PDF goto returns 500; Bing accessibility tree unavailable

[Liqq1]

Hi @tqfang
Thanks for the great open-source agent framework! I’ve run into two issues while using it and would appreciate your help.

1) goto to PDF URLs returns 500
When the agent does goto https://…/file.pdf, I get:
Request failed with status code: 500 {"error":"Failed to perform action."}
Manually opening these links works. Is this a known limitation or a Playwright/browser setting issue?

2) Bing accessibility tree is unavailable
On bing.com, I often see:
**Warning**: The accessibility tree is currently unavailable.
Other sites expose accessibility tree normally. Is this expected on Bing? Any tips to handle it?

[tqfang]

(1) The error here does not happen all the time but we do have a minor fix that has not been released in the performAction function of server.js:

      case 'goto':
            try {
                const urlToGoto = actionValue;
                
                let pathname;
                try {
                    const urlObj = new URL(urlToGoto);
                    pathname = urlObj.pathname.toLowerCase();
                } catch {
                    pathname = urlToGoto.toLowerCase();
                }
                
                const downloadableExtensions = ['.pdf', '.doc', '.docx', '.xls', '.xlsx', '.zip', '.rar', '.txt', '.csv'];
                const hasDownloadableExtension = downloadableExtensions.some(ext => {
                    const lastPart = pathname.split('/').pop() || '';
                    return lastPart.endsWith(ext) || lastPart.includes(ext + '?') || lastPart.includes(ext + '#');
                });
                
                const urlLower = urlToGoto.toLowerCase();
                const pdfUrlPatterns = [
                    '/pdf/',
                    'arxiv.org/pdf',
                    '/download/',
                    '/export?format=pdf',
                    '/fulltext',
                    'viewdoc/download',
                    '/pdfviewer/',
                    '/get_pdf/',
                    '/downloadpdf/',
                    '/getpdf',
                    '/fetch.php',
                    '/download.php',
                    '/file.php'
                ];
                const isPdfUrl = pdfUrlPatterns.some(pattern => urlLower.includes(pattern));
                
                const shouldDownload = hasDownloadableExtension || isPdfUrl;
                
                if (shouldDownload) {
                    console.log('Detected downloadable URL, using fetch to download:', urlToGoto);
                    
                    const downloadResult = await page.evaluate(async (url) => {
                        try {
                            const response = await fetch(url);
                            const blob = await response.blob();
                            
                            const blobUrl = URL.createObjectURL(blob);
                            const a = document.createElement('a');
                            a.href = blobUrl;
                            
                            const contentDisposition = response.headers.get('content-disposition');
                            let filename = 'download';
                            if (contentDisposition) {
                                const filenameMatch = contentDisposition.match(/filename[^;=\n]*=((['"]).*?\2|[^;\n]*)/);
                                if (filenameMatch && filenameMatch[1]) {
                                    filename = filenameMatch[1].replace(/['"]/g, '');
                                }
                            } else {
                                filename = url.split('/').pop().split('?')[0] || 'download';
                            }
                            
                            a.download = filename;
                            a.style.display = 'none';
                            document.body.appendChild(a);
                            a.click();
                            

                            setTimeout(() => {
                                document.body.removeChild(a);
                                URL.revokeObjectURL(blobUrl);
                            }, 100);
                            
                            return { success: true, filename };
                        } catch (e) {
                            return { success: false, error: e.message };
                        }
                    }, urlToGoto);
                    
                    if (downloadResult.success) {
                        console.log('File downloaded via fetch:', downloadResult.filename);
                        await page.waitForTimeout(3000);
                    } else {
                        console.log('Fetch download failed:', downloadResult.error);

                        await page.goto(urlToGoto, { timeout: 60000, waitUntil: "domcontentloaded" });
                    }
                    
                } else {
                    await page.goto(urlToGoto, { timeout: 60000, waitUntil: "networkidle" });
                }
                
            }

(2) This is not normal. Seem to be some anti-scraping techniques applied on your machine/IP. Suggestions would be changing the default browser search engines to some websites less likely to be blocked such as duckduckgo.com. Or you could simply rely on google search/bing search API.

[Liqq1]

thx!