One tweet brought in 100 stars: someone wrote a Recovery from scratch in Rust for Xiaomi, aiming to replace TWRP

Skyward is a third-party Recovery for Xiaomi and Redmi phones. The repo was created in Oct 25. It doesn't build AOSP or touch Xiaomi's kernel. Instead it borrows the stock firmware's services to decrypt, flash, do OTA, and back up. In its first eight days it gained 50 stars mostly from the author's own posts. In Nov 2, a Chinese X account with a large following reposted it. It went to 150 the next day, and by 11-06 it was 190. Here's what it is, where the 100 stars came from, and whether you should build something like it, and how.

Oct 19, one tweet from an account with 74 followers

In 2026 Oct 19, Fuzhou indie dev Kirk Lin (@lkirkun on X, 74 followers at the time) posted:

I rewrote a Recovery for Xiaomi phones called Skyward, all implemented in Rust from 0. Building doesn't require downloading the AOSP source.

Almost no one saw it. In Oct 25 he created the repo and released v0.1.0. The same day he posted on XDA's Xiaomi 17 Ultra board, titled "[RECOVERY][A17] Skyward 0.1.0 for Xiaomi 17 Ultra (nezha)", and got few replies. That night he released v0.1.2, with images for 19 models at once. By Nov 2 the repo had 50 stars.

In Nov 2, a little after 1 p.m. Beijing time, Geek Lite (@QingQ77), a Chinese account that specifically recommends GitHub projects, posted one line with a screenshot:

A third-party Recovery for Xiaomi and Redmi phones, written from scratch in Rust, to replace TWRP. It supports decrypting the data partition, flashing, OTA, and backup.

The tweet got 24.6k views. The repo picked up 11 forks that day, and stars jumped from 50 to 150 the next day. The top reply: "holy... I didn't recognize it at a glance, I thought it was the built-in system."

What it is

A Recovery is a small system on the phone, separate from the OS, used for flashing packages, backups, and unbricking. For over a decade, the go-to third-party Recovery in Android modding has been TWRP. TWRP works by writing a config for each phone, pulling tens of GB of AOSP source, and recompiling. When a new model comes out, someone has to port decryption, touch, and drivers one by one.

Skyward takes a different route. It takes the official full package from Xiaomi's servers, extracts only the stock recovery partition, replaces the recovery program inside with one written in Rust in-house, and keeps everything else from stock: decryption relies on the stock KeyMint, Gatekeeper, and Weaver, and networking relies on the stock WLAN service. Each model needs just one TOML config. That's why the author could release 19 images in a single night.

Features roughly match TWRP: decrypt data, flash zips and system update packages, online OTA, backup and restore, file manager and terminal, WLAN, MTP, adb, and fastbootd. After a system update overwrites the recovery partition, it can automatically reflash itself. The UI is a renderer the author wrote, styled like HyperOS's built-in settings, with Simplified and Traditional Chinese and English support.

But be clear about its current state (from the repo README, 11-06):

Release download counts line up (GitHub, 11-06): v0.1.2 has had 335 downloads in total. The most went to the two tested models: 13 Ultra 43, 17 Ultra 36. Each of the others got 8 to 22. The people who actually flashed it to a phone are probably in the dozens to one or two hundred (inference).

Why this one, and why now

The gap is real. The Xiaomi 17 series and this batch of K90 with Snapdragon 8 Elite on Android 16/17 are new phones that TWRP is slow to support. On XDA, someone posted "Unofficial TWRP 3.7.1 for Xiaomi 17 series" back in March, and the thread still says the 17 Ultra is "still in bring-up, decryption not finished." People who want Magisk, or want to back up before tinkering, have no Recovery that can decrypt. Skyward says it decrypts on the 17 Ultra, and that alone is enough to make people click.

There's a line that spreads. "Written from scratch in Rust" and "no AOSP download" are tags programmers like to repost. "Replaces TWRP, can decrypt" is what modders want. The screenshot looking like the built-in system makes a third group stop and look. Geek Lite's tweet compressed all three into one line.

The author isn't new. He previously made the Magisk unbrick module magisk-brick-guardian (148 stars) and the LSPosed certificate-bypass module TrustMe (54 stars). He has 368 followers on GitHub and some standing in the Xiaomi modding scene.

Where the stars came from: one repost, not sustained growth

Source: GitHub daily stars (10-28 to 11-06), X, XDA.

A few things worth noting:

• Before and after the repost are two different worlds. One XDA post plus the author's self-promotion got 50 stars over eight days. One big-account tweet got 100 in a day. No second wave followed, and daily gains fell back to 4 to 8. Coolapk, V2EX, Hacker News, and Reddit show no discussion, and no media has written about it. It can't get onto GitHub Trending yet, since the repo is only 12 days old.
• The fork ratio is high. 35 forks to 190 stars, about 18%. People flashing Recovery either build it themselves or want to port it to the phone in their hands. Someone has already made ports for the 13 Pro (skyward-for-nuwa) and 14 Pro (skyward-for-shennong).
• Outsiders have started working on it. In PRs, people are adding support for the Redmi K70, K100 Pro Max, and Xiaomi Pad 8 Pro. Others are fixing mount bugs, adding settings persistence, and adding 7z and zstd package flashing. Issues include "can't flash custom ROMs," "color bugs on some model screens," and "language setting lost on reboot." So people really are flashing it, and it's still very early.
• The author hasn't pushed code since Oct 26. External PRs are piling up, and it's unknown when the next version will ship.

Who it's competing with

Star counts from GitHub, 11-06.

Bottom line: on new Xiaomi models, it's competing for the same users as the unofficial TWRP. Right now it leads on the "can decrypt" claim, but it only delivers that on two devices. Its real moat isn't features, it's the way it adapts: for another model, others need a week of porting, while it needs one config file.

If it were you

C. You can learn some very practical things, but following the same path won't win many stars. The ceiling is low and the bar is high, and this time's 100 stars mostly came from one tweet.

Phone modding is a small circle, and it's shrinking. Xiaomi's bootloader unlocking is getting harder in China, and under the tweet someone wrote "too bad I'm past the age for modding, now I just want my phone to be stable." A Recovery this solid getting 190 stars in 12 days is already close to the normal ceiling for this category. What's worth taking away isn't "build another Recovery," but its method and its promotion path.

Who it suits: developers already working on Android internals (Magisk modules, kernels, drivers) who have a new phone on hand and know Rust or C++; or people building tools for other fragmented hardware who want to borrow its approach.

Who it doesn't suit: people who've never touched bootloaders and partitions, and people who want a batch of stars in three months. A serious bug in a Recovery means someone loses all the data on their entire phone.

How to do it:

1. Switch brands, don't collide with Xiaomi. Xiaomi is already taken, and PRs are still adding models. Brands like OnePlus and realme, where unlocking is relatively easy and TWRP also lags on new phones, haven't shown a project with the same approach yet (inference; search GitHub and XDA before you start). Follow its path: take the stock recovery image, swap only the upper layer, one config per device.
2. Or do the part it hasn't done. All 19 images still can't decrypt, and no external PRs have been merged. If you want to get known in this repo's circle, getting decryption working on one or two popular models (like the K90 or 15 Ultra) and releasing a port separately is easier to notice than starting a new project from scratch. Just ask in an issue first to make sure the author isn't already doing it.
3. Apply "reuse stock, swap only the upper layer" elsewhere. The core idea is to shrink the cost of per-device porting into one config file. Router firmware, TV boxes, and car head units are all places where "every machine needs its own port" applies.
4. Release a lot of models at once, and be honest about what's untested. Its 17 "never tested on a phone" images still each get 8 to 22 downloads. The model list itself is a search entry point. Users only click star when they see "my model is supported."

How to get users: the author's own path shows that XDA and his own small account only bring dozens of stars. What really scaled was a tweet from a Chinese GitHub recommendation account. Accounts like Geek Lite are always looking for new projects. Your job is to make it easy for them: the first screen of the README should say in one line what it replaces and what problem it solves, with a good-looking screenshot (Skyward's screenshot was described as "I thought it was the built-in system"), and the title should carry one tech tag ("written from scratch in Rust"). On launch day, post on the XDA board for the relevant model and on Coolapk, then send the repo link to a few recommendation accounts (inference; I didn't verify whether the author reached out directly).

When is "before"? Once TWRP decrypts the newest models, the "it decrypts, TWRP doesn't" line is gone. The unofficial TWRP has been in progress for seven months and could catch up at any time (inference). For a project aimed at another brand, the best time to ship is when the next flagship launches and TWRP falls behind again.

What would show this read was wrong: ① Skyward's next version gets decryption working on a dozen or more models and stars keep rising past 1000, which would mean the phone-modding circle isn't shrinking and it's worth going after directly, so the grade should go up; ② a similar project for another brand gets recommended and only collects dozens of stars, which would mean this time's 100 stars came mainly from the big Xiaomi user base plus one account, and other brands can't replicate it; ③ Xiaomi or another brand tightens unlocking further, and the audience for third-party Recovery shrinks along with it.

---

Data & sources: GitHub repo README, Release download counts, issues and PRs (11-06); GitHub daily stars (10-28 through 11-06); Geek Lite tweet x.com/QingQ77/status/2117130331954295141 (11-02) and replies; author tweet x.com/lkirkun/status/2112050328015376627 (10-19); two XDA threads, Skyward and the unofficial TWRP for Xiaomi 17 Series; competitor repo stars from GitHub (11-06). Where we write "inferred," it's an estimate.