The Hottest Skills & Open-Source Projects on GitHub Right Now

 The Hottest Skills & Open-Source Projects on GitHub Right Now

If you've been watching GitHub lately, you've probably noticed two things: AI is everywhere, and a new category of open source called "Agent Skills" is exploding. Here's my curated list of what's actually worth your stars right now.

---

Why GitHub Is the Center of Gravity Again

GitHub has always been where software is born, but the last twelve months have been unusual. It's not just libraries and frameworks anymore — it's agents, skills, and self-hosted AI stacks. Developers are shipping not only code but capabilities: reusable folders of instructions and scripts that teach AI agents how to do things, from "convert a PDF to slides" to "manage your browser."

Two megatrends dominate the trending charts in 2026:

Agent Skills — portable, file-based capability packs for AI agents (a `SKILL.md` plus scripts and assets). Think of them as "plugins for the agent era."

Open-source AI infrastructure — local LLM runtimes, LLMOps platforms, RAG engines, chat UIs, and browser-automation agents that put frontier-grade AI in your own hands.

Below is my hand-picked list, split into those two buckets, with links and rough star counts (approximate, as of late 2026 — stars move fast).

---

Part 1 — The Agent Skills Revolution

1. anthropics/skills — the official Agent Skills repo ⭐ ~180k

If you only bookmark one thing, make it this. Anthropic's official repository holds both the reference implementation of Agent Skills for Claude and the Agent Skills specification itself. It ships with polished, production-grade skills:

Document skills: `pdf`, `docx`, `pptx`, `xlsx` — read, summarize, and create Office documents with high fidelity.

`mcp-builder` — lets Claude design, implement, and debug its own MCP servers from a prompt.

The repo famously surged past MCP in popularity on GitHub (see AIGazine's analysis), which tells you everything about where the ecosystem is heading. It's the reference for how a skill should be structured: a folder with `SKILL.md`, a `scripts/` directory, and optional reference assets.

2. agentskills/agentskills — an open, portable skill spec

Not everyone wants to be locked into one vendor. agentskills.io is an open specification for Agent Skills that aims for portability across Claude, Codex, Gemini, and other agent runtimes. If you're building skills you want to reuse everywhere, this is the standard to watch — the same `SKILL.md` format, vendor-neutral.

3. The awesome lists — your skill discovery engine

The community moved fast, and curated lists are now the best way to browse hundreds of community skills:

vivy-yi/awesome-skills — 230+ repositories of AI agent skills, Claude Skills, and agent capabilities. Probably the most comprehensive index out there.

kenryu42/awesome-claude-skills — a focused, well-maintained Claude Skills list.

ComposioHQ's awesome-claude-skills — a 111-skill collection covering everything from GitHub automation to email drafting (via Composio, the agent tooling platform).

royalpinto007/awesome-agent-skills — another strong general-purpose list.

mkll/awesome-claude-skills-ai — a growing collection with a handy directory structure.

Pro tip: when you find a skill you like, read its `SKILL.md` first. A great skill has a clear "when to use" section, concrete instructions, and minimal dependencies.

4. davila7/claude-code-templates — skills management for Claude Code

A CLI tool for configuring and monitoring Claude Code that has grown a big skills integration (v1.24.0+). It lets you install, organize, and swap skills in Claude Code from the terminal, and it's become the de-facto community manager for local skill libraries. If you live in Claude Code, this is the fastest way to turn it into a Swiss-army agent.

---

Part 2 — Open-Source Software Everyone Is Talking About

5. Ollama — run LLMs on your own machine

The simplest way to run large language models locally: one command, hundreds of models, GPU or CPU. The local-LLM ecosystem has grown to 695 models across 15 frameworks, and Ollama is the entry point for most people. Privacy-friendly, offline-friendly, and the foundation of countless self-hosted stacks. Pair it with Open WebUI (below) for a ChatGPT-grade experience on your hardware.

6. Dify — the LLMOps platform that crossed 100k stars ⭐ 100k+

Dify is a full-stack platform for building LLM apps: visual workflow builder, RAG pipeline, model gateway, agents, and observability in one product. It officially crossed 100,000 GitHub stars and is the default choice for teams that want to ship AI products without gluing ten tools together. If "I want to build an AI app this weekend" is your vibe, start here.

7. n8n — the workflow automator that jumped to #1

n8n is a fair-code workflow automation platform, and in 2025 it jumped to #1 in JavaScript Rising Stars — driven almost entirely by its AI agent nodes. It now competes head-on with Zapier/Make while staying self-hostable, and its visual editor + code nodes make it the favorite for connecting LLMs to your real-world tools (email, Slack, databases, APIs). For "AI doing my busywork," this is the tool.

8. Open WebUI — your self-hosted AI chat frontend

An extensible, feature-rich web UI for LLMs — works with Ollama and OpenAI-compatible APIs, with RAG, web search, multi-user support, and a plugin ecosystem. It's the most-starred way to turn a local model into a polished, shareable chat experience. If Ollama is the engine, Open WebUI is the dashboard.

9. LobeChat — the beautiful open-source AI chat client

LobeChat is an open-source, modern ChatGPT-style interface with an eye-catching design, plugin system, knowledge base, and multi-model support (OpenAI, Claude, Gemini, local models, and more). Hugely popular as a desktop and web chat client, especially among users who want one clean window for all their AI services.

10. RAGFlow — deep-document RAG done right

Built on a "deep document understanding" approach, RAGFlow gives you accurate, citation-backed answers over your own documents — templates, tables, charts included. It's one of the top RAG engines on GitHub and the go-to when naive chunk-and-embed retrieval isn't cutting it for your PDFs.

11. Cline / Continue / Aider — the AI coding trio

AI-assisted coding has its own arms race:

Cline — an autonomous coding agent that lives in VS Code, plans, edits, and runs commands with your approval.

Continue — an open-source AI code assistant with deep IDE integration and a hub of models.

Aider — the terminal-native pair programmer, beloved for its git-aware workflow and speed.

All three are top-tier, and the terminal AI agent landscape from late 2025 shows this space maturing fast — Claude Code and Codex CLI set the bar, and these open-source options are closing the gap.

12. browser-use — AI that drives your browser ⭐ ~79k

One of the fastest-rising repos of the last year (~79k stars): give it a task like "book a meeting and email everyone the calendar link," and it controls a real browser — clicking, typing, filling forms — with computer-vision-based perception. It powers a whole ecosystem of browser agents and is the best open look at where "AI does my web chores" is going.

13. MaxKB & 1Panel — the Chinese ecosystem's quiet giants

Both earned spots in the 2025 LLM open-source ecosystem map:

MaxKB — an out-of-the-box knowledge-base Q&A system: upload documents, get an answerable chatbot with citations. The fastest way to a company wiki bot.

1Panel — a modern, open-source server operations panel (think "modern cPanel") that's widely used to deploy and monitor self-hosted AI apps.

Honorable mentions

vLLM — the high-throughput inference engine behind many self-hosted setups.

LM Studio — the friendliest desktop app for discovering and running local models.

Langflow — a visual builder for agent and RAG applications.

Langfuse — open-source LLM observability (tracing, evals, cost tracking).

Cherry Studio — a polished desktop multi-model AI client.

---

Summary Table

Project Category Why it's hot Rough stars*

anthropics/skills Agent Skills (official) Reference skills + spec for Claude ~180k

agentskills/agentskills Agent Skills (open spec) Vendor-neutral skill portability growing fast

vivy-yi/awesome-skills Skills index 230+ curated skill repos high

davila7/claude-code-templates Skills management CLI control of Claude Code + skills very high

Ollama Local LLM runtime One-command local models 100k+

Dify LLMOps platform Visual LLM app builder + RAG 100k+

n8n Workflow automation #1 in JS Rising Stars 2025, AI nodes 100k+

Open WebUI Chat UI (self-hosted) Polished frontend for local LLMs 100k+

LobeChat Chat client Beautiful multi-model UI + plugins 100k+

RAGFlow RAG engine Deep document understanding 50k+

Cline AI coding agent Autonomous VS Code agent 50k+

browser-use Browser agent AI operates a real browser ~79k

MaxKB Knowledge-base Q&A Instant doc-to-chatbot very high

1Panel Server operations Modern self-hosted control panel very high

*Star counts are approximate as of late 2026 and change daily — check the repos for live numbers.

---

How to Keep Finding Gold

GitHub Trending (github.com/trending) — check daily/weekly; filter by language and "today/this week."

Awesome lists — every ecosystem has one; they're the highest-signal filter the community produces.

Star History (star-history.com) — spot rockets early by watching growth rate, not just absolute stars.

Follow the sources, not the hype — official repos (`anthropics/skills`) and open specs (`agentskills.io`) tell you where things are standardized before the hype cycle does.

---

Final Thoughts

The pattern is clear: the most exciting open source right now is about giving AI agents skills — the ability to touch your documents, your browser, your code, and your workflows. The good news is that all of it is free, self-hostable, and moving at a pace that makes last year's tools look ancient.

Star the projects you use, read the `SKILL.md` files, and maybe write one yourself — the barrier to contributing has never been lower. A folder, a markdown file, and a script: that's a skill. Ship it.

Found this useful? Share it with a developer friend who's still asking "what should I self-host this year?" — and check back in a month, because this list will already be outdated.