# EP08 Website Draft

## Goal

Make subagent boundaries visible as an operating model, not a magic spawn feature.

## Hero

**Subagents are about ownership and isolation, not about multiplying noise.**

## Page Structure

1. collaboration plane overview
2. taxonomy and write scope
3. context isolation experiment
4. tool-limit comparison
5. anti-patterns
6. official links

## Interactive Blocks

### 1. Taxonomy Map

Show how a task moves from main thread to bounded sub-agent and back as a summary.

### 2. Scope Isolation Demo

Let the reader switch between:

- shared context
- isolated sub-agent context
- summary returned to main thread

### 3. Tool Limit Table

Compare:

- broad access
- bounded access
- noisy handoff

Each row should answer:

- what it is good for
- what it risks
- how to summarize it back

## Minimum Viable UX

- structured, operator-like visual language
- clear scope labels
- minimal but meaningful animation
- mobile-safe rows and cards

## Teaching Rules

- Every sub-agent must map to an artifact path.
- Every summary must be decision-ready.
- Every example must show what stays out of the main thread.

## Official Links

- sub-agents
- context window and compaction
- memory

## Exit Outcome

The reader should know how to split work without losing ownership or context quality.