feat: update CLAUDE.md orchestrator philosophy for Agent Teams era

[JacobPEvans]

Parent

Part of #14 (epic: migrate workaround orchestration patterns to native Claude Agent Teams)

Summary

Update the CLAUDE.md orchestrator patterns and model routing rules to reflect the Agent Teams feature, providing guidance on when to use subagents vs agent teams, and how to structure team-based workflows.

Problem

Current CLAUDE.md describes an orchestrator philosophy designed around subagent limitations:

• "Think of yourself as a conductor, not a musician"
• Manual delegation via Task tool only
• Model routing table doesn't account for team roles
• No guidance on inter-agent communication
• No cost awareness for team vs subagent decisions

Proposed Changes

1. Update Orchestrator Role Section

Add Agent Teams as a first-class delegation option:

### Delegation Hierarchy

1. **Agent Teams** - For complex work requiring collaboration, debate, or cross-checking
2. **Subagents (Task tool)** - For focused, independent tasks where only the result matters
3. **Direct execution** - For simple tasks within the main context

2. Add Decision Matrix

### When to Use Agent Teams vs Subagents

| Criteria | Use Subagents | Use Agent Teams |
|----------|--------------|-----------------|
| Tasks need to communicate | No | Yes |
| Work is independent | Yes | Either |
| Need cross-checking | No | Yes |
| Cost sensitivity | High | Low |
| Task count | 1-3 tasks | 4+ parallel tasks |
| Duration | Short (<5 min) | Long (5+ min) |
| Complexity | Focused | Multi-perspective |

3. Update Model Routing Table

Add team role column:

Task Type	Cloud Model	Team Role	PAL MCP Tool
Team Lead	Opus 4.6	Lead (delegate mode)	N/A
Research	Sonnet 4.5	Researcher teammate	chat, clink
Implementation	Sonnet 4.5	Implementer teammate	codereview
Validation	Haiku 4.5	Verifier teammate	chat
Code Review	Multi-model	Multiple reviewers	consensus

4. Add Team Communication Patterns

### Team Communication Guidelines

- Prefer targeted `write` over expensive `broadcast`
- Use plan approval for risky changes
- Task list is the source of truth for work status
- Teammates should self-claim tasks, not wait for assignment
- Lead should focus on coordination, not implementation (delegate mode)

5. Update Token Conservation Section

Add team cost awareness:

### Agent Team Token Costs

- Each teammate is a separate Claude instance (~20K token startup)
- N teammates = ~Nx token cost
- Use teams only when parallel collaboration adds value
- Start with 3-5 teammates; scale based on task complexity
- Use Haiku for simple teammates to reduce costs

Acceptance Criteria

• Delegation hierarchy updated with Agent Teams
• Decision matrix for subagents vs teams
• Model routing table updated with team roles
• Team communication patterns documented
• Token cost guidance for teams added
• Commands table updated with new team commands
• Passes markdownlint validation
• Token limit compliance maintained

Scope

This issue only covers CLAUDE.md in this repo. Parent repos (~/git/CLAUDE.md) are out of scope.

Dependencies

• Should be done after: agent-teams-orchestrator plugin (feat: create agent-teams-orchestrator plugin with team lifecycle management #15) is created (so commands can be referenced)

[JacobPEvans]

Parent epic: #14 | Should follow: #15, #16, #17, #18, #20, #21