Module 2 · Composition and control · Lesson 4
Skills and subagents
Package procedures as skills the model loads on demand, delegate specialist work to subagents with isolated context, and reason precisely about what crosses each boundary, in Claude Code and in eve.
By the end of this lesson you can
- Write a skill whose description reliably triggers it, and explain progressive disclosure in terms of what is in context at each stage
- Define the same specialist subagent in Claude Code, the Agent SDK and eve, and state what each one inherits from its parent
- Decide between a skill, a subagent and a plain tool for a given task, using explicit criteria
- Compute when delegating to subagents saves tokens and when it costs more
- Identify what data crosses a delegation boundary and why delegation is not an approval boundary