Tencent's BrowserSkill 0.3.0 teaches agents to click inside a canvas via screenshot coordinates
Tencent/BrowserSkill cut `ext-v0.3.0` on 2026-09-16 and `cli-v0.3.0` on 2026-09-17T02:22Z, and gained 1,350 stars today to reach 3,604. The headline of the release is a canvas visual-object-model track: discovering canvas candidates and resolving visible regions, typed visual refs with target identity validation, visible-region screenshots bound to a canvas ref, cursor continuation when there is too much to show, and screenshot-coordinate point interactions on canvas regions. It also adds real mouse-wheel, focus/blur and scroll-to-element primitives, host-managed daemons for sandboxed agents, authenticated remote gateways, and a local task execution history for audit. The pitch is driving your real logged-in browser rather than a fresh automation profile.
Source
↳ Follow the thread