I’m very interested in the “Agent Skills” feature and would like to ask a few questions:
Official Release: When will Agent Skills be officially supported in the stable version of Cursor?
Best Practices: What are the currently recommended best practices for using Agent Skills? Are there specific examples of how to structure them for maximum efficiency?
Agent Skills vs. .cursorrules (.mdc): How should we distinguish between these two features in different scenarios? When should I use Agent Skills versus defining rules in .cursorrules (or the new .mdc files)?
Replacement: Is Agent Skills intended to eventually replace .cursorrules, or are they designed to coexist and serve different purposes?
Official release: Agent Skills isn’t ready to use yet. Even though the feature was added to the docs, the team confirmed it “wasn’t quite ready for primetime”. There’s no ETA yet. Please watch the forum for updates.
Agent Skills vs .cursorrules: These are different mechanisms:
Agent Skills is an open standard for portable, reusable agent skills. They work like “agent-decided rules”, meaning the agent decides when to apply a skill based on context. They’re defined in SKILL.md files.
.cursorrules (Rules) provide system-level instructions for the Agent. They support 4 apply modes: Always Apply, Apply Intelligently, Apply to Specific Files, Apply Manually. They’re defined in .cursor/rules/RULE.md.
When to use:
Agent Skills for portable, reusable skills that work across different projects
Rules for project-specific context, coding standards, and workflow
Replacement: No, they’re designed to coexist and serve different purposes.
I’d also recommend checking out Hooks (Hooks | Cursor Docs), a system to control the agent loop using scripts.
