browser-use 0.4.0 → 0.6.0

package/dist/controller/service.js

@@ -5,8 +5,8 @@ import { ActionResult } from '../agent/views.js';
 import { ClickCoordinateEvent, ClickElementEvent, CloseTabEvent, GetDropdownOptionsEvent, GoBackEvent, NavigateToUrlEvent, ScrollEvent, ScrollToTextEvent, ScreenshotEvent, SelectDropdownOptionEvent, SendKeysEvent, SwitchTabEvent, TypeTextEvent, UploadFileEvent, WaitEvent, } from '../browser/events.js';
 import { BrowserError } from '../browser/views.js';
 import { chunkMarkdownByStructure, extractCleanMarkdownFromHtml, } from '../dom/markdown-extractor.js';
-import { extractPdfText, FileSystem } from '../filesystem/file-system.js';
-import { ClickElementActionIndexOnlySchema, ClickElementActionSchema, CloseTabActionSchema, DoneActionSchema, EvaluateActionSchema, ExtractStructuredDataActionSchema, FindElementsActionSchema, DropdownOptionsActionSchema, SelectDropdownActionSchema, GoToUrlActionSchema, InputTextActionSchema, NoParamsActionSchema, ReadLongContentActionSchema, ReadFileActionSchema, ReplaceFileStrActionSchema, ScrollActionSchema, ScrollToTextActionSchema, SearchActionSchema, SearchPageActionSchema, SearchGoogleActionSchema, ScreenshotActionSchema, StructuredOutputActionSchema, SwitchTabActionSchema, UploadFileActionSchema, WaitActionSchema, WriteFileActionSchema, SendKeysActionSchema, SheetsRangeActionSchema, SheetsUpdateActionSchema, SheetsInputActionSchema, } from './views.js';
+import { FileSystem } from '../filesystem/file-system.js';
+import { ClickElementActionIndexOnlySchema, ClickElementActionSchema, CloseTabActionSchema, DoneActionSchema, EvaluateActionSchema, ExtractStructuredDataActionSchema, FindElementsActionSchema, DropdownOptionsActionSchema, SelectDropdownActionSchema, GoToUrlActionSchema, InputTextActionSchema, NoParamsActionSchema, ReadFileActionSchema, ReplaceFileStrActionSchema, ScrollActionSchema, ScrollToTextActionSchema, SearchActionSchema, SearchPageActionSchema, SearchGoogleActionSchema, ScreenshotActionSchema, SaveAsPdfActionSchema, StructuredOutputActionSchema, SwitchTabActionSchema, UploadFileActionSchema, WaitActionSchema, WriteFileActionSchema, SendKeysActionSchema, SheetsRangeActionSchema, SheetsUpdateActionSchema, SheetsInputActionSchema, } from './views.js';
 import { Registry } from './registry/service.js';
 import { SystemMessage, UserMessage } from '../llm/messages.js';
 import { createLogger } from '../logging-config.js';
@@ -32,6 +32,16 @@ const createAbortError = (reason) => {
 const isAbortError = (error) => {
     return error instanceof Error && error.name === 'AbortError';
 };
+const resolveUniqueOutputPath = async (directory, fileName) => {
+    const parsed = path.parse(fileName);
+    let candidate = path.join(directory, fileName);
+    let counter = 1;
+    while (fs.existsSync(candidate)) {
+        candidate = path.join(directory, `${parsed.name} (${counter})${parsed.ext}`);
+        counter += 1;
+    }
+    return candidate;
+};
 const throwIfAborted = (signal) => {
     if (signal?.aborted) {
         throw createAbortError(signal.reason);
@@ -864,7 +874,7 @@ export class Controller {
     registerContentActions() {
         const registry = this.registry;
         const contentLogger = this.logger;
-        const extractStructuredDescription = "LLM extracts structured data from page markdown. Use when: on right page, know what to extract, haven't called before on same page+query. Can't get interactive elements. Set extract_links=True for URLs. Use start_from_char if previous extraction was truncated to extract data further down the page.";
+        const extractStructuredDescription = "LLM extracts structured data from page markdown. Use when: on right page, know what to extract, haven't called before on same page+query. Can't get interactive elements. Set extract_links=True for URLs. Use start_from_char if previous extraction was truncated to extract data further down the page. When paginating across pages, pass already_collected with item identifiers (names/URLs) from prior pages to avoid duplicates.";
         this.registry.action(extractStructuredDescription, {
             param_model: ExtractStructuredDataActionSchema,
         })(async function extract_structured_data(params, { page, page_extraction_llm, extraction_schema, file_system, signal }) {
@@ -989,6 +999,11 @@ export class Controller {
             }
             content = sanitize_surrogates(content);
             const sanitizedQuery = sanitize_surrogates(params.query);
+            const alreadyCollected = Array.isArray(params.already_collected)
+                ? params.already_collected
+                    .map((item) => sanitize_surrogates(String(item)).trim())
+                    .filter(Boolean)
+                : [];
             const parseJsonFromCompletion = (completion) => {
                 const trimmed = completion.trim();
                 const fencedMatch = trimmed.match(/```(?:json)?\s*([\s\S]*?)```/i);
@@ -1018,12 +1033,20 @@ You will be given a query, a JSON Schema, and the markdown of a webpage that has
 - Your response MUST conform to the provided JSON Schema exactly.
 - If a required field's value cannot be found on the page, use null (if the schema allows it) or an empty string / empty array as appropriate.
 - If the content was truncated, extract what is available from the visible portion.
+- If <already_collected> items are provided, skip any items whose name/title/URL matches those listed. Do not include duplicates.
 </instructions>`.trim();
                 const schemaJson = JSON.stringify(effectiveOutputSchema, null, 2);
+                const alreadyCollectedSection = alreadyCollected.length > 0
+                    ? `\n\n<already_collected>\nSkip items whose name/title/URL matches any of these already-collected identifiers:\n${alreadyCollected
+                        .slice(0, 100)
+                        .map((item) => `- ${item}`)
+                        .join('\n')}\n</already_collected>`
+                    : '';
                 const prompt = `<query>\n${sanitizedQuery}\n</query>\n\n` +
                     `<output_schema>\n${schemaJson}\n</output_schema>\n\n` +
                     `<content_stats>\n${statsSummary}\n</content_stats>\n\n` +
-                    `<webpage_content>\n${content}\n</webpage_content>`;
+                    `<webpage_content>\n${content}\n</webpage_content>` +
+                    alreadyCollectedSection;
                 const response = await page_extraction_llm.ainvoke([new SystemMessage(systemPrompt), new UserMessage(prompt)], undefined, { signal: signal ?? undefined });
                 throwIfAborted(signal);
                 const completion = response?.completion;
@@ -1086,6 +1109,7 @@ You will be given a query and the markdown of a webpage that has been filtered t
 - If the information relevant to the query is not available in the page, your response should mention that.
 - If the query asks for all items, products, etc., make sure to directly list all of them.
 - If the content was truncated and you need more information, note that the user can use start_from_char parameter to continue from where truncation occurred.
+- If <already_collected> items are provided, exclude any results whose name/title/URL matches those already collected. Do not include duplicates.
 </instructions>
 
 <output>
@@ -1094,7 +1118,13 @@ You will be given a query and the markdown of a webpage that has been filtered t
 </output>`.trim();
             const prompt = `<query>\n${sanitizedQuery}\n</query>\n\n` +
                 `<content_stats>\n${statsSummary}\n</content_stats>\n\n` +
-                `<webpage_content>\n${content}\n</webpage_content>`;
+                `<webpage_content>\n${content}\n</webpage_content>` +
+                (alreadyCollected.length > 0
+                    ? `\n\n<already_collected>\nSkip items whose name/title/URL matches any of these already-collected identifiers:\n${alreadyCollected
+                        .slice(0, 100)
+                        .map((item) => `- ${item}`)
+                        .join('\n')}\n</already_collected>`
+                    : '');
             const response = await page_extraction_llm.ainvoke([new SystemMessage(systemPrompt), new UserMessage(prompt)], undefined, { signal: signal ?? undefined });
             throwIfAborted(signal);
             const completion = response?.completion;
@@ -1660,363 +1690,6 @@ You will be given a query and the markdown of a webpage that has been filtered t
                 include_extracted_content_only_once: true,
             });
         });
-        this.registry.action('Intelligently read long content to find specific information. Works on current page (source="page") or files. For large content, uses search to identify relevant sections. Best for long articles, documents, or any content where you know what you are looking for.', { param_model: ReadLongContentActionSchema })(async function read_long_content(params, { browser_session, page_extraction_llm, available_file_paths, signal }) {
-            throwIfAborted(signal);
-            const goal = params.goal.trim();
-            const source = (params.source || 'page').trim();
-            const context = (params.context || '').trim();
-            const maxChars = 50000;
-            const chunkSize = 2000;
-            const fallbackSearchTerms = (() => {
-                const tokens = `${goal} ${context}`
-                    .toLowerCase()
-                    .match(/[a-z0-9][a-z0-9-]{2,}/g);
-                if (!tokens?.length) {
-                    return goal ? [goal] : ['content'];
-                }
-                return Array.from(new Set(tokens)).slice(0, 5);
-            })();
-            const extractSearchTerms = async () => {
-                const extractionLlm = page_extraction_llm;
-                if (!extractionLlm || typeof extractionLlm.ainvoke !== 'function') {
-                    return fallbackSearchTerms;
-                }
-                const prompt = `Extract 3-5 key search terms from this goal that would help find relevant sections.
-Return only the terms, one per line, no numbering or bullets.
-
-Goal: ${goal}
-
-Context: ${context}`;
-                try {
-                    const response = await runWithTimeoutAndSignal(async () => (await extractionLlm.ainvoke([new UserMessage(prompt)], undefined, { signal: signal ?? undefined })), 12000, signal, 'Timed out extracting search terms');
-                    const parsed = (response?.completion ?? '')
-                        .split('\n')
-                        .map((line) => line
-                        .trim()
-                        .replace(/^[-\d.)\s]+/, '')
-                        .trim())
-                        .filter(Boolean);
-                    const unique = Array.from(new Set(parsed)).slice(0, 5);
-                    return unique.length ? unique : fallbackSearchTerms;
-                }
-                catch (error) {
-                    if (isAbortError(error)) {
-                        throw error;
-                    }
-                    return fallbackSearchTerms;
-                }
-            };
-            const escapeRegExp = (value) => value.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
-            const searchText = (value, pattern, contextChars = 100) => {
-                let regex;
-                try {
-                    regex = new RegExp(pattern, 'gi');
-                }
-                catch {
-                    regex = new RegExp(escapeRegExp(pattern), 'gi');
-                }
-                const matches = [];
-                let match = regex.exec(value);
-                while (match != null) {
-                    const start = Math.max(0, match.index - contextChars);
-                    const end = Math.min(value.length, match.index + match[0].length + contextChars);
-                    matches.push({
-                        position: match.index,
-                        snippet: value.slice(start, end),
-                    });
-                    if (!regex.global) {
-                        break;
-                    }
-                    match = regex.exec(value);
-                }
-                return matches;
-            };
-            const chunkContent = (value, size = chunkSize) => {
-                const chunks = [];
-                for (let start = 0; start < value.length; start += size) {
-                    chunks.push({
-                        start,
-                        end: Math.min(start + size, value.length),
-                        text: value.slice(start, start + size),
-                    });
-                }
-                return chunks;
-            };
-            const getCurrentPageUrl = (page) => {
-                const value = page?.url;
-                if (typeof value === 'function') {
-                    return String(value.call(page) ?? '');
-                }
-                return typeof value === 'string' ? value : '';
-            };
-            const readPdfByPage = async (filePath) => {
-                const buffer = await fsp.readFile(filePath);
-                try {
-                    const pdfParseModule = (await import('pdf-parse'));
-                    if (typeof pdfParseModule.PDFParse === 'function') {
-                        const Parser = pdfParseModule.PDFParse;
-                        const parser = new Parser({ data: buffer });
-                        try {
-                            let numPages = 0;
-                            try {
-                                const info = await parser.getInfo?.({ parsePageInfo: false });
-                                numPages = Number(info?.total ?? 0);
-                            }
-                            catch {
-                                numPages = 0;
-                            }
-                            if (!Number.isFinite(numPages) || numPages <= 0) {
-                                const full = await parser.getText();
-                                const text = typeof full?.text === 'string' ? full.text : '';
-                                return {
-                                    numPages: 1,
-                                    pageTexts: [text],
-                                    totalChars: text.length,
-                                };
-                            }
-                            const pageTexts = [];
-                            let totalChars = 0;
-                            for (let pageNumber = 1; pageNumber <= numPages; pageNumber += 1) {
-                                const pageResult = await parser.getText({
-                                    partial: [pageNumber],
-                                });
-                                const text = typeof pageResult?.text === 'string' ? pageResult.text : '';
-                                pageTexts.push(text);
-                                totalChars += text.length;
-                            }
-                            return {
-                                numPages,
-                                pageTexts,
-                                totalChars,
-                            };
-                        }
-                        finally {
-                            if (typeof parser.destroy === 'function') {
-                                await parser.destroy();
-                            }
-                        }
-                    }
-                }
-                catch {
-                    // Fall back to the compatibility parser.
-                }
-                const parsed = await extractPdfText(buffer);
-                const text = parsed.text ?? '';
-                return {
-                    numPages: Math.max(parsed.totalPages, 1),
-                    pageTexts: [text],
-                    totalChars: text.length,
-                };
-            };
-            try {
-                let content = '';
-                let sourceName = 'content';
-                if (source.toLowerCase() === 'page') {
-                    if (!browser_session) {
-                        throw new BrowserError('Browser session missing for page content.');
-                    }
-                    const page = await browser_session.get_current_page();
-                    if (!page?.content) {
-                        throw new BrowserError('No active page available to read content.');
-                    }
-                    const html = await page.content();
-                    const pageUrl = getCurrentPageUrl(page);
-                    const markdown = extractCleanMarkdownFromHtml(html || '', {
-                        extract_links: false,
-                        method: 'page_content',
-                        url: pageUrl || undefined,
-                    });
-                    content = markdown.content;
-                    sourceName = 'current page';
-                    if (!content) {
-                        return new ActionResult({
-                            extracted_content: 'Error: No page content available',
-                            long_term_memory: 'Failed to read page: no content',
-                        });
-                    }
-                }
-                else {
-                    const allowedPaths = new Set(Array.isArray(available_file_paths) ? available_file_paths : []);
-                    const downloadedFiles = Array.isArray(browser_session?.downloaded_files)
-                        ? browser_session.downloaded_files
-                        : [];
-                    for (const filePath of downloadedFiles) {
-                        allowedPaths.add(filePath);
-                    }
-                    if (!allowedPaths.has(source)) {
-                        const message = `Error: File path not in available_file_paths: ${source}. ` +
-                            'The user must add this path to available_file_paths when creating the Agent.';
-                        return new ActionResult({
-                            extracted_content: message,
-                            long_term_memory: `Failed to read: file path not allowed: ${source}`,
-                        });
-                    }
-                    if (!fs.existsSync(source)) {
-                        return new ActionResult({
-                            extracted_content: `Error: File not found: ${source}`,
-                            long_term_memory: 'Failed to read: file not found',
-                        });
-                    }
-                    const ext = path.extname(source).toLowerCase();
-                    sourceName = path.basename(source);
-                    if (ext === '.pdf') {
-                        const pdfData = await readPdfByPage(source);
-                        const numPages = pdfData.numPages;
-                        const pageTexts = pdfData.pageTexts;
-                        const totalChars = pdfData.totalChars;
-                        if (totalChars <= maxChars) {
-                            const contentParts = [];
-                            for (let pageIndex = 0; pageIndex < pageTexts.length; pageIndex += 1) {
-                                const pageText = pageTexts[pageIndex] ?? '';
-                                if (!pageText.trim()) {
-                                    continue;
-                                }
-                                contentParts.push(`--- Page ${pageIndex + 1} ---\n${pageText}`);
-                            }
-                            const allContent = contentParts.join('\n\n');
-                            return new ActionResult({
-                                extracted_content: `PDF: ${sourceName} (${numPages} pages)\n\n${allContent}`,
-                                long_term_memory: `Read ${sourceName} (${numPages} pages, ${totalChars.toLocaleString()} chars) for goal: ${goal.slice(0, 50)}`,
-                                include_extracted_content_only_once: true,
-                            });
-                        }
-                        const searchTerms = await extractSearchTerms();
-                        const pageScores = new Map();
-                        for (const term of searchTerms) {
-                            if (!term.trim()) {
-                                continue;
-                            }
-                            const pattern = new RegExp(escapeRegExp(term), 'i');
-                            for (let pageIndex = 0; pageIndex < pageTexts.length; pageIndex += 1) {
-                                const pageText = pageTexts[pageIndex] ?? '';
-                                if (pattern.test(pageText)) {
-                                    const pageNumber = pageIndex + 1;
-                                    pageScores.set(pageNumber, (pageScores.get(pageNumber) ?? 0) + 1);
-                                }
-                            }
-                        }
-                        const pagesToRead = [1];
-                        const sortedPages = Array.from(pageScores.entries()).sort((a, b) => b[1] - a[1]);
-                        for (const [pageNumber] of sortedPages) {
-                            if (!pagesToRead.includes(pageNumber)) {
-                                pagesToRead.push(pageNumber);
-                            }
-                        }
-                        const contentParts = [];
-                        let charsUsed = 0;
-                        const pagesIncluded = [];
-                        const pageOrder = Array.from(new Set(pagesToRead)).sort((a, b) => a - b);
-                        for (const pageNumber of pageOrder) {
-                            const pageText = pageTexts[pageNumber - 1] ?? '';
-                            const pageHeader = `--- Page ${pageNumber} ---\n`;
-                            const remaining = maxChars - charsUsed;
-                            if (remaining < pageHeader.length + 50) {
-                                break;
-                            }
-                            let pageContent = `${pageHeader}${pageText}`;
-                            if (pageContent.length > remaining) {
-                                const truncationSuffix = '\n[...truncated]';
-                                pageContent =
-                                    pageContent.slice(0, remaining - truncationSuffix.length) +
-                                        truncationSuffix;
-                            }
-                            contentParts.push(pageContent);
-                            charsUsed += pageContent.length;
-                            pagesIncluded.push(pageNumber);
-                        }
-                        const partialPdfContent = contentParts.join('\n\n');
-                        return new ActionResult({
-                            extracted_content: `PDF: ${sourceName} (${numPages} pages, showing ${pagesIncluded.length} relevant)\n\n` +
-                                partialPdfContent,
-                            long_term_memory: `Read ${sourceName} (${pagesIncluded.length} relevant pages of ${numPages}) ` +
-                                `for goal: ${goal.slice(0, 50)}`,
-                            include_extracted_content_only_once: true,
-                        });
-                    }
-                    const fileBuffer = await fsp.readFile(source);
-                    content = fileBuffer.toString('utf-8');
-                }
-                if (!content.trim()) {
-                    return new ActionResult({
-                        extracted_content: `Error: No readable content found in ${sourceName}`,
-                        long_term_memory: `Failed to read ${sourceName}: no content`,
-                    });
-                }
-                if (content.length <= maxChars) {
-                    return new ActionResult({
-                        extracted_content: `Content from ${sourceName} (${content.length.toLocaleString()} chars):\n\n${content}`,
-                        long_term_memory: `Read ${sourceName} (${content.length.toLocaleString()} chars) for goal: ${goal.slice(0, 50)}`,
-                        include_extracted_content_only_once: true,
-                    });
-                }
-                const searchTerms = await extractSearchTerms();
-                const chunks = chunkContent(content, chunkSize);
-                const chunkScores = new Map();
-                for (const term of searchTerms) {
-                    const matches = searchText(content, term);
-                    for (const match of matches) {
-                        for (let index = 0; index < chunks.length; index += 1) {
-                            const chunk = chunks[index];
-                            if (chunk &&
-                                chunk.start <= match.position &&
-                                match.position < chunk.end) {
-                                chunkScores.set(index, (chunkScores.get(index) ?? 0) + 1);
-                                break;
-                            }
-                        }
-                    }
-                }
-                if (!chunkScores.size) {
-                    const truncated = content.slice(0, maxChars);
-                    return new ActionResult({
-                        extracted_content: `Content from ${sourceName} (first ${maxChars.toLocaleString()} of ${content.length.toLocaleString()} chars):\n\n${truncated}`,
-                        long_term_memory: `Read ${sourceName} (truncated to ${maxChars.toLocaleString()} chars, no matches for search terms)`,
-                        include_extracted_content_only_once: true,
-                    });
-                }
-                const sortedChunks = Array.from(chunkScores.entries()).sort((a, b) => b[1] - a[1]);
-                const selectedIndices = new Set([0]);
-                for (const [chunkIndex] of sortedChunks) {
-                    selectedIndices.add(chunkIndex);
-                }
-                const resultParts = [];
-                let totalChars = 0;
-                const orderedIndices = Array.from(selectedIndices).sort((a, b) => a - b);
-                for (const index of orderedIndices) {
-                    const chunk = chunks[index];
-                    if (!chunk) {
-                        continue;
-                    }
-                    if (totalChars + chunk.text.length > maxChars) {
-                        break;
-                    }
-                    const previousIndex = index - 1;
-                    if (index > 0 && !selectedIndices.has(previousIndex)) {
-                        resultParts.push('\n[...]\n');
-                    }
-                    resultParts.push(chunk.text);
-                    totalChars += chunk.text.length;
-                }
-                const resultContent = resultParts.join('');
-                return new ActionResult({
-                    extracted_content: `Content from ${sourceName} (relevant sections, ${totalChars.toLocaleString()} of ${content.length.toLocaleString()} chars):\n\n` +
-                        resultContent,
-                    long_term_memory: `Read ${sourceName} (${selectedIndices.size} relevant sections of ${chunks.length}) ` +
-                        `for goal: ${goal.slice(0, 50)}`,
-                    include_extracted_content_only_once: true,
-                });
-            }
-            catch (error) {
-                if (isAbortError(error)) {
-                    throw error;
-                }
-                const errorMessage = `Error reading content: ${String(error.message ?? error)}`;
-                return new ActionResult({
-                    extracted_content: errorMessage,
-                    long_term_memory: errorMessage,
-                });
-            }
-        });
         this.registry.action('Write content to a file. By default this OVERWRITES the entire file - use append=true to add to an existing file, or use replace_file for targeted edits within a file. ' +
             'FILENAME RULES: Use only letters, numbers, underscores, hyphens, dots, parentheses. Spaces are auto-converted to hyphens. ' +
             'SUPPORTED EXTENSIONS: .txt, .md, .json, .jsonl, .csv, .html, .xml, .pdf, .docx. ' +
@@ -2096,6 +1769,77 @@ Context: ${context}`;
                 },
             });
         });
+        this.registry.action('Save the current page as a PDF file. Returns the file path of the saved PDF. Use this to capture the full page content as a printable document.', { param_model: SaveAsPdfActionSchema })(async function save_as_pdf(params, { browser_session, file_system, signal }) {
+            if (!browser_session)
+                throw new Error('Browser session missing');
+            throwIfAborted(signal);
+            const paperSizes = {
+                letter: { width: 8.5, height: 11 },
+                legal: { width: 8.5, height: 14 },
+                a4: { width: 8.27, height: 11.69 },
+                a3: { width: 11.69, height: 16.54 },
+                tabloid: { width: 11, height: 17 },
+            };
+            const page = await browser_session.get_current_page?.();
+            if (!page) {
+                throw new BrowserError('No active page available for save_as_pdf.');
+            }
+            const paperKey = String(params.paper_format ?? 'Letter').toLowerCase();
+            const paperSize = paperSizes[paperKey] ?? paperSizes.letter;
+            const cdpSession = await browser_session.get_or_create_cdp_session?.(page);
+            if (!cdpSession?.send) {
+                throw new BrowserError('CDP session unavailable for save_as_pdf.');
+            }
+            const result = await cdpSession.send('Page.printToPDF', {
+                printBackground: params.print_background,
+                landscape: params.landscape,
+                scale: params.scale,
+                paperWidth: paperSize.width,
+                paperHeight: paperSize.height,
+                preferCSSPageSize: true,
+            });
+            const pdfData = result && typeof result.data === 'string' ? result.data : null;
+            if (!pdfData) {
+                throw new BrowserError('CDP Page.printToPDF returned no data.');
+            }
+            const fsInstance = file_system ?? new FileSystem(process.cwd(), false);
+            let fileName = params.file_name?.trim();
+            if (!fileName) {
+                try {
+                    const titlePromise = typeof page.title === 'function'
+                        ? page.title()
+                        : Promise.resolve('');
+                    const pageTitle = await Promise.race([
+                        titlePromise,
+                        new Promise((_, reject) => {
+                            setTimeout(() => reject(new Error('timeout')), 2000);
+                        }),
+                    ]);
+                    const safeTitle = String(pageTitle)
+                        .replace(/[^\w\s-]+/g, '')
+                        .trim()
+                        .slice(0, 50);
+                    fileName = safeTitle || 'page';
+                }
+                catch {
+                    fileName = 'page';
+                }
+            }
+            if (!fileName.toLowerCase().endsWith('.pdf')) {
+                fileName = `${fileName}.pdf`;
+            }
+            fileName = FileSystem.sanitize_filename(fileName);
+            const filePath = await resolveUniqueOutputPath(fsInstance.get_dir(), fileName);
+            await fsp.writeFile(filePath, Buffer.from(pdfData, 'base64'));
+            const fileSize = (await fsp.stat(filePath)).size;
+            const baseName = path.basename(filePath);
+            const msg = `Saved page as PDF: ${baseName} (${fileSize.toLocaleString()} bytes)`;
+            return new ActionResult({
+                extracted_content: msg,
+                long_term_memory: `${msg}. Full path: ${filePath}`,
+                attachments: [filePath],
+            });
+        });
         this.registry.action('Execute browser JavaScript on the current page and return the result.', { param_model: EvaluateActionSchema })(async function evaluate(params, { browser_session, signal }) {
             if (!browser_session)
                 throw new Error('Browser session missing');

package/dist/controller/views.d.ts

@@ -77,6 +77,14 @@ export declare const ScreenshotActionSchema: z.ZodObject<{
     file_name: z.ZodOptional<z.ZodString>;
 }, z.core.$strip>;
 export type ScreenshotAction = z.infer<typeof ScreenshotActionSchema>;
+export declare const SaveAsPdfActionSchema: z.ZodObject<{
+    file_name: z.ZodOptional<z.ZodString>;
+    print_background: z.ZodDefault<z.ZodBoolean>;
+    landscape: z.ZodDefault<z.ZodBoolean>;
+    scale: z.ZodDefault<z.ZodNumber>;
+    paper_format: z.ZodDefault<z.ZodString>;
+}, z.core.$strip>;
+export type SaveAsPdfAction = z.infer<typeof SaveAsPdfActionSchema>;
 export declare const EvaluateActionSchema: z.ZodObject<{
     code: z.ZodString;
 }, z.core.$strip>;
@@ -90,6 +98,7 @@ export declare const ExtractStructuredDataActionSchema: z.ZodObject<{
     extract_links: z.ZodDefault<z.ZodBoolean>;
     start_from_char: z.ZodDefault<z.ZodNumber>;
     output_schema: z.ZodOptional<z.ZodNullable<z.ZodRecord<z.ZodString, z.ZodUnknown>>>;
+    already_collected: z.ZodDefault<z.ZodArray<z.ZodString>>;
 }, z.core.$strip>;
 export type ExtractStructuredDataAction = z.infer<typeof ExtractStructuredDataActionSchema>;
 export declare const SearchPageActionSchema: z.ZodObject<{
@@ -112,12 +121,6 @@ export declare const ReadFileActionSchema: z.ZodObject<{
     file_name: z.ZodString;
 }, z.core.$strip>;
 export type ReadFileAction = z.infer<typeof ReadFileActionSchema>;
-export declare const ReadLongContentActionSchema: z.ZodObject<{
-    goal: z.ZodString;
-    source: z.ZodDefault<z.ZodString>;
-    context: z.ZodDefault<z.ZodString>;
-}, z.core.$strip>;
-export type ReadLongContentAction = z.infer<typeof ReadLongContentActionSchema>;
 export declare const WriteFileActionSchema: z.ZodObject<{
     file_name: z.ZodString;
     content: z.ZodString;

package/dist/controller/views.js

@@ -64,6 +64,13 @@ export const UploadFileActionSchema = z.object({
 export const ScreenshotActionSchema = z.object({
     file_name: z.string().optional(),
 });
+export const SaveAsPdfActionSchema = z.object({
+    file_name: z.string().optional(),
+    print_background: z.boolean().default(true),
+    landscape: z.boolean().default(false),
+    scale: z.number().min(0.1).max(2.0).default(1.0),
+    paper_format: z.string().default('Letter'),
+});
 export const EvaluateActionSchema = z.object({
     code: z.string(),
 });
@@ -75,6 +82,7 @@ export const ExtractStructuredDataActionSchema = z.object({
     extract_links: z.boolean().default(false),
     start_from_char: z.number().int().default(0),
     output_schema: z.record(z.string(), z.unknown()).nullable().optional(),
+    already_collected: z.array(z.string()).default([]),
 });
 export const SearchPageActionSchema = z.object({
     pattern: z.string(),
@@ -93,11 +101,6 @@ export const FindElementsActionSchema = z.object({
 export const ReadFileActionSchema = z.object({
     file_name: z.string(),
 });
-export const ReadLongContentActionSchema = z.object({
-    goal: z.string(),
-    source: z.string().default('page'),
-    context: z.string().default(''),
-});
 export const WriteFileActionSchema = z.object({
     file_name: z.string(),
     content: z.string(),

package/dist/dom/dom_tree/index.js

@@ -343,17 +343,29 @@
     }
   }
 
-  // // Add this function to perform cleanup when needed
-  // function cleanupHighlights() {
-  //   if (window._highlightCleanupFunctions && window._highlightCleanupFunctions.length) {
-  //     window._highlightCleanupFunctions.forEach(fn => fn());
-  //     window._highlightCleanupFunctions = [];
-  //   }
-
-  //   // Also remove the container
-  //   const container = document.getElementById(HIGHLIGHT_CONTAINER_ID);
-  //   if (container) container.remove();
-  // }
+  function cleanupHighlights() {
+    try {
+      const cleanupFns = Array.isArray(window._highlightCleanupFunctions)
+        ? window._highlightCleanupFunctions
+        : [];
+      for (const fn of cleanupFns) {
+        try {
+          if (typeof fn === 'function') {
+            fn();
+          }
+        } catch (error) {
+          // Ignore cleanup callback failures to keep extraction resilient.
+        }
+      }
+      window._highlightCleanupFunctions = [];
+      const container = document.getElementById(HIGHLIGHT_CONTAINER_ID);
+      if (container) {
+        container.remove();
+      }
+    } catch (error) {
+      // Ignore cleanup failures and continue with DOM extraction.
+    }
+  }
 
   /**
    * Gets the position of an element in its parent.
@@ -1391,6 +1403,7 @@
     return id;
   }
 
+  cleanupHighlights();
   const rootId = buildDomTree(document.body);
 
   // Clear the cache before starting

package/dist/filesystem/file-system.js

@@ -638,7 +638,7 @@ export class FileSystem {
                         truncationNote =
                             `\n\n[Showing ${pagesIncluded.length} of ${numPages} pages. ` +
                                 `Skipped pages: [${skippedPreview}${skippedSuffix}]. ` +
-                                'Use read_long_content with a specific goal to find relevant sections.]';
+                                'Use extract with start_from_char to read further into the file.]';
                     }
                     result.message =
                         `Read from file ${filename} (${numPages} pages, ${totalChars.toLocaleString()} chars total).\n` +