A plain-English shortlist of what builders are starring and shipping in the agent skills ecosystem — skills, harnesses, and toolkits on GitHub. No README paste, no ownership claims: just curated explainers and links.
Updated · 10 curated projects
Digest waitlist — plain-English picks of the hottest AI agent skills and harnesses. Not a newsletter blast dump; a short daily shortlist when we start sending.
Editorial ranking for builders scanning the ecosystem. Star counts are a signal, not the whole story — read the summaries, then open the repos. Live stars refresh daily via GitHub; explainers stay curated. Prefer search and filters? Use the AI Agent Skills Finder.
affaan-m/ECC
Think of ECC as an operating system for coding agents: a big library of reusable skills, specialist agents, hooks, and safety checks that plug into Claude Code, Codex, Cursor, and friends. Instead of reinventing the same “plan → build → review → secure” loop every project, teams install a shared playbook. It is less a single prompt and more a harness that keeps agents consistent across tools.
View on GitHubanthropics/skills
Anthropic’s public skills repo is the clearest “hello world” for the Agent Skills idea: folders with a SKILL.md, optional scripts, and progressive disclosure so the model only loads depth when needed. If you are writing your first skill or teaching a teammate what the format looks like, start here. It is also the closest thing to an official pattern book for how skills should be structured.
View on GitHubaddyosmani/agent-skills
A curated set of production-minded skills for coding agents — define, plan, build, verify, review, ship — written like checklists a senior eng would actually use. The value is specificity: clear exit criteria and verifiable steps instead of vague “be careful” advice. Great when you want agents to behave more like a disciplined teammate than a chatty autocomplete.
View on GitHubbrowser-use/browser-use
Browser Use is the popular open stack for letting agents click, type, and navigate real web pages — the practical cousin of “computer use” demos. Teams use it to scrape, fill forms, and test flows without writing brittle Selenium scripts by hand. If your agent needs to leave the IDE and work on the live web, this is one of the default starting points.
View on GitHubcloudflare/security-audit-skill
A coding-agent skill that turns your model into a structured security auditor: recon, parallel hunting, validation, reporting, and machine-readable findings. It is the single-repo version of ideas Cloudflare has written about for vulnerability harnesses. Useful when you want repeatable audits with evidence — not a one-off “please find bugs” prompt.
View on GitHubvercel-labs/skills
The open CLI for installing and syncing Agent Skills across Claude Code, Codex, Cursor, OpenCode, and many more. One `npx skills add` command maps a GitHub skill pack into the right folders for each harness. If the skills ecosystem is the content, this repo is the package manager people keep recommending.
View on GitHubvercel-labs/agent-browser
Agent Browser is Vercel Labs’ take on giving coding agents a reliable browser surface for navigation and checks. It sits in the same family as browser-use, but oriented toward agent workflows and modern web app testing. Reach for it when you want agents to verify UI behavior instead of guessing from source alone.
View on GitHubtrycua/cua
CUA focuses on scaling computer-use: open drivers, cross-OS fleets, and benchmarks for training and evaluation. Where browser-only tools stop at the tab, CUA aims at broader desktop and environment control. It is infrastructure for teams building or evaluating agents that operate real machines, not just chat APIs.
View on GitHubopenai/openai-agents-python
OpenAI’s Python Agents SDK is a harness for building multi-step agents with tools, handoffs, and tracing. It is the “framework” side of the stack: you write code that orchestrates models rather than dropping Markdown skills into a coding IDE. Pair it with skills when you want both procedural knowledge and a programmable runtime.
View on GitHubvercel-labs/json-render
json-render is Vercel Labs’ generative UI framework: agents emit structured JSON, and your app turns that into real components safely. It is a skills-adjacent idea — teaching models to speak a constrained UI dialect instead of freeform HTML. Worth watching if your product is an agent that should draw interfaces, not just text.
View on GitHub
