Mercury Skills: 131 Ready-Made Recipes for AI Agents That Don't Want to Start from Scratch

So here's Mercury Skills — cosmicstack-labs' shot at fixing a stupid but annoying problem: every time you build an AI agent, it starts from scratch. Doesn't know how to write a decent prompt, no framework for memory management, can't even optimize tokens. Frustrating as hell.

That's where this library steps in. It's not another random pile of AI-generated code on GitHub (there are already way too many of those). It's 131 hand-curated "skills" built against real developer and business workflows. React patterns, Docker best practices, memory management, token budgeting, even how to handle restaurant table management. Yep, really.

The kicker: one skill works everywhere. One SKILL.md format runs on Mercury, Claude Code, Codex, OpenClaw, Hermes, Cursor, Gemini — no forking per agent. You install with one command (mercury skills install frontend/react-patterns) and you're done. Your agent picks it up automatically.

Category-wise: 23 buckets, from Development (16 skills) down to Health & Wellness (5), with DevOps, Backend, Frontend, Creative, Finance, Automation, even Media Download with yt-dlp wrappers. They're organized decently, searchable on skills.mercuryagent.sh with a leaderboard, bookmarking, and a JSON API if you want to wire them into your own tools.

The philosophy: curated, not crowded. Zero AI slop, no duplicates. Every skill was written against real problems — not generated. They're atomic, versioned, and you can roll back if something breaks. MIT licensed, PRs welcome. It's the kind of thing that could actually become the "package manager" for the AI agent world, if it isn't already headed that way.