Claude skills vs agents (subagents)
A skill teaches Claude a procedure. A subagent hands a job to a separate Claude worker with its own context.
Our advice: use a skill when you keep giving Claude the same instructions or checklist. Use a subagent when a side job would flood your conversation with search results or logs.
Here, agents means subagents, the Claude Code feature for delegating work to a separate Claude. A skill is a folder with a SKILL.md file that Claude loads when a request matches. A subagent runs in its own context window with its own instructions and tool access, then returns a summary. They work well together.
Good for
- Claude users deciding whether to write a skill or set up an agent
- Claude Code users with long sessions and crowded context
- Teams packaging their procedures for everyone to reuse
Side by side
Our advice: Skills add know-how to the conversation you are in. Subagents move work out of it.
| Option | Skill | Subagent |
|---|---|---|
| What it is | Instructions in a SKILL.md folder | A separate Claude worker |
| Context | Loads into your chat; context: fork runs it in a subagent | Its own window |
| Own prompt, tools, permissions | Can pre-approve tools (allowed-tools) | Yes |
| Started by | Claude, or you with /name | Claude, by its description |
| Context cost when idle | Name and description only | Description only |
| Bundled files and scripts | Yes | Can preload skills |
| Extra usage | Body loads when used | Own requests count |
| Open standard | Yes, Agent Skills | Claude Code feature |
| Works in | claude.ai, Claude Code, API | Claude Code |
| Best for | Repeatable procedures | Side jobs with big output |
Which to pick
Our advice: Write a skill when you paste the same checklist into chat more than once. Define a subagent when the same kind of side job keeps flooding your main conversation. If a task needs both, a subagent can do the bulky work while a skill supplies the procedure.
Skills do not isolate work, and subagents work inside a single session.
Best for
- Skill: a release checklist, report format or review procedure you reuse
- Skill: sharing know-how with a team across claude.ai and Claude Code
- Subagent: searching a large codebase without filling your main chat
- Subagent: giving a job a limited set of tools or a cheaper model
Skip it if
- Skill: you need the work isolated from your main conversation
- Subagent: you only need a repeatable set of instructions, which a skill covers
Questions
What is the difference between a skill and an agent in Claude?
A skill is a set of instructions Claude loads when a request matches. A subagent is a separate Claude worker with its own context window, system prompt and tool access. The skill changes what Claude knows; the subagent changes where the work happens.
When should I use a skill instead of a subagent?
Use a skill when you keep pasting the same instructions, checklist or multi-step procedure into chat. Anthropic suggests this in its Claude Code docs. A skill's body loads only when it is used, so long reference material costs almost nothing until needed.
When should I use a subagent?
When a side task would flood your conversation with search results, logs or file contents you will not reuse. The subagent does that work in its own context and returns only a summary.
Can a skill use a subagent?
Yes. Anthropic's docs list subagent execution as a Claude Code extension to the skills format. Skills can also be invoked by name or used automatically when relevant.
Do skills and subagents use more of my usage limit?
Both can. A subagent sends its own requests, which count toward the same usage limits as your main conversation. A skill adds its instructions to the conversation when it loads.
Are skills only for Claude Code?
No. Agent Skills work in claude.ai, Claude Code and the Claude API, and follow an open standard that works across multiple AI tools. Subagents, as described here, are a Claude Code feature. See agent skills.
How does Claude decide when to use a skill or subagent?
By the description you write. Claude matches your request against each skill's description, and against each subagent's description when deciding whether to delegate. Write them to say what the thing does and when to use it.
Can I limit what a subagent can do?
Yes. A custom subagent has its own tool access and permissions, so you can restrict which tools it uses. You can also route it to a faster, lower-cost model such as Haiku.
Is Claude Code the only place with subagents?
Anthropic documents custom subagents in Claude Code. Other tools have their own agent features, so check each tool's docs. See Claude Code.
Sources
Docs, issue trackers and discussions this page is built from.
Show all 3Hide
- Extend Claude with skills · code.claude.com
- Create custom subagents · code.claude.com
- Agent Skills · platform.claude.com