Wiring Skills into Subagents
The Surprise: Subagents Don't See Your Skills
Here's something that surprises people: subagents don't automatically see your skills. When you delegate a task to a subagent, it starts with a fresh, clean context — and that context does not include the skills available in your main conversation. If you want a subagent to apply a skill, you have to wire it in explicitly.
Two key distinctions
Built-in agents (Explore, Plan) can't access skills at all. Custom subagents CAN use skills — but only the ones you explicitly list. And skills are loaded when the subagent STARTS, not on demand like in the main conversation.
The skills: Frontmatter Field
To give a custom subagent skills, list them in the skills: field of its agent markdown file (in .claude/agents/). The full content of each listed skill is preloaded into the subagent's context at startup — so the subagent carries that expertise into every task it handles.
---
name: frontend-reviewer
description: Reviews frontend code for accessibility and performance.
tools: Bash, Glob, Grep, Read, Skill
model: sonnet
skills: accessibility-audit, performance-check
---
You are a frontend code reviewer. Apply the preloaded skills to every review.Preloaded, not matched
Note the difference from the main conversation: in your main thread, a skill loads when its description matches your request. In a subagent, the skills listed in the skills: field are preloaded in full at startup — there's no matching step. First make sure those skills exist in .claude/skills/.
Two Directions, One Mental Model
Recall from the advanced section that skills and subagents combine two ways. This lesson is the second direction (subagent carries skills); the first (skill runs as a subagent via context: fork) we covered earlier. Keeping them straight prevents confusion:
| Approach | System prompt | Task | Skill role |
|---|---|---|---|
| Skill with context: fork | From the agent type | The SKILL.md content | The skill IS the task |
| Subagent with skills: field | The subagent's body | Claude's delegation message | The skills are REFERENCE knowledge |
context: fork → a skill runs as a subagent. skills: field → a subagent carries skills as reference. Different directions, complementary.
When to Wire Skills into Subagents
This pattern shines when you want isolated, expert task delegation with consistent standards baked in — rather than relying on prompts to remind a subagent of your conventions each time.
- •You want isolated task delegation with specific expertise (a reviewer that always applies your standards).
- •Different subagents need different skills (a frontend reviewer vs a backend reviewer).
- •You want to enforce standards in delegated work without depending on prompt wording.
- •Workflow: create the skills in .claude/skills/, then add the skills: field to a new or existing agent in .claude/agents/.
Section complete
You can now scope, share, and delegate skills. The final section is about operating skills day-to-day: troubleshooting, best practices, and a course wrap-up.
Key Takeaways
- ✓Subagents do NOT automatically see your skills — delegated work starts with a fresh context that excludes them.
- ✓Built-in agents (Explore, Plan) can't access skills at all; only custom subagents can, and only the ones you explicitly list.
- ✓List skills in a custom subagent's skills: frontmatter field (in .claude/agents/); their full content is preloaded at startup.
- ✓Subagent skills are PRELOADED (no matching step), unlike the main conversation where a skill loads when its description matches.
- ✓Two directions: a skill with context: fork runs AS a subagent (skill = task); a subagent's skills: field carries skills as REFERENCE.
- ✓Use this for isolated expert delegation with baked-in standards (e.g. distinct frontend vs backend reviewers) — create the skills first, then add the skills: field.
Check Your Understanding
Test what you learned in this lesson.
Q1.Do subagents automatically have access to the skills in your main conversation?
Q2.How are skills loaded into a custom subagent that lists them?
Q3.What's the difference between a skill using context: fork and a subagent using the skills: field?
Q4.When is wiring skills into a custom subagent most useful?
Practice This Lesson