
- },
754
- { type: "text", text: userText }
755
- ]
756
- }
1225
+ { role: "user", content }
757
1226
  ];
758
- this.logger.debug("Sending screenshot to vision model", { step, url: pageInfo.url });
759
- const response = await this.model.generate(messages, {
760
- temperature: 0.1,
761
- maxTokens: 1024,
762
- apiKey: opts?.apiKey,
763
- responseFormat: "json"
764
- });
1227
+ this.logger.debug("Calling model", { step, url: pageInfo.url, vision: wantVision });
1228
+ let response;
1229
+ try {
1230
+ response = await this.model.generate(messages, {
1231
+ temperature: 0.1,
1232
+ maxTokens: 1024,
1233
+ apiKey: opts?.apiKey,
1234
+ responseFormat: "json"
1235
+ });
1236
+ } catch (e) {
1237
+ failureCount++;
1238
+ this.logger.warn("Model call failed", { error: e?.message, failureCount });
1239
+ if (failureCount > this.maxFailures) {
1240
+ return await this.finalize(browser, steps, startTime, opts, extractedContent, {
1241
+ result: `Model call failed ${failureCount} times: ${e?.message}`,
1242
+ success: false
1243
+ });
1244
+ }
1245
+ actionHistory.push(`(model call failed: ${e?.message ?? "unknown"} \u2014 retrying)`);
1246
+ continue;
1247
+ }
765
1248
  if (this.costTracker && response.usage) {
766
1249
  this.costTracker.track({
767
1250
  runId: sessionId,
@@ -773,78 +1256,119 @@ ${memoryContext}` : memoryContext;
773
1256
  });
774
1257
  }
775
1258
  const raw = typeof response.message.content === "string" ? response.message.content : "";
776
- let action;
1259
+ let parsed;
777
1260
  try {
778
- action = JSON.parse(raw);
1261
+ parsed = JSON.parse(raw);
779
1262
  } catch {
780
- this.logger.warn("Failed to parse model response as JSON, retrying", { raw });
1263
+ failureCount++;
1264
+ this.logger.warn("Failed to parse model response as JSON", { raw, failureCount });
1265
+ if (failureCount > this.maxFailures) {
1266
+ return await this.finalize(browser, steps, startTime, opts, extractedContent, {
1267
+ result: `Model returned invalid JSON ${failureCount} times. Last response: ${raw.slice(0, 200)}`,
1268
+ success: false
1269
+ });
1270
+ }
781
1271
  actionHistory.push("(invalid JSON response \u2014 retrying)");
782
1272
  continue;
783
1273
  }
784
- const step_ = {
785
- index: step,
786
- action,
787
- screenshot,
788
- pageUrl: pageInfo.url,
789
- pageTitle: pageInfo.title,
790
- timestamp: /* @__PURE__ */ new Date(),
791
- dom: domSnapshot
792
- };
793
- steps.push(step_);
794
- let summary = summarizeAction(action);
795
- if (this.credentials) {
796
- summary = this.credentials.mask(summary);
797
- }
798
- const actionKey = JSON.stringify(action);
799
- if (actionKey === lastActionKey) {
800
- repeatCount++;
801
- } else {
802
- lastActionKey = actionKey;
803
- repeatCount = 1;
804
- }
805
- if (repeatCount > this.maxRepeats && action.action !== "done" && action.action !== "fail") {
806
- this.logger.warn("Stuck in a loop \u2014 same action repeated", {
807
- action: action.action,
808
- repeats: repeatCount,
809
- maxRepeats: this.maxRepeats
810
- });
811
- actionHistory.push(
812
- `\u26A0 LOOP DETECTED: "${summary}" repeated ${repeatCount} times. The agent was stuck and auto-stopped. Try a different approach or a different startUrl.`
813
- );
814
- return await this.finalize(browser, steps, startTime, opts, {
815
- result: `Stuck in a loop: "${summary}" was repeated ${repeatCount} times. The page may have a popup, consent banner, or unexpected state blocking progress.`,
816
- success: false
1274
+ const actions = (Array.isArray(parsed) ? parsed : [parsed]).slice(
1275
+ 0,
1276
+ this.maxActionsPerStep
1277
+ );
1278
+ let didTerminate = null;
1279
+ let didNavigate = false;
1280
+ lastActionWasScreenshot = false;
1281
+ for (let ai = 0; ai < actions.length; ai++) {
1282
+ const action = actions[ai];
1283
+ let summary = summarizeAction(action);
1284
+ if (this.credentials) summary = this.credentials.mask(summary);
1285
+ const actionKey = JSON.stringify(action);
1286
+ if (actionKey === lastActionKey) {
1287
+ repeatCount++;
1288
+ } else {
1289
+ lastActionKey = actionKey;
1290
+ repeatCount = 1;
1291
+ }
1292
+ if (repeatCount > this.maxRepeats && action.action !== "done" && action.action !== "fail") {
1293
+ this.logger.warn("Stuck in a loop \u2014 same action repeated", {
1294
+ action: action.action,
1295
+ repeats: repeatCount,
1296
+ maxRepeats: this.maxRepeats
1297
+ });
1298
+ actionHistory.push(
1299
+ `\u26A0 LOOP DETECTED: "${summary}" repeated ${repeatCount} times. Auto-stopping.`
1300
+ );
1301
+ didTerminate = {
1302
+ result: `Stuck in a loop: "${summary}" was repeated ${repeatCount} times. The page may have a popup, consent banner, or unexpected state blocking progress.`,
1303
+ success: false
1304
+ };
1305
+ break;
1306
+ }
1307
+ actionHistory.push(summary);
1308
+ this.logger.info(`Step ${step + 1}.${ai + 1}: ${summary}`);
1309
+ this.eventBus.emit("browser.action", { action });
1310
+ if (action.action === "done") {
1311
+ const result = this.credentials ? this.credentials.mask(action.result) : action.result;
1312
+ didTerminate = { result, success: true };
1313
+ break;
1314
+ }
1315
+ if (action.action === "fail") {
1316
+ const result = this.credentials ? this.credentials.mask(action.reason) : action.reason;
1317
+ didTerminate = { result, success: false };
1318
+ break;
1319
+ }
1320
+ let stepOk = true;
1321
+ let stepOutput;
1322
+ try {
1323
+ const exec = await this.executeAction(browser, action, actionHistory, extractedContent);
1324
+ stepOutput = exec?.output;
1325
+ if (exec?.didNavigate) didNavigate = true;
1326
+ if (action.action === "extract" && exec?.output) {
1327
+ lastExtractResult = exec.output;
1328
+ }
1329
+ if (action.action === "screenshot") {
1330
+ lastActionWasScreenshot = true;
1331
+ }
1332
+ } catch (e) {
1333
+ stepOk = false;
1334
+ failureCount++;
1335
+ this.logger.warn("Action failed", { action: action.action, error: e?.message, failureCount });
1336
+ actionHistory.push(`(error: ${e?.message ?? "unknown"})`);
1337
+ if (failureCount > this.maxFailures) {
1338
+ didTerminate = {
1339
+ result: `Action ${action.action} failed ${failureCount} times. Last error: ${e?.message}`,
1340
+ success: false
1341
+ };
1342
+ break;
1343
+ }
1344
+ }
1345
+ steps.push({
1346
+ index: steps.length,
1347
+ action,
1348
+ screenshot,
1349
+ pageUrl: pageInfo.url,
1350
+ pageTitle: pageInfo.title,
1351
+ timestamp: /* @__PURE__ */ new Date(),
1352
+ dom: domSnapshot,
1353
+ output: stepOutput,
1354
+ ok: stepOk
817
1355
  });
818
- }
819
- actionHistory.push(summary);
820
- this.logger.info(`Step ${step + 1}: ${summary}`);
821
- this.eventBus.emit("browser.action", { action });
822
- this.eventBus.emit("browser.step", {
823
- index: step,
824
- action,
825
- pageUrl: pageInfo.url,
826
- screenshot
827
- });
828
- if (action.action === "done") {
829
- const result = this.credentials ? this.credentials.mask(action.result) : action.result;
830
- return await this.finalize(browser, steps, startTime, opts, {
831
- result,
832
- success: true
1356
+ this.eventBus.emit("browser.step", {
1357
+ index: steps.length - 1,
1358
+ action,
1359
+ pageUrl: pageInfo.url,
1360
+ screenshot
833
1361
  });
1362
+ await this.sleep(this.waitAfterAction);
1363
+ if (didNavigate) break;
834
1364
  }
835
- if (action.action === "fail") {
836
- const result = this.credentials ? this.credentials.mask(action.reason) : action.reason;
837
- return await this.finalize(browser, steps, startTime, opts, {
838
- result,
839
- success: false
840
- });
1365
+ if (didTerminate) {
1366
+ return await this.finalize(browser, steps, startTime, opts, extractedContent, didTerminate);
841
1367
  }
842
- await this.executeAction(browser, action);
843
- await this.sleep(this.waitAfterAction);
844
1368
  }
845
- this.logger.warn("Max steps reached without completing task", { maxSteps: this.maxSteps });
846
- return await this.finalize(browser, steps, startTime, opts, {
847
- result: `Task not completed within ${this.maxSteps} steps. Last actions: ${actionHistory.slice(-3).join("; ")}`,
1369
+ this.logger.warn("Max steps reached without completing task", { maxSteps });
1370
+ return await this.finalize(browser, steps, startTime, opts, extractedContent, {
1371
+ result: `Task not completed within ${maxSteps} steps. Last actions: ${actionHistory.slice(-3).join("; ")}`,
848
1372
  success: false
849
1373
  });
850
1374
  } catch (error) {
@@ -857,7 +1381,8 @@ ${memoryContext}` : memoryContext;
857
1381
  steps,
858
1382
  finalUrl: "",
859
1383
  finalScreenshot: Buffer.alloc(0),
860
- durationMs: Date.now() - startTime
1384
+ durationMs: Date.now() - startTime,
1385
+ extractedContent
861
1386
  };
862
1387
  }
863
1388
  }
@@ -880,9 +1405,43 @@ ${memoryContext}` : memoryContext;
880
1405
  };
881
1406
  }
882
1407
  // ── Private helpers ──────────────────────────────────────────────────
883
- async finalize(browser, steps, startTime, opts, outcome) {
884
- const finalScreenshot = await browser.screenshot();
885
- const finalInfo = await browser.getPageInfo();
1408
+ shouldCaptureVision(step, lastActionWasScreenshot) {
1409
+ if (this.useVision === false) return false;
1410
+ if (this.useVision === true) return true;
1411
+ if (step === 0) return true;
1412
+ if (!this.useDOM) return true;
1413
+ return lastActionWasScreenshot;
1414
+ }
1415
+ detectUrlInTask(task) {
1416
+ if (!this.directlyOpenUrl) return void 0;
1417
+ const match = task.match(/https?:\/\/[^\s)<>"']+/);
1418
+ return match ? match[0] : void 0;
1419
+ }
1420
+ assertDomainAllowed(url) {
1421
+ if (!this.allowedDomains?.length && !this.prohibitedDomains?.length) return;
1422
+ let host;
1423
+ try {
1424
+ host = new URL(url).hostname.toLowerCase();
1425
+ } catch {
1426
+ throw new Error(`Cannot parse URL for domain check: ${url}`);
1427
+ }
1428
+ if (this.allowedDomains?.length && !this.allowedDomains.some((p) => matchDomain(host, p))) {
1429
+ throw new Error(`Navigation blocked: ${host} is not in allowedDomains`);
1430
+ }
1431
+ if (this.prohibitedDomains?.length && this.prohibitedDomains.some((p) => matchDomain(host, p))) {
1432
+ throw new Error(`Navigation blocked: ${host} is in prohibitedDomains`);
1433
+ }
1434
+ }
1435
+ async finalize(browser, steps, startTime, opts, extractedContent, outcome) {
1436
+ let finalScreenshot;
1437
+ let finalInfo;
1438
+ try {
1439
+ finalScreenshot = await browser.screenshot();
1440
+ finalInfo = await browser.getPageInfo();
1441
+ } catch {
1442
+ finalScreenshot = Buffer.alloc(0);
1443
+ finalInfo = { url: "", title: "" };
1444
+ }
886
1445
  if (opts?.saveStorageState) {
887
1446
  try {
888
1447
  await browser.saveStorageState(opts.saveStorageState);
@@ -903,7 +1462,8 @@ ${memoryContext}` : memoryContext;
903
1462
  finalUrl: finalInfo.url,
904
1463
  finalScreenshot,
905
1464
  durationMs: Date.now() - startTime,
906
- videoPath
1465
+ videoPath,
1466
+ extractedContent
907
1467
  };
908
1468
  if (this.memoryManager) {
909
1469
  const sessionId = opts?.sessionId ?? `browser_${startTime}`;
@@ -927,49 +1487,194 @@ ${memoryContext}` : memoryContext;
927
1487
  });
928
1488
  return output;
929
1489
  }
930
- async executeAction(browser, action) {
931
- try {
932
- switch (action.action) {
933
- case "click":
1490
+ /**
1491
+ * Execute a single action. Returns `{ output?, didNavigate? }` for the
1492
+ * caller's state tracking. May throw — the loop handles failure-budget
1493
+ * accounting in that case.
1494
+ */
1495
+ async executeAction(browser, action, _actionHistory, extractedContent) {
1496
+ switch (action.action) {
1497
+ case "click": {
1498
+ if (typeof action.index === "number") {
1499
+ const ok = await browser.clickByIndex(action.index);
1500
+ if (ok) return {};
1501
+ }
1502
+ const keyword = this.extractClickKeyword(action.description);
1503
+ if (keyword && await browser.clickByText(keyword)) {
1504
+ this.logger.debug("Clicked by text", { keyword });
1505
+ return {};
1506
+ }
1507
+ if (typeof action.x === "number" && typeof action.y === "number") {
934
1508
  await browser.click(action.x, action.y);
935
- break;
936
- case "type": {
937
- const resolvedText = this.credentials ? this.credentials.resolve(action.text) : action.text;
938
- if (action.x != null && action.y != null) {
939
- await browser.clickAndType(action.x, action.y, resolvedText);
940
- } else {
941
- await browser.type(resolvedText);
942
- }
943
- if (resolvedText.includes("\n")) {
944
- await browser.pressKey("Enter");
945
- }
946
- break;
1509
+ return {};
947
1510
  }
948
- case "scroll":
949
- await browser.scroll(action.direction, action.amount);
950
- break;
951
- case "navigate":
952
- await browser.navigate(action.url);
953
- break;
954
- case "back":
955
- await browser.back();
956
- break;
957
- case "wait":
958
- await this.sleep(Math.min(action.ms, 1e4));
959
- break;
960
- case "screenshot":
961
- break;
962
- default:
963
- this.logger.warn("Unknown action", { action });
1511
+ throw new Error(`click action has no resolvable target (index/text/coordinates)`);
964
1512
  }
965
- } catch (error) {
966
- this.logger.warn("Action execution failed", { action: action.action, error: error.message });
1513
+ case "type": {
1514
+ const resolvedText = this.credentials ? this.credentials.resolve(action.text) : action.text;
1515
+ const submit = action.submit ?? resolvedText.includes("\n");
1516
+ const cleanText = submit ? resolvedText.replace(/\n+$/, "") : resolvedText;
1517
+ if (typeof action.index === "number") {
1518
+ const ok = await browser.inputByIndex(action.index, cleanText, {
1519
+ clear: action.clear,
1520
+ submit
1521
+ });
1522
+ if (ok) return {};
1523
+ }
1524
+ if (typeof action.x === "number" && typeof action.y === "number") {
1525
+ await browser.clickAndType(action.x, action.y, cleanText);
1526
+ if (submit) await browser.pressKey("Enter");
1527
+ return {};
1528
+ }
1529
+ await browser.type(cleanText);
1530
+ if (submit) await browser.pressKey("Enter");
1531
+ return {};
1532
+ }
1533
+ case "scroll": {
1534
+ if (typeof action.index === "number") {
1535
+ await browser.scrollIntoViewByIndex(action.index);
1536
+ return {};
1537
+ }
1538
+ await browser.scroll(action.direction, action.amount);
1539
+ return {};
1540
+ }
1541
+ case "navigate": {
1542
+ this.assertDomainAllowed(action.url);
1543
+ await browser.navigate(action.url);
1544
+ return { didNavigate: true };
1545
+ }
1546
+ case "back":
1547
+ await browser.back();
1548
+ return { didNavigate: true };
1549
+ case "wait":
1550
+ await this.sleep(Math.min(action.ms, 1e4));
1551
+ return {};
1552
+ case "screenshot":
1553
+ return {};
1554
+ case "send_keys":
1555
+ await browser.sendKeys(action.keys);
1556
+ return {};
1557
+ case "find_text":
1558
+ await browser.findText(action.text);
1559
+ return {};
1560
+ case "evaluate": {
1561
+ if (!this.allowEvaluate) {
1562
+ throw new Error("evaluate is disabled. Set allowEvaluate: true on BrowserAgent to enable.");
1563
+ }
1564
+ const out = await browser.evaluate(action.code);
1565
+ return { output: `evaluate \u2192 ${out}` };
1566
+ }
1567
+ case "dropdown_options": {
1568
+ const options = await browser.dropdownOptions(action.index);
1569
+ const formatted = options.map((o, i) => `${i + 1}. "${o.label}" (value="${o.value}")${o.selected ? " [selected]" : ""}`).join("\n");
1570
+ return { output: `Dropdown [${action.index}] options:
1571
+ ${formatted || "(no options found)"}` };
1572
+ }
1573
+ case "select_dropdown": {
1574
+ const ok = await browser.selectDropdown(action.index, action.text);
1575
+ if (!ok) throw new Error(`Could not select "${action.text}" on dropdown [${action.index}]`);
1576
+ return {};
1577
+ }
1578
+ case "upload_file": {
1579
+ const ok = await browser.uploadFileByIndex(action.index, action.path);
1580
+ if (!ok) throw new Error(`Could not upload "${action.path}" to [${action.index}]`);
1581
+ return {};
1582
+ }
1583
+ case "extract": {
1584
+ const pageText = await browser.pageText({ extractLinks: action.extractLinks });
1585
+ const model = this.pageExtractionLLM ?? this.model;
1586
+ const messages = [
1587
+ {
1588
+ role: "system",
1589
+ 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."
1590
+ },
1591
+ {
1592
+ role: "user",
1593
+ content: `Query: ${action.query}
1594
+
1595
+ Page content:
1596
+ ${pageText}`
1597
+ }
1598
+ ];
1599
+ const response = await model.generate(messages, { temperature: 0, maxTokens: 2048 });
1600
+ const out = typeof response.message.content === "string" ? response.message.content : "";
1601
+ const masked = this.credentials ? this.credentials.mask(out) : out;
1602
+ extractedContent.push(masked);
1603
+ return { output: masked };
1604
+ }
1605
+ case "tool": {
1606
+ const tool = this.tools.find((t) => t.name === action.name);
1607
+ if (!tool) throw new Error(`Tool "${action.name}" is not registered on this BrowserAgent`);
1608
+ const ctx = new RunContext({
1609
+ sessionId: `browser_${this.name}_${Date.now()}`,
1610
+ eventBus: this.eventBus
1611
+ });
1612
+ const result = await tool.execute(action.args ?? {}, ctx);
1613
+ const out = typeof result === "string" ? result : JSON.stringify(result);
1614
+ return { output: `${action.name} \u2192 ${out}` };
1615
+ }
1616
+ // `done` and `fail` are intercepted by the run loop before we get
1617
+ // here. Listing them keeps the discriminated union exhaustive.
1618
+ case "done":
1619
+ case "fail":
1620
+ return {};
967
1621
  }
968
1622
  }
969
1623
  sleep(ms) {
970
1624
  return new Promise((resolve) => setTimeout(resolve, ms));
971
1625
  }
1626
+ /**
1627
+ * Parse a quoted target keyword from a click action's `description`.
1628
+ * Returns `undefined` for generic / ambiguous labels (login buttons,
1629
+ * close, OK, etc.) where a substring text match could fire on the
1630
+ * wrong element.
1631
+ */
1632
+ extractClickKeyword(description) {
1633
+ if (!description) return void 0;
1634
+ const match = description.match(/['"\u2018\u2019\u201C\u201D]([^'"\u2018\u2019\u201C\u201D]{1,80})['"\u2018\u2019\u201C\u201D]/);
1635
+ if (!match) return void 0;
1636
+ const keyword = match[1].trim();
1637
+ if (!keyword || keyword.length < 2) return void 0;
1638
+ const skip = /* @__PURE__ */ new Set([
1639
+ "log in",
1640
+ "login",
1641
+ "sign in",
1642
+ "sign up",
1643
+ "submit",
1644
+ "close",
1645
+ "ok",
1646
+ "okay",
1647
+ "cancel",
1648
+ "yes",
1649
+ "no",
1650
+ "x",
1651
+ "continue",
1652
+ "next",
1653
+ "back",
1654
+ "accept",
1655
+ "dismiss",
1656
+ "got it",
1657
+ "agree",
1658
+ "i agree",
1659
+ "allow",
1660
+ "deny"
1661
+ ]);
1662
+ if (skip.has(keyword.toLowerCase())) return void 0;
1663
+ return keyword;
1664
+ }
972
1665
  };
1666
+ function matchDomain(host, pattern) {
1667
+ const h = host.toLowerCase();
1668
+ let p = pattern.toLowerCase().replace(/^https?\*?:\/\//, "").replace(/^\/+/, "");
1669
+ if (p.includes("/")) p = p.split("/")[0];
1670
+ if (p === "*") return true;
1671
+ if (p === h) return true;
1672
+ if (p.startsWith("*.")) {
1673
+ const base = p.slice(2);
1674
+ return h === base || h.endsWith(`.${base}`);
1675
+ }
1676
+ return false;
1677
+ }
973
1678
 
974
1679
  // src/credential-vault.ts
975
1680
  var CredentialVault = class {