Blog

Their from-scratch table left AginxBrowser out. Same 940-page Amiibo crawl, pool 25, this Mac: 0.6.2 at 51.6–52.4s and about 400 MiB RSS. Peers were not re-run on Oct 22. The 4.5s figure stays on their AWS table.

Lightpanda 10 月 19 日的从零浏览器表里没有 AginxBrowser。同一道 940 页 Amiibo、25 路，这台 Mac 上 0.6.2 三次冷启动 51.6–52.4 秒，峰值 RSS 约 400 MiB。对手 10 月 22 日没有重跑。4.5 秒留在他们的 AWS 表里。

A 35k-star Zig headless browser shipped 0.4.1; reading their release notes against our git log, several things they filed tickets for this week we had shipped days earlier. No adjectives — timestamps and links only, including the row where they filed first. The living scoreboard lives in docs/RECEIPTS.md.

对着 0.4.1 的 release notes 翻自己的 git log：这周开票的几个角落，我们几天前就 ship 了。只有时间戳和链接——包括一条他们先开票的，照样放进来。活版记分牌在 docs/RECEIPTS.md。

The "single binary, zero Chromium" promise left no room for headless Chrome. So we run the page's JS in our own V8 and hand the final DOM to Dioxus Blitz for layout and paint. The blank-screen root cause (upstream #636), the CJK hang and pin decision (parley#752), and the self-contradicting crop trap.

Kitesurf validated that agents need their own browser — then openly listed hard gaps: real TLS fingerprinting and persistent authenticated sessions. Those are exactly what AginxBrowser was built for — plus a plain HTTP API.

全白屏根因链、CJK 挂死的钉版决策、元素坐标与 paint_scene 裁剪的自相矛盾——纯 Rust 渲染栈的真实踩坑记录。

Kitesurf 验证了方向，也公开承认了三个最难的场景它不做——正是 AginxBrowser 守着的山头。