Two weeks ago I wrote that every hot GitHub repo was an agent. This week? It’s worse. Or better, depending on where you sit.
I pulled the full GitHub Trending data for daily, weekly, and monthly rankings on March 26, 2026. Here’s what the numbers say, and what they mean if you’re building anything with AI right now.
The Numbers Are Absurd
The top two repos on GitHub this month have a combined 224,000+ stars. Both are agent skill frameworks. obra/superpowers sits at 115k stars with 51,000 gained in a single month. affaan-m/everything-claude-code is at 109k with 55,000 gained in the same period.
For context, React has 234k stars. These repos are approaching React-level popularity in weeks, not years. The difference: React took a decade. These took a quarter.
Out of the 15 weekly trending repos, 11 are directly AI-related. The remaining four are a Minecraft world generator, a money-making automation tool, and two PDF/document parsers (which exist to feed data into AI pipelines). So really, it’s 13 out of 15.
The 10 Repos You Should Know About
1. obra/superpowers (115k stars, +17,540 this week)
What it is: An agentic skills framework and software development methodology. Think of it as a structured way to give coding agents reusable skills, so they don’t start from zero every session.
Why it matters: This isn’t a model or a wrapper. It’s a methodology. The idea that agents need persistent, composable skills is becoming the dominant paradigm. If you’ve worked with skill-based agent systems, you know why. Without skills, every AI session is Groundhog Day.
github.com/obra/superpowers | Shell
2. affaan-m/everything-claude-code (109k stars, +22,442 this week)
What it is: A performance optimization system for agent harnesses. Skills, instincts, memory, security, and development methodology for Claude Code, Codex, Opencode, Cursor, and other coding agents.
Why it matters: The Claude Code ecosystem is spawning its own tooling layer. This repo, plus claude-hud (monitoring), claude-skills (plugin library), and learn-claude-code (build your own) all trended this month. It’s like watching the npm ecosystem form in real time, except for agents instead of packages.
github.com/affaan-m/everything-claude-code | JavaScript
3. bytedance/deer-flow (47.8k stars, +13,951 this week)
What it is: ByteDance’s open-source long-horizon SuperAgent. It handles tasks that take minutes to hours, using sandboxes, persistent memory, tool access, subagents, and a message gateway to coordinate complex work.
Why it matters: This is a Big Tech company open-sourcing their internal agent infrastructure. Long-horizon agents (ones that work for hours, not seconds) are the next frontier. The fact that ByteDance is betting on this architecture tells you where the industry is headed. If you’ve been building agents in Python, deer-flow shows you the end state: agents that don’t just answer questions but run entire projects.
github.com/bytedance/deer-flow | Python
4. TauricResearch/TradingAgents (42.2k stars, +8,939 this week)
What it is: A multi-agent LLM framework for financial trading. Multiple specialized agents (analyst, risk manager, portfolio manager) collaborate on trade decisions.
Why it matters: Financial AI agents are having a moment. This repo plus virattt/dexter (18.7k stars, autonomous financial research) plus a Chinese fork (TradingAgents-CN, 21.6k stars) signal massive demand. People want AI that can research stocks, assess risk, and suggest trades. Whether it works in live markets is a different question, but the 8,939 stars in one week tell you the appetite is real.
github.com/TauricResearch/TradingAgents | Python
5. ruvnet/RuView (42.8k stars, +1,001 today)
What it is: WiFi-based human pose estimation, vital sign monitoring, and presence detection. No cameras. It uses commodity WiFi signals to track human movement and health metrics in real time.
Why it matters: This is the kind of project that makes you stop scrolling. DensePose from WiFi signals, not video. The privacy implications are interesting both ways: you can monitor elderly patients without cameras (good), or you can detect people through walls (concerning). It gained 34,000 stars this month, which suggests people are genuinely excited about camera-free sensing.
github.com/ruvnet/RuView | Rust
6. supermemoryai/supermemory (19.5k stars, +2,064 this week)
What it is: A memory engine and API built for AI applications. Extremely fast, scalable, designed to be the persistence layer that agents need.
Why it matters: Every agent framework has a memory problem. Sessions end, context is lost, the agent starts over. SuperMemory positions itself as the database layer specifically for AI memory: not a generic vector store, but purpose-built for how agents actually store and retrieve context. Combined with volcengine/OpenViking (context database for agents, 19.3k stars this month), it’s clear that agent memory infrastructure is becoming its own category.
github.com/supermemoryai/supermemory | TypeScript
7. Crosstalk-Solutions/project-nomad (17.2k stars, +13,848 this week)
What it is: A self-contained, offline survival computer packed with critical tools, knowledge bases, and local AI. Designed to work without internet access.
Why it matters: The offline-first AI movement is growing. Between this and moeru-ai/airi (35.7k stars, self-hosted AI companion), there’s a clear trend: people want AI that runs on their hardware, without cloud dependencies. Call it digital sovereignty or just paranoia, but the stars don’t lie. 13,848 in a single week for an offline tool.
github.com/Crosstalk-Solutions/project-nomad | TypeScript
8. jarrodwatts/claude-hud (13.6k stars, +6,816 this week)
What it is: A Claude Code plugin that shows what’s happening in real time: context usage, active tools, running agents, and todo progress.
Why it matters: When your coding agent runs 47 subagents and you can’t see what any of them are doing, you have an observability problem. claude-hud is the DevTools for agent workflows. As agents get more autonomous and long-running, monitoring them becomes a necessity, not a nice-to-have. This is the beginning of “agent ops.”
github.com/jarrodwatts/claude-hud | JavaScript
9. NousResearch/hermes-agent (13.5k stars, +4,365 this week)
What it is: An agent framework from Nous Research built around the idea that the agent should grow with you. Self-evolving, learning from interactions over time.
Why it matters: Nous Research is known for fine-tuned open models (Hermes series). Them shipping an agent framework signals that model companies are moving up the stack. The “agent that grows with you” pitch is essentially personalized AI: an agent that gets better the more you use it, because it remembers what worked.
github.com/NousResearch/hermes-agent | Python
10. langchain-ai/open-swe (8.5k stars, +2,531 this week)
What it is: An open-source asynchronous coding agent from LangChain. Designed to handle software engineering tasks end-to-end.
Why it matters: LangChain building their own SWE agent means the framework layer and the application layer are merging. open-swe competes with Devin, SWE-agent, and similar tools, but as a fully open-source offering. The “async” part is key: it runs in the background while you do other things, then comes back with a pull request. That’s a fundamentally different workflow from pair-programming with a chatbot.
github.com/langchain-ai/open-swe | Python
Five Patterns Worth Watching
1. Agent Skills Are the New Packages
The top two repos on all of GitHub are agent skill frameworks. Not models. Not apps. Infrastructure for teaching agents reusable capabilities. Skills, not prompts. This is the shift from “write a good prompt” to “build a good skill library.”
2. Claude Code Is Spawning an Ecosystem
Five of the top monthly repos are Claude Code-specific tools. A monitoring dashboard, a skills library, a best-practices guide, a “build your own” tutorial, and a performance optimizer. This level of ecosystem activity around a single tool is unusual. Cursor had this moment. VS Code had this moment. Now it’s happening for agent harnesses.
3. Financial Agents Have Real Demand
42k stars on a trading agent framework is not a toy. The Chinese fork hitting 21k stars independently confirms the demand crosses markets and languages. If you’re building anything in fintech, multi-agent architectures for analysis and decision-making are where the energy is.
4. Memory Is the Missing Piece
Three separate repos in the monthly charts are specifically about agent memory: supermemory, OpenViking, and the memory subsystems in deer-flow and superpowers. Agents that forget everything between sessions are annoying. The race is on to build the persistence layer that makes agents actually useful over time.
5. Offline and Self-Hosted AI Is Growing
project-nomad and airi both trended hard. People want AI that works without an internet connection, runs on their hardware, and doesn’t phone home. This isn’t just a privacy play. It’s also about reliability: if your AI tools depend on an API that has a rate limit, you have a fragile workflow.
What This Means for Builders
If you’re building AI products right now, the signal from GitHub is loud and consistent:
- Don’t build another chatbot. Build agent infrastructure. Skills, memory, monitoring, orchestration.
- Vertical agents win. The generic frameworks exist. The opportunity is in domain-specific agents (finance, legal, healthcare, DevOps) that encode real expertise.
- Memory is a product. If you can solve agent memory better than the current options, you have a business.
- Observability is early. claude-hud got 6,800 stars in a week. The “Datadog for AI agents” hasn’t been built yet.
The GitHub trending page used to be libraries and frameworks. Now it’s agents all the way down. The question isn’t whether this trend continues. It’s whether you’re building on it, or watching it happen.