Every week, GitHub publishes its trending list. Raw numbers, repo names, stars climbing. But when you look at this week’s lineup — the week of July 22, 2026 — a pattern emerges.
Five projects. Three converging themes. One message: the way we code is changing, and the tools are changing with it.
No fluff. Here’s what’s rising, why it’s rising, and what it says about us.
1. mattpocock/skills — The Keyring for AI Agents
⭐ 182,000 stars · +10,651 this week
Mattpocock isn’t unknown. His newsletter reaches 60,000 developers. But his skills repo jumped hard this week — and it’s earned.
The idea is straightforward: AI agents (Claude Code, Cursor, Codex) are powerful, but without precise instructions, they go off the rails. This repo provides “skills” — context files you drop into your project to guide the agent.
A quick npx skills@latest add mattpocock/skills gives you access to 30+ skills. My favorite? /grill-me: an alignment interview where the agent asks you questions about your project before touching a single line of code. It’s the exact opposite of “generate me a REST API in 30 seconds.”
There’s also /tdd (forced test-driven development), /improve-codebase-architecture (architectural audit), and a clear philosophy: engineering fundamentals matter more than ever with AI.
What this says: Developers who pilot their agents win. Those who give them the right instructions win even more. The skill curve doesn’t disappear — it shifts.
2. Nutlope/hallmark — The Killer of Generic AI Design
⭐ 15,400 stars · +8,948 this week
You’ve seen those “AI-generated” websites? Same layout, same gradients, same animations. That’s what Nutlope calls design slop — the junk design produced by AI tools.
Hallmark is their answer. A skill for Claude Code, Cursor, and Codex that does exactly 4 things: build, audit, redesign, study. No predefined themes, no templates. Just briefs that force the agent to think before coding.
20 themes, 57 anti-slop test gates. Each brief produces a unique design — not a variant of the same template.
What struck me: every usage generates a different design. Not a better template. A design that didn’t exist before.
What this says: AI-generated design has become visual noise. The tools that cut through the noise — the ones that force reflection — are taking over.
3. OpenCut-app/OpenCut — CapCut, But Open Source
⭐ 77,400 stars · +8,341 this week
CapCut is everywhere. But it’s proprietary, with its limitations, pricing changes, and ByteDance dependency. OpenCut is the open-source answer: free video editing, web + desktop + mobile, MIT license.
The project is being rewritten from scratch. The core is moving to Rust, the architecture is becoming plugin-first. What’s coming: an MCP server for AI agents to control video editing, a headless mode (everything via command line), and a scripting tab for custom workflows.
1,584 commits, active community. The classic version still works during the rewrite — no break.
What this says: Consumer video editing is starting to have its open-source equivalent. Firefox did the browser, OpenCut wants to do CapCut.
4. diegosouzapw/OmniRoute — The AI Gateway That Costs Almost Nothing
⭐ 24,500 stars · +6,245 this week
One endpoint. 268+ providers. 500+ models. And crucially: 90+ free tiers — about 1.53 billion free tokens per month.
OmniRoute is an MIT AI gateway that runs locally. Plug it into Claude Code, Cursor, Copilot, or any compatible client. Define routing strategies (18 available), and the tool automatically switches between providers when one goes down or hits its limit.
“Caveman compression” reduces token consumption by 15-95%. That means your $20/month AI subscription lasts 3-10x longer.
Desktop, PWA, Docker, Termux — it runs almost everywhere.
What this says: AI isn’t just about models. It’s about infrastructure. Pipeline optimization tools are becoming as important as the models themselves.
5. koala73/worldmonitor — The World Dashboard, Real-Time
⭐ 67,800 stars · +5,517 this week
Worldmonitor is a global intelligence dashboard. 500+ curated RSS feeds, 15 categories, 25 languages, dual map engine (3D globe + WebGL flat). It’s not a news aggregator — it’s a geopolitical, economic, and technological monitoring system.
The serious part: a Country Instability Index that cross-references geopolitical, economic, and security data. A financial radar with intraday snapshots. Local AI via Ollama — everything runs on your machine, no data sent.
The MCP server lets AI agents query the dashboard in real-time. Python, Ruby, and Go SDKs offer programmatic integration. The desktop app (Tauri 2) is lightweight and fast.
What this says: Geopolitical intelligence won’t be the monopoly of specialized agencies anymore. Tools that give a developer or business owner the same visibility as an analyst — that’s the next wave.
The Common Thread
Five repos. But one thread: the infrastructure layer around AI is being built right now.
Models (GPT, Claude, Gemini) aren’t enough anymore. You need skills to guide them, gateways to route them, dashboards to pilot them, content tools to resist their generic output, and open-source alternatives to avoid depending on a single vendor.
The developer of 2026 doesn’t just code anymore. They orchestrate, they pilot, they curate. And the repos that are rising — those — are the tools of this new posture.
Want to know how your business can leverage these trends? Take the Izri.Online free diagnostic — we identify concrete quick wins for your business in 10 minutes.
Tags: GitHub, trends, AI, AI agents, open source, dev tools, productivity, 2026
