AI INTELLIGENCE / DAILY SIGNAL
0011011011
2026-05-15 / GITHUB SIGNAL

GitHub Trending Digest — May 15, 2026

mattpocock/skills — 84,695 stars (+3,155 today, +17K this week). Fifth consecutive 3K+ day. The .claude skills collection is on a tear — gained 17K this week alone and shows no signs of slowing. At this pace it crosses 1...

BRIEFING STRUCTURE19 SIGNALS

Inside this digest

  • Hottest Repos
  • AI & ML
  • Developer Tools
  • Hidden Gems
  • Notable Trends
← Back to archive

Hottest Repos

  • mattpocock/skills — 84,695 stars (+3,155 today, +17K this week). Fifth consecutive 3K+ day. The .claude skills collection is on a tear — gained 17K this week alone and shows no signs of slowing. At this pace it crosses 100K before month-end. https://github.com/mattpocock/skills

  • ruvnet/RuView — 57,317 stars (+1,865 today). WiFi-as-radar keeps compounding. Turns ESP32 WiFi signals into spatial intelligence and vital sign monitoring — no cameras. 7.4K forks. The non-AI breakout of May. https://github.com/ruvnet/RuView

  • obra/superpowers — 192,661 stars (+1,646 today). The 193K colossus. Jesse Vincent's agentic skills framework gains 1,500-1,800/day like clockwork. 16.8K forks. The gravity well of the agent skills category. https://github.com/obra/superpowers

  • tinyhumansai/openhuman — 8,878 stars (+1,272 today). Day 5: 2.5K → 8.9K. The "personal AI super intelligence" with 118+ OAuth integrations and Memory Tree has nearly quadrupled in under a week. Rust-powered, Obsidian-compatible. On track to cross 10K tomorrow. https://github.com/tinyhumansai/openhuman

  • CloakHQ/CloakBrowser — 11,760 stars (+1,286 today, +8.4K this week). Day 8: tripled from 3.9K. The stealth Chromium with source-level C++ patches crossed 10K yesterday and keeps accelerating. Drop-in Playwright replacement. The bot-detection arms race has a clear leader. https://github.com/CloakHQ/CloakBrowser

AI & ML

  • anthropics/financial-services — 23,224 stars (+12,529 this week). NEW: #1 weekly trending. Anthropic's official financial services toolkit — reference agents for pitch decks, equity research, GL reconciliation, KYC screening, month-end close. Works as Cowork plugins or via Managed Agents API. Partners include LSEG and S&P Global. This is the playbook for how model vendors capture vertical value. https://github.com/anthropics/financial-services

  • Hmbown/DeepSeek-TUI — 29,807 stars (+11,303 this week). Terminal coding agent for DeepSeek V4 explodes. Rust-built TUI with 1M-token context, streaming reasoning blocks, approval gates, and auto mode that picks model/thinking-level per turn. Install via npm, cargo, or brew. The DeepSeek ecosystem now has its own Claude Code equivalent. https://github.com/Hmbown/DeepSeek-TUI

  • github/spec-kit — 100,136 stars (+951 today). Crossed 100K. Called it yesterday — GitHub's spec-driven development toolkit hit the milestone. Specs become executable code via the specify CLI. 8.5K forks. https://github.com/github/spec-kit

  • K-Dense-AI/scientific-agent-skills — 22,361 stars (+643 today). 135 scientific skills for any agent. Genomics, drug discovery, materials science, clinical research, geospatial — plus 100+ database connectors (PubChem, ChEMBL, UniProt, COSMIC). Works with Claude Code, Cursor, Codex. The most comprehensive domain skill pack on GitHub. https://github.com/K-Dense-AI/scientific-agent-skills

  • NousResearch/hermes-agent — 151,884 stars. Crossed 152K. The self-improving agent with 200+ model support and 24K forks is now firmly the largest open-source agent project. 11.5K open issues signal a full ecosystem. https://github.com/NousResearch/hermes-agent

Developer Tools

  • pranshuparmar/witr — 16,336 stars (+494 today). "Why is this running?" Answers the question existing tools (ps, top, lsof) don't: not just what is running, but why — tracing causality chains through supervisors, containers, services, and shells. Interactive TUI mode. Written in Go, works on Linux/macOS/Windows/FreeBSD. Genuinely useful. https://github.com/pranshuparmar/witr

  • garrytan/gstack — 97,509 stars (+1,127 today). YC president's Claude setup approaching 100K. 23 opinionated tools (CEO, Designer, Eng Manager, QA). At this rate, crosses 100K within 3 days — the "celebrity skills pack" phenomenon is real. https://github.com/garrytan/gstack

  • decolua/9router — 10,616 stars (+6,024 this week). AI model router still climbing. Smart proxy connecting Claude Code, Codex, Cursor to 40+ providers with 3-tier auto-fallback. RTK token compression saves 20-40% input tokens. 489 open issues = real adoption pain. https://github.com/decolua/9router

  • supertone-inc/supertonic — 5,937 stars (+712 today). Day 4: on-device TTS keeps climbing. 31-language TTS via ONNX Runtime. SDKs for 7 languages. No cloud, no API calls. Voice Builder for permanent voice clones. The edge-AI audio category is legit. https://github.com/supertone-inc/supertonic

Hidden Gems

  • NVIDIA-AI-Blueprints/video-search-and-summarization — 1,108 stars (+305 today). NVIDIA's reference architecture for GPU-accelerated vision agents. Natural language video search, real-time alert verification, long video summarization. Uses Cosmos-Reason2-8B and Nemotron. MCP-compatible tools. If you work with video data, this is serious infrastructure. https://github.com/NVIDIA-AI-Blueprints/video-search-and-summarization

  • joeseesun/qiaomu-anything-to-notebooklm — 2,631 stars (+465 today). Claude Skill that converts anything (WeChat articles, YouTube, podcasts, paywalled sites) into NotebookLM podcasts, PPTs, mind maps, and quizzes. Includes paywall bypass for 300+ sites. The "content → any format" pipeline as a single skill. https://github.com/joeseesun/qiaomu-anything-to-notebooklm

  • richkuo/go-trader — 279 stars (+41 today). Lean crypto trading bot in Go with backtesting, paper trading, and live trading with risk management. Clean MIT license. Early but trending in Go daily — one to watch if you're in the trading bot space. https://github.com/richkuo/go-trader

Notable Trends

  • Agent skills are now a full-stack ecosystem. Five skills repos above 20K stars: anthropics/skills (135K), garrytan/gstack (97K), mattpocock/skills (85K), addyosmani/agent-skills (42K), K-Dense scientific-agent-skills (22K). The category has graduated from "config files" to "infrastructure layer." Domain-specific skill packs (scientific, financial) are the new growth frontier.

  • Model vendors are going vertical. Anthropic's financial-services repo — complete with named agents (Pitch Agent, GL Reconciler, KYC Screener), partner-built plugins (LSEG, S&P Global), and deployment via Managed Agents API — is the first major domain toolkit shipped directly by a model vendor. Expect every frontier lab to follow with their own vertical agent libraries. The era of general-purpose-only is over.