Two different jobs
A rule tells the agent how to behave. A skill teaches the agent how to do a specific task. Cursor’s docs describe a skill as “a portable, version-controlled package that teaches agents how to perform domain-specific tasks.”
| Aspect | Rule | Skill |
|---|---|---|
| Format | One .mdc file with frontmatter |
A folder with SKILL.md, optional scripts/, references/, assets/ |
| Where (project) | .cursor/rules/ |
.cursor/skills/ or .agents/skills/ |
| Where (personal) | User Rules in Customize | ~/.cursor/skills/ or ~/.agents/skills/ |
| How it gets used | Always, by globs, when the agent judges it relevant, or on @-mention | When the agent judges it relevant, or when you type / and pick it |
| Can bring scripts | No | Yes |
| Read by other tools | No, .mdc is Cursor’s format |
Cursor also loads .claude/skills/ and .codex/skills/ |
What a skill looks like
A skill is a folder whose name matches the name field in its SKILL.md. The frontmatter has two required fields and a few optional ones:
name(required): lowercase letters, numbers, and hyphens, equal to the folder name.description(required): what the skill does and when to use it. The agent reads this to decide relevance.paths(optional): glob patterns that limit the skill to matching files.disable-model-invocation(optional): whentrue, the skill only runs when you call it with/skill-name.icon,color, andmetadata(optional).
The same team knowledge looks different in each form. As a rule in .cursor/rules/migrations.mdc:
---
description: Database migrations
globs: db/migrations/**
alwaysApply: false
---
Never edit a migration that is already merged. Add a new one.
As a skill in .cursor/skills/add-migration/SKILL.md:
---
name: add-migration
description: Add a new database migration. Use when a schema change is needed.
---
1. Run scripts/new-migration.sh with a short name.
2. Write the up and down steps in the new file.
3. Run the migration locally and commit both files.
The rule states a standing agreement for matching files. The skill walks through a task and can call its own script.
When to use which
- A standing agreement (“use our error class”, “never commit secrets”, “answer in Dutch”): a rule. If it always holds, Always Apply. If it holds for some files, globs.
- A procedure with steps (“how we cut a release”, “how to add a database migration”): a skill. The agent discovers the skill from its name and description, then loads the full instructions when the task comes up.
- A procedure that needs a script (a check, a generator, a deploy helper): a skill, because rules cannot carry files.
- Something you only want on command: a skill with
disable-model-invocation: true, or a rule set to Apply Manually. - Something several AI tools must share: a skill in
.claude/skills/(Cursor and Claude Code both load it) or plain instructions inAGENTS.md.
Moving rules to skills
Cursor has a built-in command for this: /migrate-to-skills. It converts dynamic rules, meaning rules with alwaysApply false or empty and no globs, into skills. Slash commands, both user-level and workspace-level, become skills with disable-model-invocation: true. It does not migrate rules with alwaysApply: true or with globs, and it does not migrate user rules, because those are not stored on the file system. That split matches the advice above: standing rules stay rules, procedures become skills.
The common mistake: skills follow you to the team
A skill in ~/.cursor/skills/ reaches your own Cloud Agents only after you enable Sync Skills for Cloud Agents. Many people assume that means the team gets it too. The docs are explicit: synced skills stay private to you and are not shared with your team, and automation runs do not load them. To give a skill to the team, or to an automation, commit it to the repository’s .cursor/skills/ folder, or publish it to the team marketplace, where teammates install it. Only ~/.cursor/skills/ syncs to Cloud Agents; ~/.agents/skills/ does not.
What this means for a team
Rules and skills in the repository are shared; rules and skills on your own account are not. Share a personal skill by committing it to the repository or publishing it to the team marketplace, where teammates install it. For rule setup itself, see How do you use Cursor rules. For sharing across the team, see How do you share context and team rules in Cursor.
Where Bombos fits
Rules and skills hold what you write down on purpose. Cursor project rules use Cursor’s format; Agent Skills can work across compatible tools. Bombos is a plugin that gives every AI tool a team uses (Cursor, Claude Code, ChatGPT, Copilot, Grok, Codex, your own agents) one shared memory with team knowledge, rules it learns from corrections, and what already sits in email, documents, and repos, traceable, in the cloud or on your own server. One developer can try it first with their own agents.
Related questions
Can I convert my rules into skills?
Yes, with /migrate-to-skills. It converts dynamic rules (no alwaysApply and no globs) into skills, and slash commands into skills with disable-model-invocation: true. Rules with alwaysApply or globs, and user rules, are not migrated.
Does Cursor load Claude Code skills?
Yes. For compatibility Cursor also loads skills from .claude/skills/, .codex/skills/, ~/.claude/skills/, and ~/.codex/skills/.
Are my personal skills shared with my team?
Not by themselves. Skills in ~/.cursor/skills/ reach your own Cloud Agents only after you enable Sync Skills for Cloud Agents, and stay private there. To share a skill, commit it to .cursor/skills/ in the repository or publish it to the team marketplace.
Can a skill only apply to some files?
Yes. The optional paths field takes glob patterns that limit the skill to matching files.
