# AI Skill Scoping Deciding where an [[AI Agent Skills|AI skill]] lives: user-level, project-level, team-shared, or published. The scope determines who can use it, what context it assumes, and how it's maintained. ## Scope levels | Scope | Location (Claude Code) | Who uses it | Example | |-------|----------------------|-------------|---------| | **User** | `~/.claude/commands/` | You, across all projects | Writing style skill, personal review checklist | | **Project** | `.claude/commands/` or `.claude/skills/` | Anyone working on this project | Note writer skill, project-specific publisher | | **Plugin/Team** | Published plugin or shared repo | Team members who install it | Code review conventions, team PR template | | **Public** | Registry or marketplace | Anyone | Generic Markdown formatter, universal commit skill | ## How to choose **Make it user-level when:** - It reflects personal preferences (writing style, communication patterns) - It works identically across all your projects - It contains personal context (identity, voice, habits) **Make it project-level when:** - It depends on project structure (vault layout, folder conventions, tech stack) - It encodes project-specific business rules or domain knowledge - Other contributors to the project should use it too **Make it team/plugin-level when:** - Multiple projects share the same conventions - The skill needs coordinated updates across projects - It's part of team standards or org policy **Make it public when:** - It's genuinely generic (no project/team/personal assumptions) - Others would benefit and you're willing to maintain it - It doesn't leak proprietary knowledge ## Layering and precedence Scopes layer like [[Context File Hierarchy]]: user skills provide defaults; project skills specialize; plugin skills add team conventions. When scopes conflict, more specific wins: project overrides user, just like a subfolder CLAUDE.md overrides the root. ## Common mistakes - **Over-scoping**: putting personal preferences in project skills (forces your style on the team) - **Under-scoping**: keeping project-critical skills in your user directory (teammates can't use them) - **Leaking context**: a skill at public scope that assumes your vault structure or personal notes - **Duplication**: same skill copied across projects instead of shared at team level ## References - ## Related - [[AI Agent Skills]] - [[AI Skill Distribution]] - [[AI Skill Portability]] - [[AI Skill Composability]] - [[Context File Hierarchy]] - [[Context Inheritance]] - [[Context Layering]] - [[Claude Code Skills]] - [[Claude Code Plugins]] - [[Claude Code Memory]] - [[Personal Context Management (PCM)]] - [[Team Context Management (TCM)]] - [[Project Context Management (PCM)|Project Context Management]]