Why Go with the Browser's Native Print?

If you're doing everything on the frontend (no backend), there are basically three ways to generate a PDF:

1. Server-side rendering (Puppeteer/Playwright) — not an option for static sites
2. Client-side JS generation (html2pdf.js, jsPDF + html2canvas) — outputs a bitmap image
3. Browser native print (window.print()) — gives you a real vector PDF, and the browser handles page breaks

I started with html2pdf.js, but quickly hit some major pain points:

• Bitmap output: The PDF is essentially a screenshot — you can't select or search any text
• Messy page breaks: Long code blocks and nested lists get chopped right in the middle (this was the dealbreaker)
• Performance issues: Long documents need a ton of Canvas rendering, which can easily freeze mobile devices

After switching to the browser's native print, the text-chopping problem was completely gone. Plus, text is selectable and searchable, links still work, file sizes are smaller, and page breaks are way more reliable.

The Core Setup: Overlay + iframe

If you just call window.print() directly, it'll print your entire page UI — buttons, nav, everything. So we use an iframe to isolate the content. Here's how it works:

1. User clicks "Download PDF"
2. A fullscreen overlay with a toolbar pops up
3. Print styles are generated from a template
4. The styled HTML gets written into the iframe
5. User hits print, which triggers iframe.contentWindow.print()

This way, you get a nice preview without messing up the main page's styles.

Key CSS Rules for Page Breaks

@media print {
  tr, pre, blockquote, img, .katex-display {
    page-break-inside: avoid;
    break-inside: avoid;
  }
  h1, h2, h3, h4, h5, h6 {
    page-break-after: avoid;
    break-after: avoid;
  }
  table { page-break-inside: auto; }
  thead { display: table-header-group; }
  body {
    -webkit-print-color-adjust: exact;
    print-color-adjust: exact;
  }
}

These rules prevent content from getting cut off, repeat table headers on every page, and preserve background colors.

Guiding Users Through the Print Dialog

It's a good idea to show users a quick tip in the UI telling them how to set things up:

1. Destination: Save as PDF
2. Background graphics: Check this on
3. Headers and footers: Turn these off
4. Paper size: A4 or Letter
5. Margins: Default

Wrapping Up

As long as you're okay with the print dialog popping up, the browser's native print is hands down the most stable, hassle-free, and highest-quality option out there. If you've got a better approach or any questions, feel free to leave a comment!

Related Links:

• Frontend-Only Markdown to PDF: From html2pdf.js to Browser Native Print
• md-to.com Markdown to PDF Tool
• Markdown to PDF User Guide

QClaw official website: https://claw.guanjia.qq.com/

Mac Download Links

macOS (Apple Silicon) https://cdn.m.qq.com/qclaw/arm/5001/QClaw-latest-arm64.dmg macOS (Intel) https://cdn.m.qq.com/qclaw/intel/5001/QClaw-latest-x64.dmg

Windows Download

QClaw Windows: https://cdn.m.qq.com/qclaw/win/5001/QClaw-Setup-latest.exe

2. How to Get a QClaw Invite Code

1. Fill out the official application form from Tencent: QClaw invite code application: https://wj.qq.com/s2/26010208/ltnx/ (new application link)

https://wj.qq.com/s2/25871229/abe7/ (old application link)

1. Join the beta testing group on WeCom (Enterprise WeChat).

3. QClaw Usage FAQ

1. What if the invite code doesn't work? Double-check for easily confused characters such as I and l, which can look identical on mobile devices. Try swapping the character, or copy-paste the code directly.

Codex APP for Mac

You can get it here: https://chatgpt.com/codex/get-started Or: https://persistent.oaistatic.com/codex-app-prod/Codex.dmg

Codex APP for Windows

https://apps.microsoft.com/detail/9plm9xgg6vks

2. Which ChatGPT plans support Codex APP? Can I try it for free?

Codex is available for ChatGPT Plus ($20/month) and Pro ($200/month) subscribers.

Good news! Starting February 3rd, there's a free trial going on — even Free ($0) and Go ($8/month) users can use Codex for free.

3. Error when launching the Codex app on macOS: You can’t open the application “Codex” because this Mac does not support it.

Reason: The current version of the Codex app does not support Intel-based Macs.

Solution: Wait for an official Intel Mac version, or use the Codex CLI or the Codex IDE plugin instead.

4. Error: "Your access token could not be refreshed"

Easy fix: Just log out and log back in. That should do the trick.

5. How do I check my Codex APP usage?

Head over to this page: https://chatgpt.com/codex/settings/usage

My experience and resolution process:

1. About 1-2 hours after domain registration, DNS resolution, and server binding, I submitted the site to Bing. The URL inspection returned a DNS issue (arrow 1)
2. Clicked "Live URL", and the test page displayed normally. Then directly clicked the "Request Indexing" button (arrow 2)
3. When I checked again at 1 PM, it showed "Successfully indexed".

2. Alert: Discovered but not crawled

Solution: First resolve any discovered SEO issues, then submit an appeal.

3. Bing Webmaster Support Error: No Data Available

When accessing the Bing Webmaster support page (https://www.bing.com/webmasters/help/webmasters-support-24ab5ebf), the page appears blank with a red bar at the bottom displaying "No Data Available". I suspect Bing may have made an error with this URL during a website redesign. Solution: Visit the correct feedback page: https://www.bing.com/webmasters/support How to fill out the feedback form? See item 4 in this article

Agent Skills currently supports the following code editors:

• Claude Code
• VS Code (with GitHub Copilot)
• ChatGPT Codex
• Cursor
• Google Antigravity（from v1.14.2 supports）

How to Install Agent Skills (Installation Commands)

The general installation command format is:

npx skills-installer install <skill-name> --client <client-name>

Here's an example using @anthropics/claude-code/frontend-design:

1. Installing a Skill in Claude Code

npx skills-installer install @anthropics/claude-code/frontend-design --client claude-code 

2. Installing a Skill in Cursor

npx skills-installer install @anthropics/claude-code/frontend-design --client cursor

3. Installing a Skill in ChatGPT Codex (using VS Code Codex extension or Codex CLI)

npx skills-installer install @anthropics/claude-code/frontend-design --client codex

4. How to install a Skill in Google Antigravity

Simply copy the skill directory to the ~\.gemini\antigravity\skills directory.

Troubleshooting: "npx skills-installer: command not found" Error

This error usually occurs when copying the command npx skills-installer - there might be an issue with the space between the words. Simply delete the space and type it again manually.

How to Check if Skills Are Installed Successfully?

Simply ask: "What skills have been installed?"

How to Use Skills?

Skills are typically triggered when related keywords are mentioned in conversation. You can also directly use the skill name, for example: "Please use the frontend-design skill to help me redesign the website style."

1. First, install the VS Code Chinese language extension. Simply search for "Chinese" in the Extension search bar, and you can choose to install Simplified Chinese or Traditional Chinese, as shown below:

1. Press Ctrl + Shift + P (Windows) or Command + Shift + P (Mac), search for "language", click Configure Display Language, then click "Simplified Chinese" or "Traditional Chinese" (as shown below). Next, click the "Restart" button, and after restarting Google Antigravity, it will display in Chinese. If you want to switch back to English, follow the same steps.

2. Set Google Antigravity to Use Chinese Responses Step by Step

The main approach is to set rules for using Chinese responses in the global rules. The steps are as follows:

1. Click the three-dot icon in the upper right corner of the Agent chat window;
2. Then click Customizations;
3. Click "+ Global", which will generate a GEMINI.md file in the .gemini/ directory under the user directory;
4. Close the Global Antigravity Rules File prompt;
5. Write the rules directly in the GEMINI.md file using plain text, such as "Agent should respond in Simplified Chinese", then save it;
6. After saving, you can see the edited Google Antigravity global rules content in the Rules section;
7. Finally, Lanxk entered "Please help check if the rules have been applied?" in the Agent dialog box, and Google Antigravity replied: I have confirmed that these rules have been applied.

Cursor has a free plan that includes a 7-day Pro trial. After the trial ends, it limits the number of Agent requests and Tab completions. Cursor's paid plans for individuals include: Pro ($20), Pro+ ($60), and Ultra ($200). The main difference between these plans is the usage limits for OpenAI, Claude, and Gemini models.

2. Which AI Model Works Best in Cursor?

For coding, I personally recommend Claude Sonnet 4.5 (Thinking). It produces high-quality code and is very efficient at fixing bugs. In comparison, Claude Opus 4.5 is too expensive, and the improvement isn't particularly noticeable at the moment.

3. How to Set Up Chinese Language in Cursor?

First, install the VS Code language extension. Search for "Chinese" in the Extension search bar, and you can choose to install either Simplified Chinese or Traditional Chinese, as shown below:

Press Ctrl + Shift + P, search for "Language", click "Configure Display Language", then select "中文（简体）" or "中文（繁体）" (as shown below). Next, click the "Restart" button. After restarting Cursor, it will display in Chinese. To switch back to English, follow the same steps.

4. Which is Better: Cursor VS GitHub Copilot?

I personally choose Cursor. I had previously paid for an annual GitHub Copilot subscription, but for the past few months, I've been using Cursor almost exclusively and rarely use GitHub Copilot anymore. The main reason I chose Cursor initially was that its Tab completion is excellent and far superior to GitHub Copilot. Later, Cursor introduced many new features that have left GitHub Copilot far behind.

5. Which is Better: Cursor VS ChatGPT Codex?

I use both Cursor and Codex. Most of the time, I use Cursor because it's fast, convenient, and handles most problems well enough. When I encounter more complex issues or problems that Cursor can't solve, I turn to Codex. Also, when creating a new project, I prefer to use Codex first because it produces better UI. The downside of Codex is that it's too slow, so it's not ideal for handling simple problems.

6. Which is Better: Cursor VS Google Antigravity?

I think each has its own strengths. After using both for a while, here's what I've noticed:

1. Using Gemini 3 Pro in Google Antigravity works better than using Gemini 3 Pro in Cursor;
2. Using Claude Sonnet 4.5 and Claude Opus 4.5 in Cursor works better than using them in Google Antigravity;
3. Google Antigravity can call Nano Banana to generate images, which is a feature that Cursor lacks.

7. Cursor Error: Model not available. This model provider doesn't serve your region.

For the solution, please see here: Model not available. This model provider doesn't serve your region

8. Cursor Error: Unauthorized request. User is unauthorized

This error occurs because your computer's system time is incorrect. Solution: Adjust your computer's date and time settings to the correct values (you may need to restart the Cursor client).

1. Open the Beijing etax website and navigate to the login page (as shown below). Select "电子营业执照", then scan the QR code with WeChat on your phone. After scanning successfully, the "电子营业执照" mini-program will ask you to select the electronic license for your business. After selection, it will prompt you to enter the "执照密码". If you haven't changed the password, the default password is "123456". After entering, click "确认", then an "授权登录" page will pop up. Click "确认登录" again, wait a few seconds, and the mini-program will show "授权成功". Wait a few more seconds, and the webpage will automatically log in successfully.

2. After successful login, you will be redirected to this page: National Tax Administration Electronic Tax Bureau. Click "社保费业务" directly (as shown below), or visit directly: https://etax.beijing.chinatax.gov.cn:8443/xxzx/view/nsrxxzx/sbf .

3. Click "日常申报" on the right.

4. On the "日常申报" page, all fees to be paid will be displayed. Check all of them first, then click "提交申报".

5. A prompt box will pop up. Click "立即获取", or wait about ten seconds for it to automatically redirect.

6. If the declaration record is blank at this point (as shown in the image), click "查询" again.

7. After clicking "查询", the recent declaration records will be displayed. Find the declaration record that needs payment and click "缴款" behind it (if you only see "作废" but not "缴款", scroll the webpage to the right), then click "缴款" (Figure 7).

8. On the "缴费" page, check all items that need to be paid, then click "立即缴费".

9. Click the "确认" button (Figure 9).

10. In the "账户信息" section, I don't select anything here, just click "确认" directly.

11. Next, you will be redirected to the "银行端查询缴税凭证" page. Copy these three numbers for later use: (1)银行端查询缴税凭证序号 (2)纳税人识别号 (3)税务机关代码. If you can't find this page, you can go to the Payment Certificate Print Record page and click "重新打印".

12. I usually copy these three items and send them to "文件传输助手" on WeChat.

Next, go to the bank counter or use the bank client on your computer/mobile phone to make the payment.

2. Detailed Steps for Paying Social Insurance Fees via Mobile Banking Tax Payment

Here I'll use the "招商银行" mobile APP as an example:

1. Search for "缴税通" and click to enter the "缴税通" page.

2. As shown in the image below, enter the corresponding (1)银行端查询缴税凭证序号 (2)纳税人识别号 (3)税务机关代码 according to the prompts in the image, then click the "查询" button.

3. After confirming everything is correct, scroll to the bottom of the page and click "缴费".

4. After completing the payment in the mobile APP, return to the webpage, open the "申报记录" page (https://sbf.beijing.chinatax.gov.cn/dwsbf-web/#/dwsbjlcx ), click "查询", and you can see that this declaration has been successfully paid (the "缴款状态" has changed to "已缴费").

3. Common Questions About Social Insurance Payment

1. Payment Time (Prompt: Not Within Declaration Time)

Pay the previous month's social insurance fees from the 10th to the 25th of each month (including both the 10th and 25th). For example, pay November's social insurance in December. If you declare before the 10th or after the 25th, you will be prompted that it's not within the declaration time.

2. Can I Pay Social Insurance Without Installing the Social Insurance Management Client?

Paying social insurance on the Electronic Tax Bureau website and paying through the Social Insurance Management Client are actually the same thing. You can choose either one (if you have special requirements, please choose according to your needs). Similar question: Can Beijing individual business owners pay social insurance themselves without using the social insurance payment client, only using the Electronic Tax Bureau website? (Yes, refer to the answer above)

3. Prompt: "您尚未进行社保费缴费信息关联，请与主管税务机关联系。"

If you just registered as an individual business owner this month, don't worry. It usually automatically associates on the 10th of next month. Wait until the 10th of next month and log in to the Tax Bureau website again to check. If you're not a newly registered individual business owner, contact your local tax authority to ask what's going on.

4. What to Do If "缴费记录" Shows Empty?

There are two solutions: Method 1: Come back another day to query again. If you're lucky, it will show up (or it might still not display). Method 2: Visit the "单位缴费记录打印" page ( https://sbf.beijing.chinatax.gov.cn/dwsbf-web/#/dwjfzm ), click "查询", wait one or two minutes, and the payment record will appear.

5. Beijing Social Insurance Payment Prompt: Unit Payment Data Query Failed, Failure Reason [No Valid Summary Preprocessing Information]

Reason 1: It's now December 2025. If you're an existing user who has already paid social insurance last month, and you encounter this problem today, it's likely due to too many people accessing the website (this problem usually occurs most easily on the 10th-12th of each month). Solution: Try again at a different time.

Reason 2: If this is your first payment, you might need to perform "年度缴费工资调整".

6. What to Do If I Forgot (Can't Find) the Bank-side Tax Payment Certificate Serial Number?

You can go to the Payment Certificate Print Record page and click "重新打印".

Just like Cursor, Google Antigravity also supports Global Rules and Project Rules.

Here's a step-by-step guide on how to add Global Rules (with screenshots):

1. Click the three-dot icon in the top right corner of the Agent chat window;
2. Click on "Customizations";
3. Click "+ Global" - this will create a file called GEMINI.md in the .gemini/ folder under your user directory;
4. Close the "Global Antigravity Rules File" prompt;
5. Write your rules in plain text directly in the GEMINI.md file, then save it;
6. After saving, you can see your updated Global Rules in the Rules section;
7. To verify, type something like "Please check if the rules are applied?" in the Agent chat. Google Antigravity should confirm that the rules have been applied.

2. What Are the Common Global Rules?

Here are some useful Global Rules you might want to set:

1. Agent should reply in your preferred language (e.g., Simplified Chinese);
2. Code comments should be written in your preferred language;
3. UI text should follow your project's i18n settings (or use a default language if not specified).

3. Recommended Project Rules for Google Antigravity

You can easily convert your existing Cursor project rules to Google Antigravity format. For example, I've converted my Cursor project rules for Vue 3 frontend development into Google Antigravity project rules.

Download link: https://www.lanxk.com/files/GEMINI-project-rules-for-vue3-en.md

After downloading, rename the file to GEMINI.md and place it in the .gemini/ folder at your project root.

Google Antigravity is currently free to use. You can use the following large models for free:

• Gemini 3 Pro (High)
• Gemini 3 Pro (Low)
• Claude Sonnet 4.5
• Claude Sonnet 4.5 (Thinking)
• Claude Opus 4.5 (Thinking)
• GPT-OSS 120B (Medium)

2. Which model works best in Google Antigravity?

Personally, I recommend Gemini 3 Pro (High). Its UI generation quality is excellent.

3. Google Antigravity vs Cursor: which one is better?

In my opinion, each has its own strengths. You can try both and see which one fits you better. After using them for a while, my feeling is:

1. Using Gemini 3 Pro inside Google Antigravity gives better results than using Gemini 3 Pro inside Cursor.
2. Using Claude Sonnet 4.5 and Claude Opus 4.5 inside Cursor works slightly better than using Claude Sonnet 4.5 and Claude Opus 4.5 inside Google Antigravity.

4. How do I switch Google Antigravity to Simplified Chinese or Traditional Chinese?

1. First, install the VS Code language pack extension. In the Extensions search box, search for "Chinese", then choose and install either Simplified Chinese or Traditional Chinese, as shown below:

1. Press Ctrl + Shift + P, choose Configure Display Language, then select “中文（简体）” or “中文（繁体）” (as shown below). After that, click the Restart button. Once Google Antigravity restarts, the UI will be in Chinese.
   To switch back to English, repeat the same steps and choose English.

5. How to fix Google Antigravity login issues?

Error: region not supported

Your current account is not eligible for Antigravity, because it is not currently available in your location.

Error: account not eligible

Your current account is not eligible for Antigravity. Try signing in with another personal Google account. Learn more by visiting the FAQ.

Message: setting up your account

Setting Up Your Account

For detailed solutions, see this article (Chinese):
Resolved Google Antigravity Sign In Issues

6. Where can I download Google Antigravity? (Official link)

Official download link:

https://antigravity.google/download

7. Google Antigravity Agent shows an empty model list (No Model Selected) and says: Select Model to Send Message Select a model using the model selector in the input box

This usually happens when you have successfully signed in before, and then restart the client later.

The fix is simple: set your VPN to a supported region and enable TUN mode, then restart the Google Antigravity client. After that, the model list should load normally.

8. Error: Your current account is not eligible for antigravity. To use antigravity you must be 18 years old or older. If you think you are receiving this message in error, please ensure you have verified your age and try to log in again. Learn more by visiting the faq.

Reason: Google requires users to be at least 18 years old to use Antigravity. Solution: Visit the Google Account Personal Information page, click Birthday, change it to a date before December 2007, then log in to the Google Antigravity client again.

9. How to Set Up Antigravity Global Rules and Project Rules?

For detailed setup steps, please refer to this article: How to Add Rules to Google Antigravity?

10. Antigravity Prompt: Agent terminated due to error

You can prompt the model to try again or start a new conversation if the error persists. See our troubleshooting guide for more help.

Solutions:

1. Click Retry to try again (the success rate is not very high, and errors will likely continue);
2. Try switching models, for example, change from Claude Opus 4.5 (Thinking) to Claude Sonnet 4.5 (Thinking), or try Gemini 3 Pro (High) or Gemini 3 Flash;
3. If it still doesn't work, click New Conversation to start a new conversation and begin again;
4. Try at a different time, for example, wait for the 5-hour quota to refresh and try again.

11. Antigravity Prompt: Model quota limit exceeded

You have reached the quota limit for Claude Opus 4.5 (Thinking). You can resume using this model at 1/13/2026,1:28:06 PM. You can upgrade to the Google AI UItra plan to receive the highest rate limits.

Reason: Reached the quota limit (weekly) for Claude Opus 4.5 (Thinking).

Solutions:

1. Wait for the quota to update;
2. Upgrade to the Google AI UItra plan to receive the highest rate limits.
3. Try a different model;
4. Try a different account.

After spending a day changing my Google account region from Hong Kong to Taiwan, I finally managed to sign in to Antigravity. Steps I used:

1. Change your Google account's region to a supported region (for example: United States, Japan, Taiwan — detailed steps below);
2. Switch your VPN to a supported region (preferably the same one I used) and enable TUN mode (global mode is not required);
3. After the region change is processed (you'll receive an email notification), sign in again. Good luck!

Error: Your current account is not eligible for Antigravity, because it is not currently available in your location.

This means the Google account you're using is not in a supported region. The solution is to change the account's associated region.

How to Change Your Google Account's Associated Region

Visit Google's "Request to Change Associated Region" page: https://policies.google.com/country-association-form

1. First check the current associated region — if it's already in a supported region, you don't need to change it;
2. Choose the target region you want to change to (for example: United States, Japan);
3. Select a reason (recommended: avoid saying you're using a VPN);
4. Submit the request (this requires manual review, you will receive an email notification in about an hour).

Error: Your current account is not eligible for Antigravity. Try signing in with another personal Google account. Learn more by visiting the FAQ.

This “Your current account is not eligible for Antigravity” error happens because of the Google account you’re using. Common reasons:

1. You’re using a student account.
2. You’re using a organization account.
3. The birthday on your personal account shows you are under 18.

How to fix the “Your current account is not eligible for Antigravity” error:

• If it’s reason 1 or 2, sign out and sign in with a personal Google account.
• If it’s reason 3, try updating the birthday in your Google account settings.

Error: Your current account is not eligible for antigravity. To use antigravity you must be 18 years old or older. If you think you are receiving this message in error, please ensure you have verified your age and try to log in again. Learn more by visiting the faq.

Reason: Google requires users to be at least 18 years old to use Antigravity. Solution: Visit the Google Account Personal Information page, click Birthday, change it to a date before December 2007, then log in to the Google Antigravity client again.

Error: There was an unexpected issue setting up your account. We apologize for the inconvenience. Please try again later.

An unexpected issue occurred during account setup. The exact cause is unknown. Here are some solutions you can try:

1. Wait a while and try logging in again;
2. Switch to a different VPN location and log in again;
3. Try logging in with a different account.

Google Antigravity Download

https://antigravity.google/download

----------The original text below — optional reading-------

Google released Antigravity, an AI programming tool. After downloading and installing, I followed the setup prompts and successfully signed in, which redirected me to https://antigravity.google/auth-success. However, when returning to the Antigravity window, it kept showing "Setting Up Your Account", like the image below:

Members in a WeChat group also reported the same issue. I checked domestic and international tech forums and found many users encountering this. Suggested fixes varied — some said to use a different account, others suggested changing IP addresses. I tried several approaches but didn't find a definitive fix yet.

Hope Google resolves this soon so everyone can try Antigravity (and use Gemini 3 Pro within Antigravity).

The Cursor team is working so hard that they're still updating on Saturday, and now it's been updated to v2.0.9(It's been updated to v2.0.34 on October 29th).

Cursor v2.0 Changelog:

The most obvious update is that you can now select multiple models in the chat container (previously you could only select one), and you can choose 1x/2x/3x/4x.

After selecting multiple models, you can let them compete and provide multiple results. If you select 2x after Models, that Model will produce 2 results. Below are the 3 results after selecting claude-4.5-sonnet 1x and gpt-5 2x:

There's also a voice input button. If you speak in a non-English language, it will automatically translate to English text in the input box (I tried speaking Chinese and it translated to English, not sure if other languages can be recognized). The recognition quality is quite good.

Layout (Agent Layout & Editor Layout)

Editor is the original UI. A new Agent UI has been added. After switching, only the Agent-related dialog is displayed, as shown in the following image:

Cursor v2.0 Download

Cursor Mac v2.0.34 Download Links

1. darwin-universal

2. darwin-x64

3. darwin-arm64

Cursor Windows v2.0.34 Download Links

1. win32-x64-user

2. win32-arm64-user

3. win32-x64-system

4. win32-arm64-system

Cursor Linux v2.0.34 Download Links

1. linux-x64

2. linux-arm64

When I'm debugging websites, I constantly run into situations where I need to format JSON data to make it readable. You know how it is - you're working on a project and suddenly you've got this messy JSON response that's impossible to read.

For years, my workflow was pretty straightforward: open up DevTools in Edge or Chrome, go to the Network tab, copy the JSON response, then open VS Code, create a temp file, paste it in, and format it. It was a bit tedious, but I got used to it and it worked fine.

But then I hit a wall. I started dealing with JSON strings that needed unescaping, and my usual routine just wasn't cutting it anymore. I'd copy the JSON into VS Code, try to format it, and... nothing. I'd have to find some online tool to unescape it first, then go back to VS Code to format it.

The problem was that most online JSON formatters are pretty unforgiving. If your JSON isn't perfectly formatted, they just give up. I'd spend way too much time hunting down syntax errors, fixing them, and then trying again.

So I thought, "Why not just build my own tool that handles all this stuff?" And that's exactly what I did. I built this JSON formatter website.

I built in all the features I wished existed: automatic unescaping, removing those annoying quotes at the beginning and end (both single and double), and tons of error tolerance - like allowing unquoted keys, single quotes, and trailing commas. Basically, I made it handle all the messy JSON that other tools reject.

How to Use It

It's dead simple. Just copy your messy JSON data, click the "Paste" button (or use the keyboard shortcut), and boom - it automatically unescapes, cleans up quotes, and formats everything into beautiful, readable JSON.

Try It Out (Available in 4 Languages)

• English: https://jsonformatter.lanxk.com/
• Simplified Chinese: https://jsonformatter.lanxk.com/zh-cn/
• Traditional Chinese: https://jsonformatter.lanxk.com/zh-tw/
• Japanese: https://jsonformatter.lanxk.com/ja/

If you run into any issues or have feature requests, drop me a line!

1. Jun 21 Discovered that site: lanxk.com returned zero results on Bing. Yikes.

2. Around 2 AM on Jun 22 Filed the first appeal. Got an auto-reply from Copilot saying there were SEO issues. Chose to submit the form anyway. Received the system confirmation email.

3. Jun 22 – Jul 14 Used the waiting time to audit the site in Bing Webmaster Tools. Found and fixed most SEO issues (from 83 errors down to a little over 20).

4. Jul 14 Filed the second appeal. This time, Copilot didn’t flag SEO issues. Got the auto-reply again.

5. Jul 22 (about 8 days later) Received the first human response. They said the site was sent to the product review team for further evaluation. They expected the review to take a few weeks.

6. Aug 19 (about 4 weeks later) Got the second email: good news — the issue was resolved!

7. Total time From discovery to final resolution: about 2 months From the first appeal to the second appeal: 3 weeks From the second appeal to resolution: about 5 weeks

The process: what actually happened

On Jun 21, I noticed that searching site: lanxk.com on Bing showed no results at all (both the CN and the international versions). Not a fun surprise:

So around 2 AM on Jun 22, I started the appeal process. This time felt a bit different: Copilot popped up first with an automated message saying my site had some SEO issues, like this:

I chose to submit the form anyway and then got an automated confirmation email.

No news for the next few days, so I dove into Bing Webmaster Tools to see what I could fix on my end.

URL Inspection surfaced several SEO issues:

The Recommendations page showed 83 SEO errors:

Since I was in the waiting game anyway, I spent some time fixing most of the flagged issues:

After waiting 3 weeks without a response, I submitted another appeal on Jul 14. This time, Copilot didn’t flag any SEO issues, which felt more promising:

I immediately got another auto-reply, same as the one on Jun 22.

About 8 days later, I finally received the first email from a human. In short: “We’ve reviewed your site and sent it to our product review team for further evaluation. Bingbot will continue to evaluate it against the Bing Webmaster Guidelines to determine whether it should be indexed. The review usually takes a few weeks, depending on your site’s complexity. We’ll email you once it’s done.”

At that point, there wasn’t much left to do but wait. Those weeks were tough — I’d already fixed what I could. Interestingly, Bing kept crawling and indexing my site (the content was in the index); it just didn’t show up in search results.

After another 4 weeks, on Aug 19, the good news finally landed: the issue was resolved.

A friend's new website Viddo AI - AI Image and Video Generator just launched, and I provided some feedback on SEO and frontend issues along with optimization recommendations. I'm documenting these suggestions on my blog, hoping they might help other developers too. If you spot any areas where I could improve my analysis, feel free to let me know!

SEO Issues and Optimization Recommendations

1. The title tag is too long, exceeding 60 characters (currently 86 characters, recommended to keep it between 50-60).
2. The description is also too long, recommended to keep it between 150-160 characters, currently at 251.
3. Keywords meta tag isn't really necessary anymore, but if you choose to use it, the current one has way too many keywords.
4. Many images are missing alt attributes.
5. Internal links in the footer section could use more descriptive title attributes. Don't just repeat the anchor text - the title should provide additional context. Think about what would entice users to click that link.

<a class="cursor-pointer" title="Veo3" href="/veo3?model=veo3">Veo3</a>


<a class="cursor-pointer" title="Use model Veo3 generate video" href="/veo3?model=veo3">Veo3</a>
<a class="cursor-pointer" title="Generate video by model Veo3" href="/veo3?model=veo3">Veo3</a>

1. Missing hreflang attributes for internationalization.
2. Missing structured data markup. I'd recommend using "@type": "Organization" schema. Reference: Google Search supported structured data
3. Top navigation internal links are missing title attributes.
4. Tool pages have great long-form content at the bottom - that's excellent! Consider adding the homepage footer there too, and naturally incorporate links to other pages within that content.
5. Blog page URLs (like /gpt-oss) could be a bit longer with more keywords. Also, consider adding some internal links to tool pages within the blog content.

UI/UX and Frontend Issues & Recommendations

1. Change "Log In" to "Sign In" (Google uses "Sign in with Google").

2. This grayed-out icon shouldn't scale on hover anymore. The cursor should be default (cursor: default) or not-allowed (cursor: not-allowed).

3. For images, only show pointer cursor (cursor: pointer) on draggable areas. Use default cursor (cursor: default) everywhere else to avoid confusing users who think they can click but nothing happens. Other images have this same issue - users expect something to happen when they click, but nothing does.

4. If the sound icon doesn't actually play sound, consider hiding it. Users will feel disappointed when they enable sound and nothing happens.

5. The images in the blog homepage list are way too large in both file size and dimensions. This is unnecessary and causes two problems: slow page loading (bad UX) and wasted bandwidth costs.

6. The blog detail pages could use some work. Consider writing more articles about your website tools and cross-linking them with the tool pages.

7. In Edge browser, the Google login popup in the top-right corner has positioning issues.

8. Consider removing the purple background from icons and just using white. The current purple background combined with the page's black background looks a bit harsh and uncomfortable.

9. The website loads a bit slowly. Check out PageSpeed recommendations for improvements.

10. Having a button element inside a hyperlink feels a bit odd (though it's not invalid). Consider just adding a class to the a tag and styling it as a button instead.

Overall, this website is already really impressive! I'd encourage everyone to try out Viddo AI and create some beautiful images and videos.

Cursor v1.5 changelog:

Added a Toggle AI Pane icon on the right side of the top menu bar

Startup page now displays Chat sidebar (see comparison screenshots of v1.4 and v1.5 startup pages below)

Cursor settings page's "Tools & Integrations" changed to "MCP & Integrations"

Cursor v1.5 Download

Cursor Mac v1.5.5 Download Links

1. darwin-universal

2. darwin-x64

3. darwin-arm64

Cursor Windows v1.5.5 Download Links

1. win32-x64-user

2. win32-arm64-user

3. win32-x64-system

4. win32-arm64-system

Cursor Linux v1.5.5 Download Links

1. linux-x64

2. linux-arm64

What's the difference between the 8 GPT-5 variants in Cursor?

• GPT-5: Balanced capability and speed. A solid default for most tasks.
• GPT-5-fast: Similar capability to GPT-5 but faster. Great for chats, short Q&A, and iterative dev work.
• GPT-5-high: Deeper reasoning and higher reliability for complex architecture, algorithms, and long reasoning chains. Slower and more expensive.
• GPT-5-low: Lightweight and cost-efficient for simple rewrites and batch jobs. Cheap, but weaker on detail and robustness.
• GPT-5-high-fast: Strong reasoning with lower latency. Faster than "high," more stable than "fast." Still relatively pricey.
• GPT-5-low-fast: Low cost + higher speed. Good for batch workloads like large-scale generation, log cleanup, and reranking.
• GPT-5-nano: Tiny, very fast, and the cheapest. Best for formatting, extraction, and simple transforms. Not suitable for complex coding or reasoning.
• GPT-5-mini: Between "low" and "nano." Inexpensive and more stable. Handy for small daily tasks and scaffolding.

How to choose

• General coding and Q&A: GPT-5 or GPT-5-fast.
• System design, complex reasoning, strict reviews: GPT-5-high (or GPT-5-high-fast for lower latency).
• Batch processing with good value: GPT-5-low-fast.
• Simple copy and scripts: GPT-5-mini or GPT-5-nano.

Pricing in Cursor

My Cursor Pro plan uses the legacy 500-request quota. In my account, non-fast GPT-5 consumes 2 requests per call, while the fast variants consume 4 requests. If you go over the quota, or you're on the new pricing, the fast variants are billed at 2× the standard price.

There's also a launch promo: during GPT-5's "launch week" (the first week after it goes live), paid users (Pro/Teams/Ultra) can use GPT-5 for free. I tried a few GPT-5 calls just now and they didn't deduct from my 500-request.

Cursor CLI

Bring Agent to any IDE or remote machine.

Today I discovered that Cursor has automatically updated to v1.4.0, and just now I received another update notification. After restarting, I found it had been updated to v1.4.1.

Cursor v1.4 changelog:

Past Chats

Cursor's official team hasn't released the v1.4 changelog yet. Currently, the only difference I've noticed is the addition of a "Past Chats" list at the bottom of the Chat panel on the right side.

However, I've discovered that this feature still has bugs - clicking "View All" doesn't actually load all the historical chat records.

Cursor Mac v1.4.4 Download Links

1. darwin-universal:

https://downloads.cursor.com/production/dd05130a0bf8d5f1a5372e6e6bb7c4184786eaa9/darwin/universal/Cursor-darwin-universal.dmg

1. darwin-x64:

https://downloads.cursor.com/production/dd05130a0bf8d5f1a5372e6e6bb7c4184786eaa9/darwin/x64/Cursor-darwin-x64.dmg

1. darwin-arm64:

https://downloads.cursor.com/production/dd05130a0bf8d5f1a5372e6e6bb7c4184786eaa9/darwin/arm64/Cursor-darwin-arm64.dmg

Cursor Windows v1.4.4 Download Links

1. win32-x64-user:

https://downloads.cursor.com/production/dd05130a0bf8d5f1a5372e6e6bb7c4184786eaa9/win32/x64/user-setup/CursorUserSetup-x64-1.4.4.exe

1. win32-arm64-user:

https://downloads.cursor.com/production/dd05130a0bf8d5f1a5372e6e6bb7c4184786eaa9/win32/arm64/user-setup/CursorUserSetup-arm64-1.4.4.exe

1. win32-x64-system:

https://downloads.cursor.com/production/dd05130a0bf8d5f1a5372e6e6bb7c4184786eaa9/win32/x64/system-setup/CursorSetup-x64-1.4.4.exe

1. win32-arm64-system:

https://downloads.cursor.com/production/dd05130a0bf8d5f1a5372e6e6bb7c4184786eaa9/win32/arm64/system-setup/CursorSetup-arm64-1.4.4.exe

Cursor Linux v1.4.4 Download Links

1. linux-x64:

https://downloads.cursor.com/production/dd05130a0bf8d5f1a5372e6e6bb7c4184786eaa9/linux/x64/Cursor-1.4.4-x86_64.AppImage

1. linux-arm64:

https://downloads.cursor.com/production/dd05130a0bf8d5f1a5372e6e6bb7c4184786eaa9/linux/arm64/Cursor-1.4.4-aarch64.AppImage

What's New in Claude Opus 4.1?

Anthropic has provided an official comparison chart showing how Claude Opus 4.1 stacks up against Claude Opus 4, Claude Sonnet 4, OpenAI o3, and Gemini 2.5 Pro:

As for pricing, Anthropic states that it's the same as Claude Opus 4. So we can expect Cursor's billing method and pricing to remain the same as when selecting Claude-4-Opus.

I just tested it out by using Claude-4.1-Opus (Thinking) to translate this blog post into English. The results were pretty good, and it took about 20 requests to complete.

Cursor just updated v1.1.7 on July 1st, less than 24 hours, and it was automatically updated to v1.2.0 again.

What's new in Cursor v1.2?

At present, Cursor has not officially released the changelog, but from the settings, you can see these 2 options:

1. To-Do List

Allow Agent to use To-Do lists to track progress on tasks

1. Queue Messages When enabled and the model is streaming, subsequent messages will be queued and sent in the order they were received

Cursor Mac v1.2.2 download

1. darwin-universal:

https://downloads.cursor.com/production/faa03b17cce93e8a80b7d62d57f5eda6bb6ab9fa/darwin/universal/Cursor-darwin-universal.dmg

1. darwin-x64:

https://downloads.cursor.com/production/faa03b17cce93e8a80b7d62d57f5eda6bb6ab9fa/darwin/x64/Cursor-darwin-x64.dmg

1. darwin-arm64:

https://downloads.cursor.com/production/faa03b17cce93e8a80b7d62d57f5eda6bb6ab9fa/darwin/arm64/Cursor-darwin-arm64.dmg

Cursor Windows v1.2.2 download

1. win32-x64-user:

https://downloads.cursor.com/production/faa03b17cce93e8a80b7d62d57f5eda6bb6ab9fa/win32/x64/user-setup/CursorUserSetup-x64-1.2.2.exe

1. win32-arm64-user:

https://downloads.cursor.com/production/faa03b17cce93e8a80b7d62d57f5eda6bb6ab9fa/win32/arm64/user-setup/CursorUserSetup-arm64-1.2.2.exe

1. win32-x64-system:

https://downloads.cursor.com/production/faa03b17cce93e8a80b7d62d57f5eda6bb6ab9fa/win32/x64/system-setup/CursorSetup-x64-1.2.2.exe

1. win32-arm64-system:

https://downloads.cursor.com/production/faa03b17cce93e8a80b7d62d57f5eda6bb6ab9fa/win32/arm64/system-setup/CursorSetup-arm64-1.2.2.exe

Cursor Linux v1.2.2 download

1. linux-x64:

https://downloads.cursor.com/production/faa03b17cce93e8a80b7d62d57f5eda6bb6ab9fa/linux/x64/Cursor-1.2.2-x86_64.AppImage

1. linux-arm64:

https://downloads.cursor.com/production/faa03b17cce93e8a80b7d62d57f5eda6bb6ab9fa/linux/arm64/Cursor-1.2.2-aarch64.AppImage

Latest Cursor version: v1.1.3(Update date: 2025-06-15)

Release date: 2025-06-12

What's new in Cursor v1.1.0:

• Slack now supports Background Agents

You can create agents in Slack by just mentioning @Cursor and asking it to make a new Background Agent.

You can also reply to follow-ups and open pull requests right in Slack, without leaving the app.

Cursor Mac v1.1.3 Download

1. darwin-universal:

https://downloads.cursor.com/production/979ba33804ac150108481c14e0b5cb970bda3266/darwin/universal/Cursor-darwin-universal.dmg

1. darwin-x64:

https://downloads.cursor.com/production/979ba33804ac150108481c14e0b5cb970bda3266/darwin/x64/Cursor-darwin-x64.dmg

1. darwin-arm64:

https://downloads.cursor.com/production/979ba33804ac150108481c14e0b5cb970bda3266/darwin/arm64/Cursor-darwin-arm64.dmg

Cursor Windows v1.1.3 Download

1. win32-x64-user:

https://downloads.cursor.com/production/979ba33804ac150108481c14e0b5cb970bda3266/win32/x64/user-setup/CursorUserSetup-x64-1.1.3.exe

1. win32-arm64-user:

https://downloads.cursor.com/production/979ba33804ac150108481c14e0b5cb970bda3266/win32/arm64/user-setup/CursorUserSetup-arm64-1.1.3.exe

1. win32-x64-system:

https://downloads.cursor.com/production/979ba33804ac150108481c14e0b5cb970bda3266/win32/x64/system-setup/CursorSetup-x64-1.1.3.exe

1. win32-arm64-system:

https://downloads.cursor.com/production/979ba33804ac150108481c14e0b5cb970bda3266/win32/arm64/system-setup/CursorSetup-arm64-1.1.3.exe

Cursor Linux v1.1.3 Download

1. linux-x64:

https://downloads.cursor.com/production/979ba33804ac150108481c14e0b5cb970bda3266/linux/x64/Cursor-1.1.3-x86_64.AppImage

1. linux-arm64:

https://downloads.cursor.com/production/979ba33804ac150108481c14e0b5cb970bda3266/linux/arm64/Cursor-1.1.3-aarch64.AppImage

Cursor Version: 0.51.2

Updated: 2025-06-03

What's New in Cursor v0.51?

1. The Cursor Settings page got a major upgrade - now you can see your subscription email and account plan (like Pro Plan) right in the settings.

2. Added Generate Memories feature, but you'll need to turn off privacy mode to use it.

3. New model added: DeepSeek R1 0528

4. And other updates to Network checks, Indexing & Docs, and more.

Mac Installer

1. darwin-universal:

https://downloads.cursor.com/production/f364e608fc11d38303429b80fd1e1f32d7587d43/darwin/universal/Cursor-darwin-universal.dmg

1. darwin-x64:

https://downloads.cursor.com/production/f364e608fc11d38303429b80fd1e1f32d7587d43/darwin/x64/Cursor-darwin-x64.dmg

1. darwin-arm64:

https://downloads.cursor.com/production/f364e608fc11d38303429b80fd1e1f32d7587d43/darwin/arm64/Cursor-darwin-arm64.dmg

Windows Installer

1. win32-x64-user:

https://downloads.cursor.com/production/f364e608fc11d38303429b80fd1e1f32d7587d43/win32/x64/user-setup/CursorUserSetup-x64-0.51.2.exe

1. win32-arm64-user:

https://downloads.cursor.com/production/f364e608fc11d38303429b80fd1e1f32d7587d43/win32/arm64/user-setup/CursorUserSetup-arm64-0.51.2.exe

1. win32-x64-system:

https://downloads.cursor.com/production/f364e608fc11d38303429b80fd1e1f32d7587d43/win32/x64/system-setup/CursorSetup-x64-0.51.2.exe

1. win32-arm64-system:

https://downloads.cursor.com/production/f364e608fc11d38303429b80fd1e1f32d7587d43/win32/arm64/system-setup/CursorSetup-arm64-0.51.2.exe

Linux Installer

1. linux-x64:

https://downloads.cursor.com/production/f364e608fc11d38303429b80fd1e1f32d7587d43/linux/x64/Cursor-0.51.2-x86_64.AppImage

1. linux-arm64:

https://downloads.cursor.com/production/f364e608fc11d38303429b80fd1e1f32d7587d43/linux/arm64/Cursor-0.51.2-aarch64.AppImage

As shown in the image below:

I was originally using Agent mode. When I selected "deepseek-r1-0528" in the Chat, I noticed a message appeared at the bottom of the chat window: Model deepseek-r1-0528 doesn't have good agent support yet. Start a new thread using Manual mode for better results.

I tried both Manual and Agent modes, and both responded pretty quickly with basically the same speed and results.

I feel like deepseek-r1-0528 is much faster than Claude when handling simple questions or tasks with clear instructions. But for complex problems, it's still not as good as Claude Sonnet 3.7 and Claude Sonnet 4.

Recently, after Cursor started supporting Claude 4 Sonnet, many users of the student version and the 14-day trial have reported frequently encountering the issue mentioned above.

However, none of the officially subscribed Cursor Pro users I know have encountered this problem. So I guess Cursor is currently prioritizing access to Claude 4 Sonnet for officially subscribed users and only making it available to student version and 14-day trial users during off-peak times (this is my speculation and might not be accurate).

Related Issues or Error Messages:

1. We're experiencing high demand for Claude 3.5 Sonnet right now. Please upgrade to Pro, switch to the 'auto-select' model, another model, or try again in a few moments.

2. We're experiencing high demand for Claude 4 Opus right now. Please switch to the 'auto-select' model, another model, or try again in a few moments.

Solutions (Recommended: Solution 1 and Solution 3):

1. Switch to the "auto-select" model or other free models.

2. Wait a while or a few hours and try again.

3. Upgrade to Cursor Pro with a paid subscription.

1. It gets dev needs better—no more random or repetitive tweaks like 3.7, fewer unnecessary changes, so it feels faster than 3.7;

2. It can think ahead—spot a bug on a page (like missing null checks) and fix similar issues across the board;

3. Less likely to mess up—used it to optimize a performance issue, and the tweaks were spot-on, barely needed any follow-ups;

4. It auto-generates a PERFORMANCE_OPTIMIZATION.md doc (though it misplaced it) detailing the optimization suggestions, process, code changes, and next steps.