Claude Skills vs Slash Commands
Both give you shortcuts in your AI agent. But they work very differently — and understanding the distinction saves you hours of getting inconsistent output from quick commands.
Slash Commands
Short prompts you type once to trigger a quick action. Works in-session. No persistent state — the agent figures out what to do each time.
Best for
Quick, well-understood, one-off tasks
Claude Skills
Persistent instruction files installed into your agent. Loaded automatically into every session. Define a full process, required inputs, and structured output format.
Best for
Repeatable workflows you do more than once
Side-by-side comparison
| Dimension | Slash Commands | Claude Skills |
|---|---|---|
| Setup | Type it in the chat | Install once with npx skills add |
| Persistence | In-session only | Permanent — loads every session automatically |
| Process definition | None — agent improvises | Full multi-step process defined in the file |
| Required inputs | Not enforced | Skill asks for missing inputs before starting |
| Output format | Variable | Defined — same structure every time |
| Reuse effort | Retype each time | Zero — always available |
| Complexity ceiling | Low — gets unwieldy fast | High — supports reference files and sub-processes |
| Shareable | No | Yes — publish to GitHub, others can install |
| Versioned | No | Yes — tracked in git alongside your codebase |
How each one works
1 Slash commands: quick triggers, no memory
Slash commands work like shortcuts in the chat. Type /summarise and the agent figures out what that means based on your conversation context. Useful for short, well-understood requests where the agent already has everything it needs.
The limitation: no process, no input validation, no structured output. You get whatever the model decides to do. For simple tasks this is fine. For anything complex or repeated, you'll get different results each time.
2 Skills: permanent process definitions
A Claude Skill is a SKILL.md file installed to ~/.claude/skills/. Claude Code reads every installed skill at session start and uses those instructions when the relevant work comes up. No typing needed — skills activate automatically when you describe the task.
A well-written skill defines: when to activate, what inputs to ask for if missing, a specific step-by-step process, and the exact output format. The agent follows it consistently — not just once, but every time, for every session.
When to use each
Use a slash command when…
- · The task is simple and quick (< 30 seconds)
- · You only need it once or twice
- · No specific input gathering is needed
- · Output format doesn't matter much
- · You just need a quick nudge to the agent
Use a Claude Skill when…
- ✓ You do this task more than a few times
- ✓ Consistent, structured output matters
- ✓ The agent needs specific context before starting
- ✓ Multiple steps or checks are involved
- ✓ You want to share the process with a team
Real examples
Translate a paragraph
Use commandCommand approach
Slash command works fine — quick, one-off, no structure needed.
Skill approach
Overkill. You don't need a process file for a 5-second task.
Review a pull request
Use skillCommand approach
Works once, but every review will have a different structure. You'll miss the same categories of issues repeatedly.
Skill approach
Ideal. A PR review skill defines check order (security → correctness → performance), output format, and what "done" looks like.
Write a cold email
Use skillCommand approach
You'll get something, but it won't ask about your target audience, pain point, or CTA — so it'll be generic.
Skill approach
Ideal. The skill gathers recipient context, enforces word count, and produces subject line + body + follow-ups in one structured output.
Summarise a meeting
Either worksCommand approach
Works for a quick informal summary.
Skill approach
Worth it if you run meetings regularly and want consistent format: action items, owners, deadlines, decisions.
Do skills and commands work together?
Yes. A skill can define its own trigger phrase — effectively creating a named command with a full process behind it. Type review this PR and your installed code review skill activates with all its structured logic. You get the convenience of a command and the consistency of a skill.
Think of skills as "commands with process definitions." For simple tasks, plain commands are fine. For anything you do more than once with quality expectations, a skill is the right tool.
Frequently asked questions
Are slash commands built into Claude Code?
Claude Code has a set of built-in slash commands for common operations. You can also define custom commands in your project or user settings. Skills are separate — installed as files, not commands — but they can effectively function as named triggers for complex processes.
Do I have to type anything to activate a skill?
No. Well-written skills define their own trigger conditions and activate when you describe the relevant task naturally. You don't have to type a command — the agent recognises the work type from your description and applies the skill automatically.
Can I convert a slash command into a skill?
Yes, and for frequently used commands this is usually worth doing. Take the task you're triggering, write out a proper process definition (what inputs to gather, what steps to take, what the output should look like), and save it as a SKILL.md. You get the same trigger with much more consistent results.
How many skills can I install?
There's no hard limit. Claude Code reads all installed skills at session start. In practice, 5–15 skills covering your core repeated tasks is a reasonable working set. Beyond that, be selective — each skill adds context overhead.
Related comparisons
Browse 405+ Claude Skills
Install a skill and replace your most-used commands with consistent, structured processes.