Three mechanisms for routing different tasks to different LLM models:

1. delegate_task(model=..., provider=...)

Route individual subagent calls to any model:

delegate_task(goal='summarise docs', model='google/gemini-flash-1.5')

2. Skill frontmatter model: field

Skills can declare their own model in SKILL.md frontmatter. When loaded via delegate_task(skill='...'), the skill's model is used automatically.

---
name: code-review
model: anthropic/claude-opus-4-6
---

3. Config aliases

delegation:
  supervisor_model: anthropic/claude-opus-4-6   # main orchestrating agent
  execution_model: google/gemini-flash-1.5       # default for all subagents