# Cursor Rules vs Skills

> Pick between a Cursor rule and a skill with a 4-question decision tree: always-on vs on-demand, short ban vs long playbook, path-scoped vs task-scoped.

Canonical: https://www.agentickit.co/compare/cursor-rules-vs-skills-decision
Last updated: 2026-11-18

If the instruction must fire on every turn or on a specific file path, write a rule. If it's a longer playbook the agent should pull in only when the task matches, write a skill. Most solo founders start with ten rules and end up moving eight of them into skills once the rules file gets too long to read. Last updated 2026-10-26 by Rakshit Yadav.



## The decision tree
<ol>- **Does it need to apply on every single turn, with no exceptions?** Rule. (Tenant isolation checks, "never commit .env", output format bans.)
- **Does it only apply when editing files matching a path, like app/api/** or *.sql?** Rule, scoped with a glob in frontmatter.
- **Is it longer than a screen, or does it read like a how-to rather than a ban?** Skill. Stripe webhook setup, RAG chunking strategy, and on-page SEO checklists are playbooks, not rules.
- **Should the agent decide on its own when to pull this in, based on what the task looks like?** Skill. A rule can't be selectively loaded by task description; a skill's own description field is exactly that selector.
</ol>If you answered "rule" to questions 1 or 2, stop there. If not, it's a skill. Nothing here should be both — see "Don't duplicate" on the fuller [Rules vs Skills vs Commands](https://www.agentickit.co/compare/cursor-rules-vs-skills) page.

## Why founders get this wrong
The common mistake is writing every project convention as a rule because rules were the first mechanism Cursor shipped and `.cursorrules` is what most tutorials still show. That works until the rules file hits a few hundred lines and loads on every single message regardless of relevance, burning context budget on things like a webhook integration guide that's only relevant twice a month. The fix isn't deleting the content — it's moving the long-form, situational parts into skills and leaving only the always-true bans behind.

## A concrete example
A SaaS founder's rules file has: "never expose the service-role Supabase key to the client" (keep as a rule — always true, short), and a 40-line walkthrough of how this project's Stripe webhook handler is structured (move to a skill — long, only relevant when touching billing code). After the split, the rules file loads three lines on every turn instead of forty, and the Stripe context only loads when a billing task actually matches.



## FAQ

### Can I just keep everything as rules and skip skills entirely?
You can, but every rule loads on every turn regardless of relevance, so a growing rules file quietly eats context budget. Skills load only when the task matches, which is why long or situational content belongs there instead.

### Does a skill override a rule if they conflict?
Don't rely on precedence. If a rule and a skill say different things, you have a bug, not a priority order. Keep each instruction in exactly one file.

### How do I convert an existing .cursorrules file into skills?
Grade it first with the Rules File Grader to see what's actually a playbook versus a true always-on ban, then move the playbook sections into .cursor/skills/<name>/SKILL.md files.

## Related
- [Cursor Rules vs Skills vs Commands (full breakdown)](https://www.agentickit.co/compare/cursor-rules-vs-skills)
- [Cursor Skills vs Commands](https://www.agentickit.co/compare/cursor-skills-vs-commands)
- [The same question for Claude Code: CLAUDE.md vs Skills vs Subagents vs Hooks](https://www.agentickit.co/compare/claude-code-skills-vs-subagents-vs-hooks)
- [Rules File Grader](https://www.agentickit.co/tools/rules-grader)
- [Best Cursor skills: SKILL.md and 12 to install](https://www.agentickit.co/blog/best-cursor-skills)
- [Cursor Skills pillar](https://www.agentickit.co/skills)
