Browser & web agents — what's accelerating
The browser is how an agent touches the real, logged-in web — and that's exactly where things break. Below: the fastest-climbing tooling for driving pages, scraping at scale, and surviving real-world page churn, separated from the orchestration and skill-pack noise that keeps washing into the same searches.
Top mover
Turns any website into a CLI an agent can drive, reusing your already-logged-in browser session. That last part is the whole point: most useful web work lives behind auth, and reusing the session you already have sidesteps the login/captcha wall instead of fighting it.
---
The browser layer
real substanceA browser-automation CLI built specifically for AI agents, from Vercel Labs. A Rust core plus a named backer is a credibility signal in a category crowded with thin wrappers.
Gives an agent read/search access across Twitter, Reddit, YouTube, GitHub, Bilibili and XiaoHongShu through one CLI with zero per-platform API keys. The value is consolidation: one surface instead of six brittle integrations each with its own auth and rate limits.
The widely-adopted "make websites accessible for AI agents" library — still climbing hard at near-100k stars. At this size the velocity matters more than the absolute count: a large project still adding ~167 stars/day is one the ecosystem is actively standardising on.
An API to search, scrape and interact with the web at scale. Where the others drive a browser, Firecrawl is the ingestion layer — turning messy pages into clean, agent-readable data.
---
Adjacent, worth a glance
Two repos sit next to the browser layer rather than in it: JCodesMore/ai-website-cloner-template (⭐16,419 · ↑193.2/day · TypeScript) clones any site with one command via a coding agent — useful, but a build-time template, not a runtime agent — and KeygraphHQ/shannon (⭐44,316 · ↑175.9/day · TypeScript), a white-box AI pentester that reads your source and probes web apps and APIs. Shannon is a genuine web agent, just a security one; flagged here so the core list stays about general automation.
Noise filter
not browser infraFast climbers that share the search but aren't web-agent tooling: ComposioHQ/awesome-claude-skills (⭐63,433 · ↑273.4/day) is a curated skill list; Yeachan-Heo/oh-my-claudecode (⭐35,879 · ↑242.4/day) is multi-agent orchestration for Claude Code; earendil-works/pi (⭐60,269 · ↑200.9/day) is a general agent toolkit. All high-velocity, none of them about driving a browser — listed so the ranking above stays honest.
---
How this was made
Live GitHub pull, bucketed by browser/web-automation keywords, each repo verified not-archived and pushed recently, ranked by stars/day, then curated to separate real browser infrastructure from orchestration and skill-pack noise. Star counts pulled at publish — they move daily; re-verify before reposting.
Accelbrief · catch acceleration, not stars · all editions