GitHub Trending Repos This Week — May 16, 2026

TL;DR: This week's GitHub trending is dominated by AI agent skills and developer tools — from OpenHuman's privacy-first desktop agent to superpowers' agentic framework and RuView's WiFi-based spatial intelligence. Here are the 8 repos gaining the most traction.

The open-source landscape this week is defined by three themes: AI agent skills (reusable agent capabilities), local-first tools (on-device AI, code graphs, offline TTS), and agentic frameworks that give agents structure. This follows last week's trend of agent-focused repos, but with a sharper shift towards reusable skills that agents can consume directly — paralleling the skills and memory architectures we explored in our Hermes guide series. Here are the repos making the biggest moves this week.

Weekly Trending Summary

RepositoryStarsWhat It Does
tinyhumansai/openhuman~4.5kDesktop AI agent that understands you from day one via OAuth context
obra/superpowers194kAgentic skills framework and software development methodology
ruvnet/RuView58kWiFi signals → spatial intelligence, vital signs, presence detection
mattpocock/skills75kCurated CLAUDE.md skills from a TypeScript expert
K-Dense-AI/scientific-agent-skills23kAgent skills for research, science, engineering, and analysis
colbymchenry/codegraph2.4kPre-indexed code knowledge graph for Claude Code
supertone-inc/supertonic6.7kLightning-fast on-device multilingual TTS via ONNX
Anil-matcha/Open-Generative-AI14kOpen-source AI video/image generation with 200+ models

1. tinyhumansai/openhuman — The Agent That Reads You First

OpenHuman takes an inverted approach to AI agents: instead of asking you what you want, it builds context about you from day one. It requests OAuth access to your calendar, email, documents, and browsing history — then uses that context to make informed decisions on your behalf. The approach clearly resonated: it topped GitHub Trending this week with rapid star growth.

Built with privacy as a design principle (all processing happens locally), OpenHuman uses a plugin architecture for integrating with third-party services. It's written in TypeScript with a Python backend for local LLM orchestration.

2. obra/superpowers — Agentic Skills Framework

superpowers is the most-starred entry on this list at 194k, and it's still gaining momentum. It's an agentic skills framework that defines a software development methodology for AI agents. Instead of giving agents loose instructions, superpowers provides structured skill definitions that agents can execute reliably. It's gained 1,200+ stars today alone.

The repos is written in Shell with skill definitions in markdown, making it accessible to any agent that can read structured prompts.

3. ruvnet/RuView — Spatial Intelligence From WiFi

RuView is one of the most surprising repos this week. It uses commodity WiFi signals to generate real-time spatial intelligence — tracking movement, monitoring vital signs, and detecting presence — all without any cameras or sensors. It works by analyzing how WiFi signal reflections change as people move through a space.

At 58k stars, RuView demonstrates the growing interest in unconventional AI applications. It's written in Python with signal processing libraries at its core.

4-5. The Skills Explosion: mattpocock/skills and scientific-agent-skills

Two skills repositories are driving major engagement this week. mattpocock/skills (75k stars) comes from the well-known TypeScript educator, offering CLAUDE.md configurations that improve Claude Code's behavior based on real-world engineering experience. K-Dense-AI/scientific-agent-skills (23k stars) provides ready-to-use agent skills for research, science, engineering, financial analysis, and technical writing.

The trend confirms something we've been tracking: reusable agent skills are becoming the new "open-source library" — not for human developers, but for AI agents to consume.

6. colbymchenry/codegraph — Smarter Code Context for Claude Code

codegraph pre-indexes your codebase into a knowledge graph that Claude Code can query efficiently. By reducing the number of tokens and tool calls needed to understand a codebase, it makes AI-assisted coding faster and cheaper. It operates entirely locally — no data leaves your machine.

At 2.4k stars, it's the smallest repo on this list but one of the fastest-growing, gaining nearly 400 stars today from developers looking to optimize their Claude Code workflow.

7. supertone-inc/supertonic — On-Device TTS

supertonic delivers multilingual text-to-speech running locally via ONNX runtime. It's optimized for speed — "lightning-fast" in the project's own description — and supports multiple languages without any cloud dependency. Voice cloning and voice design are handled entirely on-device. This is part of a broader shift toward local-first AI that doesn't require API calls for basic inference.

8. Anil-matcha/Open-Generative-AI — AI Media Studio

Rounding out the list is Open-Generative-AI, an open-source alternative to proprietary AI video platforms. It includes over 200 models for image and video generation, plus a studio interface for editing. At 14k stars, it's the go-to self-hosted option for AI media creation.

Frequently Asked Questions

What's driving the explosion of agent skills repos?

The rise of agentic coding tools (Claude Code, Cursor, Codex) has created demand for reusable, version-controlled agent configurations. Skills repos are the new "dotfiles" — they're how developers share the agent behaviors they've refined through trial and error.

Which repo grew fastest this week?

OpenHuman had the fastest star velocity, hitting 776 stars in a single day at its peak. It topped GitHub Trending and was the most-discussed new project across developer communities.

Sources

← Back to all posts