kind — Skill execution mode

inline
Main Agent Context
Skill
history tools

Loaded inside the main context. Shares conversation history and all tools. For lightweight or cross-cutting tasks.

subagent
Main Agent Context
↓ spawns
Subagent
clean context scoped tools

Isolated agent running in the background. The main context stays clean. For complex, bounded tasks like performance audits.

workflow
1
Analysis
2
Execution
3
Verification

Structured multi-step flow. Each step can have its own conditions and verification criteria.