The WeChat Official Account backend Codex couldn't get into, this one could: how Tencent's BrowserSkill managed to rack up 140k, and what you can take from it

BrowserSkill is a small tool open-sourced by Tencent's WeChat team. Install a command-line program and a browser extension, and agents like Claude Code, Codex, and Cursor can work in a separate window using the Chrome you're already logged into. The extension has 70k users on Chrome and 70k on Edge, and it's free. Tencent's brand and official launch pushed it, and followers can't copy those. But its install approach and the way it finds demand are both worth copying.

One line: "cannot be bypassed"

In mid-August, a CSDN author called "AI一手" wanted AI to help him figure out why his Official Account's recent numbers were down. He first tried OpenAI's official Codex Chrome extension. He was already logged into the backend, but after waiting a while Codex replied:

The WeChat Official Account platform is blocked from automated reading by browser security policy and cannot be bypassed.

He checked GitHub issues and figured out it wasn't a local misconfiguration. OpenAI maintains a server-side list of blocked sites, and WeChat domains like mp.weixin.qq.com are on it, so no local setting would help. He switched to Tencent's WorkBuddy with BrowserSkill, same backend and same login: "got in, clicked, analysis done" (CSDN, "WorkBuddy+BrowserSkill: a smoother browser control approach than Codex+Chrome").

Around the same time, a Bilibili creator tested it with a real account: the agent went into the Douyin creator backend to pull comments, clicked the wrong thing once, then re-read the page on its own to find the entry again, and finally retrieved 203 comments (a summary of a long post on Smzdm 8-18).

Both point to the same need: agents can write code and read docs, but they stall at any page that requires login. And the browser extensions big companies ship either only work with their own agent, or decide in the cloud which sites you're not allowed to visit.

What it actually is

Four parts: the bsk command line written in Rust, a local background process, a Skill description file for agents to read, and a Chrome / Edge extension. The agent calls bsk through the shell, the command is passed through the local process to the extension, and the extension opens an "Agent Window" with an orange border to do the work, without touching the window you're using.

The store page is blunt: "Let AI agents use your logged-in browser in a separate Agent Window—without interrupting your work." Its selling points:

• It reuses your login state, so you don't hand your account password to the agent;
• To use tabs you already have open, it has to "borrow" them and return them afterward; the on/off switch for this lives in browser settings and can't be changed from the command line;
• When it hits a CAPTCHA or a payment confirmation, it stops and hands control back to you, then continues once you're done;
• It all runs locally, is open source under MIT, is free, and has no in-app purchases.

The developer name is wxg.prc.cpg, which is the WeChat Business Group. On GitHub it sits under the Tencent organization, and the top ten contributors are all still committing. A new version comes out every one to two weeks, from Jun 22 (0.1.2) to Nov 5 (0.4.2). This is not a side project by one person.

How much it grew

(stars from web archives and GitHub, by date)

On the store side: Chrome 70k users, and this number stayed the same from 11-01 to 11-06 (the store only shows up to the tens-of-thousands level, so changes between 70k and 80k aren't visible). Reviews 4.52,23; over 6 days, 4 were added, and the score dropped from 4.95 to 4.52. The Edge store shows 69,996 active installs and a rating of 4.2 (Edge Add-ons, 11-06).

The most interesting part is the ratio: 10k stars, 140k extension users. The reason is the install flow, explained below.

Where users come from

Tencent's official account posted first. On Sep 16, @TencentAI_News (23k followers) posted: "We open sourced BrowserSkill, a bridge between your agent and your actual browser", followed by "most tools give the agent a blank browser. We let it borrow a tab from yours, then hand it back". The closing line, "it can't be talked around", got quoted a lot. The post has 208k views, 2,070 likes, and 2,767 bookmarks, more bookmarks than likes (X, 11-06 views). Stars went from 1.7k to 4.7k in two days and reached 8k in three weeks. The project README also shows a badge saying it placed #4 on GitHub Trending's daily overall chart in September (the project's own claim, which we have not verified independently).

Chinese-language communities heated up first. A month before the post, Zhihu (8-10, 8-12), CSDN, smzdm (8-18), Tencent Cloud developer community (8-20), and Bilibili already had reviews and tutorials. The smzdm piece said "discussion didn't really take off until August", while "WeChat Official Accounts, Toutiao, 36 Cailian, and TMTPost each had only 0 posts", so it was still at the stage of developers trying it themselves. The August wave took stars from a few hundred to over a thousand and built up a body of Chinese tutorials, which Medium, Reddit, and YouTube then kept covering secondhand after the English launch in September.

Tencent's own agent users. The README and website list WorkBuddy and CodeBuddy as first-class supported tools, alongside Claude Code and Cursor. A batch of tutorials are titled "WorkBuddy+BrowserSkill" and written for WeChat Official Account and e-commerce back-office operators, not programmers. Edge installs match Chrome's, which points to a large base of Windows office users and users in China (inferred).

The install flow turns every user into an extension user. The README's first step is one sentence: "Following the instructions in …/AGENT_INSTALL.md, install and configure browser-skill locally", which you paste to your agent. The agent reads the docs, installs the CLI, and configures the Skill itself. Then you run bsk doctor. The checklist usually has only one red item left, "extension connected", and that's when the person goes to the store and clicks "Add to Chrome". So everyone who actually uses it ends up installing the extension, while the star is something only some of them bother to give.

No paid ads found.

Who it's competing with

This category was already crowded before BrowserSkill. It is neither the first nor the largest.

Looking at Chrome alone, its 70k users sit one tier below OpenCLI's and Browser MCP's 100k. It doesn't own anything exclusive. It just holds a share of a category that is growing fast.

Three things set it apart: it isn't tied to any one agent; it uses an isolated window with consent required for borrowing a tab and hands key steps to the human; and it runs fully locally, so domestic Chinese sites are reachable, which also ties it to WorkBuddy users. These are enough to keep people using it, but none are hard to copy. OpenCLI was built by one individual developer and is bigger. What pushed BrowserSkill up was mainly the four words "Tencent open source" and that official launch.

One more thing to know: back in Sep 20 it had a vulnerability. CVE-2026-94111,0.3.0 and earlier versions let a local process bypass the WebSocket origin check, CVSS 6.9. For a tool that acts on your logged-in browser, security is both the selling point and the biggest risk.

If it were you

C: worth learning from, not worth chasing. Building another general "let an agent use your browser" puts you against official extensions from Anthropic, OpenAI, Kimi, and Microsoft, plus open-source projects like OpenCLI and Browser MCP that already have 100k users. BrowserSkill's own growth came from Tencent's name, its official account, and WorkBuddy's users, none of which you can borrow. It doesn't make money either: the store shows no revenue at all.

Who it suits: developers already running a specific back-office operation (WeChat Official Accounts, Douyin, Xiaohongshu, Taobao/Pinduoduo seller back-ends, cross-border stores), or people with an operator audience who can sell courses or templates to it, aged 1–2; people who want to build an open-source project that earns stars and already know a domestic platform's back-end well.

Who it doesn't suit: people who want to build yet another general browser bridge to chase user numbers; people who want to charge for the extension itself (this whole category is free, and users won't pay for "can connect to my browser"); people who aren't prepared for the security responsibility of a tool that holds users' login sessions.

How to do it: don't build the base layer, build the layer on top. BrowserSkill and OpenCLI are both open source and already solve "open the back-end". Nobody has done "what to do once you're inside" well. Pick one platform and write a Skill or set of commands, for example "pull the reads and conversions for each WeChat article from the last 30 days and write a review", "export Douyin comments and categorize them by question", or "scrape viral headlines from similar accounts", so WorkBuddy or Claude Code can run them out of the box. Open source earns reputation; money comes from outside: paid template packs, configuration services for operations teams, and courses.

How to get users: copy two moves. First, install by "paste this to your agent", so a user copies one line and is set up, low enough that an operator will try it. Second, saturate the Chinese-speaking market first: post real-account test videos and image-text posts on Zhihu, CSDN, and Bilibili (the review that captured 203 Douyin comments is the model), and only publish in English once you have enough content. Write headlines as direct comparisons like "what Codex can't do". That CSDN article spread on exactly that line.

When this window closes: before the first half of 2027. Every agent vendor is now adding browser capabilities. Once WorkBuddy, CodeBuddy, or BrowserSkill itself ships ready-made skills for things like "WeChat post review", big companies will have filled this layer themselves (inferred).

What would show you're wrong: ① BrowserSkill or WorkBuddy officially launches a ready-made skill marketplace for platforms like WeChat and Douyin, which narrows the path above; ② OpenAI removes WeChat and similar domains from its blocklist, so the reason "only local tools can reach domestic sites" disappears; ③ conversely, if a domestic platform's dedicated skill pack gets several thousand GitHub stars within a month or two, that shows the layer really is empty, and you should move fast.

Sources

• Chrome store listing and reviews (daily user counts and rating counts, 11-01 to 11-06); Edge store BrowserSkill page (11-06).
• GitHub Tencent/BrowserSkill: README, AGENT_INSTALL.md, release history, contributors (11-06); star counts from web archives, 8-20 to 10-05.
• @TencentAI_News post (9-16); view, like, and bookmark counts as seen at 11-06.
• CSDN "AI一手", "WorkBuddy+BrowserSkill: a smoother browser-control approach than Codex+Chrome"; smzdm, "Tencent open-sources BrowserSkill, AI can finally 'log in' to your browser" (8-18); Zhihu reviews (8-10, 8-12); Tencent Cloud developer community (8-20).
• CVE-2026-94111 (cve.org).
• Competitor user counts from the Chrome store, star counts from GitHub, both as of 11-06.