https://cursor.com/changelog/2-4
Cursor has released support for Skills, Subagents, and image generation. It’s time to redo commands and rules ¯\_(ツ)_/¯
Skills are an implementation of the Agent Skills specification. In terms of functionality, they don’t differ much from rules with the “Apply Intelligently” type. However, they allow covering both automatically invoked actions and those invoked only manually (disable-model-invocation=true).
Subagents are an attempt to more accurately select context for a specific development stage that can be isolated without “cluttering” the main agent’s context. Plus, more suitable models can be used. Out of the box, these are:
explore(searches and analyzes codebases)bash(runs series of shell commands)browser(controls browser via MCP tools)
You can write your own subagents. They are the same markdown files with frontmatter and a set of their own properties: name, description, model (you can use fast, inherit, or a specific model), readonly, is_background. For now, you cannot specify a list of tools, but this might be added in the future.
In my projects, I implemented subagents through agent skills as a convenient format. But the Cursor developers decided to make their own for some reason.
It turns out Hooks were released earlier. I’ll go read and experiment.