Agent skills are the new plugins

If you have been building with AI coding agents for any amount of time, you have probably hit the same wall: your agent can do a lot, but it does not know how you work. It does not know your team conventions, your deployment process, or the twelve-step review workflow your org requires. That gap between what agents can do and what they actually know is where skills come in.

The past few months have seen an entire ecosystem spring up around this idea. There is now an open standard for agent skills backed by Anthropic and adopted by 25+ tools including Cursor, Claude Code, and Gemini CLI. Dedicated directories like skills.sh let you browse and install them the way you would browse npm packages. Marketplaces with hundreds of ready-made plugins are live. Full configuration bundles package agents, hooks, commands, and rules into a single install. The plugin era for AI agents is here, and it is moving fast.

Here is my take: skills are going to be the connective tissue between agents and real work. Not because agents are not capable on their own, but because the vast majority of times what is missing is not intelligence — it is context. Portable, version-controlled context that travels with you across tools.

Some of what is in this roundup is practical right now — install a skill, give your agent a new capability in seconds. Some of it is more forward-looking, like orchestrators that spin up parallel agents in isolated workspaces. Either way, these tools represent what I think is the most important shift happening in the AI tooling space: making agents actually useful for the specific way you and your team work.