Today's GitHub trending is less about new model releases and more about the infrastructure being built around them. Developers are shipping browsers purpose-built for agents, curating reusable skills at massive scale, and reaching for offline-first tools that keep sensitive work off third-party servers — a sign that the AI tooling layer is quietly but steadily growing up.
Ego-Lite: A Browser Purpose-Built for AI Agents
citrolabs/ego-lite is a JavaScript browser runtime designed specifically for AI agents doing web automation — and unlike headless solutions, it's built to preserve existing logged-in sessions without disrupting the user's own browsing. It picked up nearly 1,000 stars today. The problem it solves is real: agents that need to operate authenticated web sessions typically either require the user to hand over credentials or spin up a separate isolated browser that doesn't share cookies. Ego-lite threads that needle, making it easier to build agents that work within your existing web context.
Matt Pocock's Skills Library: 188K Stars and Still Climbing
mattpocock/skills — a Shell-based library of professional engineering skills extracted from developer agent configurations — added 1,740 stars today, continuing its trajectory as one of the most-starred repos in the skills-for-agents space. The project collects reusable, composable agent behaviors that developers would otherwise rediscover and rewrite independently. At 188K total stars, it has clearly become a go-to reference for teams building or customizing agentic workflows.
Block/Buzz: A "Hive Mind" Communication Platform in Rust
block/buzz is a new Rust-based communication platform billed as a "hive mind" — a term that suggests it's designed around collective, agent-assisted communication rather than conventional chat. It led today's trending list with 2,491 new stars, an unusually sharp spike for a day-one appearance. Details are sparse, but the combination of Block's infrastructure pedigree and the positioning around collective intelligence makes it worth watching as more context emerges.
Alibaba's Open-Code-Review: LLM-Powered Code Review With Deterministic Guardrails
alibaba/open-code-review is a Go-based tool that pairs deterministic rule pipelines with LLM agents to deliver line-level code review comments, with a security-focused ruleset baked in. It gained 431 stars today. What makes the design interesting is the explicit separation of concerns: fast, predictable rules handle the things rules are good at, while the LLM layer handles nuance and context. For teams wary of fully trusting an AI reviewer, this hybrid approach is a sensible middle ground.
github.com/alibaba/open-code-review
Harper: Offline, Privacy-First Grammar Checking in Rust
Automattic/harper is a fast, open-source grammar checker that runs entirely offline — no data leaves your machine. Built in Rust and powering tools like VS Code extensions and local editors, it gained over 500 stars today. As more developers route work through AI tools that send text to external APIs, Harper's appeal is straightforward: you get grammar checking that's genuinely private, and at roughly millisecond-per-document speeds, there's no meaningful performance tradeoff for keeping it local.