Today's GitHub trending is dominated by one theme: AI infrastructure growing up. Developers are formalizing best practices for agentic coding, demanding privacy-first alternatives to cloud-dependent tools, and building community archives to understand what's actually running inside the AI systems they rely on every day.
Zackriya-Solutions/meetily: The Case for Fully Local AI Meetings
Meetily surged to nearly 20,000 total stars and added 2,494 today — making it the week's fastest-rising project. Built in Rust, it delivers real-time meeting transcription using Parakeet or Whisper at claimed 4× baseline speed, with speaker diarization and Ollama-powered summarization, all running entirely offline. For teams wary of routing sensitive conversations through third-party cloud APIs, the appeal is direct.
github.com/Zackriya-Solutions/meetily
asgeirtj/system_prompts_leaks: The Unofficial AI Transparency Archive
With over 52,000 stars and nearly 1,400 added today, this repository has become a living reference for how major AI systems are actually instructed. It covers system prompts across Anthropic, OpenAI, Google, and xAI — including recent additions like Claude Sonnet 5 (July 1) and GPT-5.5 Codex (June 18). It's less a security concern than a transparency project: as AI tools are embedded deeper into professional workflows, understanding their default behaviors and guardrails has real engineering value.
github.com/asgeirtj/system_prompts_leaks
addyosmani/agent-skills: Senior Engineering Discipline for AI Agents
Addy Osmani's agent-skills library packages 24 production-grade development workflows as reusable Claude Code skills, covering the full lifecycle from spec and planning through testing, security hardening, and deployment. The project has topped 71,000 stars and added 1,100 today. Its core premise — that AI coding agents default to shortcuts and skip the practices that separate production code from prototypes — directly addresses one of the most persistent complaints about agentic coding in real-world use.
github.com/addyosmani/agent-skills
ruvnet/RuView: WiFi Signals as Presence and Health Sensors
RuView turns commodity WiFi hardware into a spatial sensing platform: a $9 ESP32 module can detect presence through walls, estimate breathing and heart rates, track 17-point body pose, and detect falls — all without cameras or wearables. At 78,000 total stars with 470 added today, the project integrates natively with Home Assistant, Apple Home, and Google Home via Matter, and its pretrained model fits in just 8KB. It's a compelling answer to the privacy tradeoffs of camera-based occupancy sensing in smart homes, healthcare, and retail environments.
firecrawl/firecrawl: Web Scraping Infrastructure for the AI Era
Firecrawl hit 147,000 stars with version 2.11 released recently, adding 867 today. The platform converts any website into LLM-ready formats — markdown, HTML, or structured JSON — with 96% web coverage including JavaScript-heavy pages, and supports interactive browsing, autonomous crawling, and batch processing across multiple SDKs. For developers building AI applications that need grounded, real-time web data, it remains the go-to answer to "how do I feed the web into my model" without building scraping infrastructure from scratch.