Weekly open-source repository ranking
This is the 17:00 hourly edition. Follow the article link to view the full version in Portuguese, English and Spanish.
---
Each week, GitHub Trending serves as a barometer of what matters in the open-source world. Today's ranking paints a more nuanced picture than recent weeks: less framework hype, more concrete tools — from security to agent-oriented design systems, personal finance, and local inference — indicating a mature ecosystem focused on solving real problems rather than announcing platforms.
1. GoogleChrome/modern-web-guidance-src — TypeScript (2,678★)
799 stars just today. This isn't a framework, a library, or a runtime. It's a guidance repository: a set of best practices for modern web development, specifically designed to keep coding agents (like Cursor, Windsurf, and Claude Code) up to date with the state of the art in web development. With the massive increase in projects being generated or maintained by AI, Google Chrome has addressed the most common bottleneck: agents trained on outdated data applying 2023 patterns to 2026 projects.
What it solves is the "obsolete agent" problem — rather than the developer hunting through documentation to find which is the modern API, the repository centralizes verified recommendations. The explosive number of stars today (799) reflects the practical urgency of the team: it's the same reasoning that led Apple to develop the Neural Engine, but now applied to code guides rather than hardware.
2. openai/codex-security — TypeScript (12,296★)
663 stars today. OpenAI couldn't be more explicit about what this project solves: it's a CLI and TypeScript SDK to find, validate, and fix security vulnerabilities in code. With Codex (and its successors) being increasingly used in production pipelines, the attack surface has expanded accordingly. This package — now published as `@openai/codex-security` on npm — acts as a real-time guardian, identifying insecure patterns before merge.
The project's traction (12,296 stars in a few months) reflects the sector's awareness: the productivity of code agents is undeniable, but security cannot be an afterthought. The trend of integrating security linting directly into the code-generation tool — rather than relying on post-merge audits — is the same principle that moved the CI/CD industry to the left a decade ago. Now, security is going even further: to the source.
3. nvm-sh/nvm — Shell (95,982★)
107 stars today. It seems strange to see such a mature project (nearly 100k stars, active since 2014) on the weekly ranking. But the fact is that nvm continues to attract contributions and updates, and the installed base grows along with the Node.js ecosystem. In a market where new bundlers and frameworks emerge every month, nvm remains critical infrastructure — and its position in the ranking confirms that, at the bottom of the ecosystem, the fundamental tools remain what sustains everything.
4. pookjw/HermesEnabler — Objective-C (1,480★)
300 stars today. This project allows using Hermès (Apple's watch) watch faces on any Apple Watch device. It's a classic example of the open-source community filling gaps that manufacturers don't address — and the rapid traction shows there's real demand for hardware customization that the mainstream market ignores.
5. we-promise/sure — Ruby (13,771★)
974 stars today. The description is simple: "The personal finance app for everyone." In a world saturated with finance apps that charge subscriptions, sure positions itself as an open-source, accessible alternative — and the explosive number of stars indicates the public is hungry for financial control tools that aren't walled gardens. The choice of Ruby as the language (rather than Go or Rust, which you'd expect in 2026) reflects the project's philosophy: prioritizing developer productivity and code accessibility.
6. QwenLM/qwen-code — TypeScript (29,719★)
151 stars today. Qwen Code is QwenLM's open-source coding agent — it runs in the terminal, integrates with editors, and allows developers to write, debug, and iterate using Qwen models. With 29,719 stars, it's already the leading open-source alternative to GitHub Copilot and Cursor. The modest traction today (151 stars) is expected: the project is already consolidated.
7. ashhart/TensorFold — Zig (1,902★)
94 stars today. LLM inference engine for Metal, CUDA, and Vulkan. In a CUDA-only dominated market, TensorFold offers efficient inference on Apple (Metal), NVIDIA (CUDA), and AMD (Vulkan) hardware with a single codebase. The use of Zig — a systems language with manual memory management and performance near C — is no accident: it's the right choice for an inference engine that needs raw performance across multiple GPU backends.
8. nodejs/node — JavaScript (123,562★)
133 stars today. The central runtime of the JavaScript ecosystem continues to be updated and maintained rigorously. Its position in the ranking reflects the constant vitality of Node.js — it's no novelty, but it's essential infrastructure.
9. addyosmani/agent-skills — JavaScript (116,575★)
812 stars today. Production-grade engineering skills for AI coding agents. Addy Osmani (Google) publishes here prompt skills that define code patterns, architecture, and best practices that any coding agent can follow. It's the materialization of the concept of "agent as team member": it's not enough to generate code; the agent needs to understand the context, constraints, and quality standards of the project.
With 116,575 stars, it's one of the most-read projects on the ranking — and the peak of 812 stars today shows that the adoption of coding agents in production is far from exhausted.
10. meta/astryx — TypeScript (26,343★)
1,789 stars today. Meta's open-source design system, called Astryx, is fully customizable and agent-ready. Unlike other design systems that serve only human developers, Astryx was designed from the start to be consumed by both people and AI agents. The result: components with tokens, documentation, and structured constraints that allow an agent to generate interfaces that are already correct from the start.
With 1,789 stars in a single day, it's the most-starred repository on today's ranking — and not by coincidence. Design systems are the area that benefits most from agents in the development workflow: when an agent can read design tokens and apply constraints automatically, prototyping speed explodes.
---
What the collection reveals
Today's ranking outlines a clear map: coding agents have moved beyond experimentation and become infrastructure. The agent-skills (addyosmani), codex-security (OpenAI), modern-web-guidance-src (Google Chrome), and astryx (Meta) are four projects that, at different levels, treat agents not as an optional feature, but as a first-class citizen of modern development.
The trend is convergence: tools that create standards (skills), guides (modern-web-guidance-src), security guardians (codex-security), and design systems (astryx) that, together, form the backbone of an ecosystem where humans and AI agents work together. It's the same reasoning that led Apple to develop the Neural Engine, but now at ecosystem scale — artificial intelligence has moved from being an internal component to becoming the architecture around which all software engineering is organized.
The second axis of the ranking is maturity: nvm-sh/nvm and nodejs/node, projects with tens of thousands of stars, continue to attract contributions and updates. The fundamental infrastructure hasn't disappeared — it has evolved, and AI agents are the latest evolution.
The question that remains is: when all developers are using agents as part of their daily workflow, what differentiates a software engineer from an "agent orchestrator"?
Sources: GitHub Trending, GitHub - GoogleChrome/modern-web-guidance-src, GitHub - openai/codex-security, GitHub - addyosmani/agent-skills
✓ Independent sources cross-checked and verified before publishing
