agent-browser plugins

by Replay

Record browser sessions with the Replay Browser through the host agent browser, analyze recordings through Replay MCP tools or Replay QA, and upload pending recordings when the turn ends.

by YONGJAE LEE

Motion forensics for the animated web. Four jobs against a live URL: decode (analyse motion + build, no code), clone (URL → React + Tailwind), verify (AE/SSIM + motion-parity score vs. an existing impl), extract (raw JSON dump). Extracts real CSS and animation params (GSAP / Framer Motion / Lenis / anime.js) directly from JS bundles, then verifies with AE/SSIM diff — no vision tokens. Three skills: ui-reverse-engineering (decode/clone/verify/extract entry points, supports Webflow IX2 / Next.js / Tailwind v4), ui-capture (reference + transitions), visual-debug (AE/SSIM diff + computed-style mismatch diagnosis).

by Tony Jiang

中文版:SOP 截图前/改完一批 UI 后,把全站可交互元素系统性点一遍——agent-browser(Rust CLI,自带 Chrome,a11y 快照)驱动 + 通用化遍历引擎(config 外置、默认拦截名单只增不减、八分类记账 + 可选孤儿功能对账:代码清单 vs 遍历实际到达,查「功能实现了但前端没入口/入口打不通」)。安全边界:破坏性按钮只记不点、confirm/prompt 一律驳回、出域双层防护(--allowed-domains 限显式导航 + 引擎逐击域名校验,出域即拉回),只用于自家或已获授权的站点。

by Tony Jiang

English version: before SOP screenshots or after a batch of UI changes, systematically click every interactive element on the site — driven by agent-browser (Rust CLI, bundled Chrome, a11y snapshots) plus a generalized traversal engine (externalized config, an append-only default denylist, eight-category ledger, plus an optional orphan-feature audit: code inventory vs what the sweep actually reached, finding features with no UI entry point or a broken one). Safety boundary: destructive buttons are logged, never clicked; confirm/prompt dialogs are always dismissed; two-layer domain guard (--allowed-domains for explicit navigation + per-click hostname check, off-domain clicks pulled back on the spot); only for sites you own or are authorized to test.

by Falconiere Barbosa

Efficient, token-lean browser automation for agents via agent-browser (skill + CLI wrapper) — accessibility-tree snapshots with @eN refs instead of HTML/screenshots. Standalone, no dependencies.

by Bymax One

Real-browser verification and assisted UI testing for web apps. /bymax-web-verify:test brings the full stack up — reuses a running backend instance or starts one, opens the frontend in the Claude Desktop Browser pane (preview) — and walks a UI flow step by step while the user watches, verifying via page state, console, network (2xx gate), and server logs; falls back to the agent-browser CLI in a terminal. /bymax-web-verify:verify drives a real browser to confirm a single change works. /bymax-web-verify:setup installs the agent-browser CLI + Chrome for Testing in one shot. A SessionStart hook detects a missing CLI and offers to install it. Depends on `agent-browser` but never bundles it. /bymax-web-verify:record captures a UI flow as reviewable video evidence: the project's own Playwright with video forced on for one isolated run, human pacing (slowMo with a resized timeout), ffmpeg lead-in trim with a verified first frame, an ffprobe pacing gate, optional --mp4, and a plain-text walkthrough derived from the spec that ran.