20 AI Agent Skills With ~800K GitHub Stars That Form a Real Agent Stack

The weekend is just around the corner, and we know your fingers are already itching to build something even cooler than last time.

The repositories we’ll walk through today showcase a set of skills that are genuinely fun to experiment with, while also giving you a clear view into how agentic software is being built right now.

We’ll explore research workflows, codebase understanding, memory, design constraints, validation loops, and shipping tools, the pieces that teach an agent not just what to do, but how to work.

Well-designed skill packages a procedure the agent would otherwise improvise, and that alone can wipe out an entire category of repetitive work.

That’s why these 20 skills below form a surprisingly coherent loop:

See the problem. Map the system. Make the thing. Inspect what happened. Ship the next version!

The stack

Before getting into the details, I want to share the full list with you, grouped by layer, with approximate GitHub star counts as of this writing.

Editor’s note: Find the right AI role and ship what it asks for. Join the Agent Foundry.

Research

• agent-reach (85.0K)…