Guide for distinguishing Claude Code features (Commands, Skills, Agents, Hooks, Rules) and selection criteria. Use when decomposing context files, deciding where to add new features, or customizing Claude Code.
/plugin marketplace add masseater/claude-code-plugin/plugin install mutils@masseater-pluginsThis skill inherits all available tools. When active, it can use any tool Claude has access to.
agents.mdcommands.mdhooks.mdreferences.mdrules.mdskills.mdGuide for choosing between 5 major Claude Code features.
| Feature | Trigger | Essence | Best For |
|---|---|---|---|
| Commands | Manual (/command) | Inject saved prompts instantly | Repeatable standard workflows by user execute |
| Skills | Auto (Claude decides) | Knowledge loaded on demand | On-demand knowledge (tool usage, guidelines) |
| Agents | Manual/delegation | Specialists with independent context | Bulk data processing & parallel tasks |
| Hooks | Event-driven | Shell scripts auto-executed around tools | Deterministic automation |
| Rules | Conditional/always | Modularized rules | Path-conditional constraints |
1. "Should it auto-execute on events like file edits?"
→ Yes: Hooks (implement as shell commands)
2. "Needs independent context for bulk data processing?"
→ Yes: Agents (don't pollute main conversation)
3. "Knowledge not always needed, but should surface when relevant?"
→ Yes: Skills (tool usage guides, domain expertise, patterns)
4. "Standard procedure user explicitly executes?"
→ Yes: Commands (instant /command execution)
5. "Constraints for specific paths or entire project?"
→ Yes: Rules (conditional with paths:)
6. Cannot classify
→ Keep in CLAUDE.md
See references.md.