Today's GitHub trending list tells a coherent story: "skills" are emerging as the dominant unit of distribution for agentic work. Whether it's a broad harness, a domain-specific toolkit, or a tiny focused module that fixes one annoying output problem, developers are treating reusable agent capabilities the way they once treated npm packages — small, composable, shareable.
affaan-m/ECC: Agent Performance Optimization at Scale
ECC describes itself as an agent harness performance optimization system with skills, instincts, memory, and security built in. It's one of the largest agent-infrastructure projects on GitHub by raw star count, sitting at over 254,000 total and adding nearly 1,500 in a single day. The framing — "instincts" alongside skills and memory — suggests an attempt to model something closer to trained reflexes than explicit rules, which is an interesting architectural bet as the field debates how much agent behavior should be hard-coded versus learned.
obra/superpowers: A Methodology, Not Just a Framework
Superpowers bills itself as both an agentic skills framework and a software development methodology, which sets it apart from tools that just wrap prompt templates. At 283,000 stars — one of the highest totals in the trending list — it's clearly found traction beyond any single workflow. The project's premise is that reliable agentic software development requires opinionated conventions alongside the tooling, not just more capabilities piled on top of a raw LLM call.
openai/skills: Codex Gets a Catalog
OpenAI published its own official Skills Catalog for Codex, providing a structured collection of reusable capabilities for its coding agent. The repository gained nearly 500 stars today, crossing 26,700 total. It signals that first-party skill distribution is becoming a real product surface for the major labs — not just a community artifact — and raises the question of whether skills libraries will become a platform differentiator the way app stores once were.
ayghri/i-have-adhd: A Skill That Fixes One Specific Thing
Not every trending repo aims to be infrastructure. The i-have-adhd skill for Claude Code does exactly one thing: it stops the coding agent from burying the actual answer deep in a long explanation. Framed as an ADHD-friendly output mode, it's accumulated over 32,000 stars and added 656 today. Its success is a useful data point — sometimes the highest-value tooling improvement isn't a new capability, it's a constraint on an existing one.
cathrynlavery/diagram-design: Editorial Diagrams for Coding Agents
This repository collects 38 editorial diagram types — flowcharts, decision trees, timelines, and more — implemented as self-contained HTML and SVG, designed to slot into Claude Code, Codex, and similar agent environments. At 35,800 stars with over 700 added today, it addresses a gap that becomes obvious once you use agents for documentation or technical communication: the defaults are rarely production-quality. By packaging polished, reusable diagram primitives as skills, it makes a visual standard portable across agent environments.