team

name	team
description	AgentWorld development team. Trigger: '组建团队', '开会', '/team', 'build team', 'review architecture'
argument-hint	[task description]
disable-model-invocation	true

AgentWorld Development Team

10-agent expert team for building, reviewing, and growing the agent-world multi-agent simulation framework.

任务

$ARGUMENTS

可用 Agent

Tier 1: Strategic Architecture

Agent	文件	人物原型	职能	产出目录
Framework Architect	framework-architect	DHH	Plugin system, API design, conventions	docs/framework-architect/
LLM Architect	llm-architect	Andrej Karpathy	Brain plugin, prompt engineering, agent cognition	docs/llm-architect/

Tier 2: Core Engineering

Agent	文件	人物原型	职能	产出目录
Systems Engineer	systems-engineer	Werner Vogels	EventBus, tick loop, performance, resilience	docs/systems-engineer/
Cognitive Engineer	cognitive-engineer	Daniel Kahneman	Memory plugin, decision models, agent psychology	docs/cognitive-engineer/
Simulation Designer	simulation-designer	Will Wright	Scenario design, game mechanics, emergence	docs/simulation-designer/
Visualization Engineer	visualization-engineer	Bret Victor	2D rendering, interactive viz, WebSocket streaming	docs/visualization-engineer/

Tier 3: Quality Assurance

Agent	文件	人物原型	职能	产出目录
Quality Engineer	quality-engineer	Kent Beck	TDD, plugin testing, integration testing	docs/quality-engineer/
Contrarian Reviewer	contrarian-reviewer	Charlie Munger	Architecture review, finding flaws, inversion thinking	docs/contrarian-reviewer/

Tier 4: Growth & Community

Agent	文件	人物原型	职能	产出目录
DX Lead	dx-lead	Guillermo Rauch	Developer experience, docs, examples, onboarding	docs/dx-lead/
Community Strategist	community-strategist	Evan You	Open source growth, ecosystem, community building	docs/community-strategist/

预设编队 (Preset Formations)

1. Architecture Review 架构评审

场景：Major design decisions, new package structure, API changes 成员：framework-architect + systems-engineer + contrarian-reviewer 流程：Architect proposes → Systems validates feasibility → Munger finds flaws 产出：docs/framework-architect/architecture-review-{topic}.md

2. Brain Design 大脑设计

场景：Brain plugin improvements, prompt engineering, cognitive architecture 成员：llm-architect + cognitive-engineer + contrarian-reviewer 流程：Karpathy designs LLM approach → Kahneman validates cognitive model → Munger stress tests 产出：docs/llm-architect/brain-design-{topic}.md

3. New Scenario 新场景开发

场景：Creating a new simulation scenario (大逃杀, 股市, 三体 etc.) 成员：simulation-designer + llm-architect + visualization-engineer 流程：Wright designs mechanics → Karpathy designs agent brains → Victor designs visualization 产出：docs/simulation-designer/scenario-{name}.md

4. Plugin Development 插件开发

场景：Building a new plugin (interaction, world, brain, etc.) 成员：framework-architect + quality-engineer + systems-engineer 流程：DHH designs API → Beck writes tests → Vogels reviews resilience 产出：docs/framework-architect/plugin-{name}.md

5. Launch Prep 发布准备

场景：Preparing for npm publish, GitHub release, community launch 成员：dx-lead + community-strategist + quality-engineer 流程：Rauch polishes docs/README → Evan plans launch strategy → Beck verifies test coverage 产出：docs/dx-lead/launch-{version}.md

6. DX Audit 开发者体验审计

场景：Reviewing and improving developer experience 成员：dx-lead + framework-architect + contrarian-reviewer 流程：Rauch identifies friction → DHH proposes API improvements → Munger challenges assumptions 产出：docs/dx-lead/dx-audit-{date}.md

7. Full Review 全面评审

场景：Comprehensive project review before major milestone 成员：ALL 10 agents 流程：

• Batch 1 (architecture): framework-architect + systems-engineer + llm-architect
• Batch 2 (quality): quality-engineer + contrarian-reviewer
• Batch 3 (growth): dx-lead + community-strategist
• Batch 4 (design): simulation-designer + cognitive-engineer + visualization-engineer 产出：docs/*/full-review-{date}.md per agent

8. Memory System Design 记忆系统设计

场景：Designing or improving the agent memory architecture 成员：cognitive-engineer + llm-architect + systems-engineer 流程：Kahneman designs cognitive model → Karpathy maps to LLM constraints → Vogels ensures performance 产出：docs/cognitive-engineer/memory-design-{topic}.md

9. Visualization Sprint 可视化冲刺

场景：Building or improving the frontend visualization 成员：visualization-engineer + dx-lead + simulation-designer 流程：Victor designs visual layers → Rauch ensures onboarding UX → Wright ensures game feel 产出：docs/visualization-engineer/viz-{topic}.md

10. Community Growth 社区增长

场景：Planning content, outreach, and community events 成员：community-strategist + dx-lead + simulation-designer 流程：Evan plans strategy → Rauch aligns docs → Wright suggests viral scenarios 产出：docs/community-strategist/growth-{topic}.md

执行步骤

1. 分析任务，选择成员

• Match task to preset formation or custom-pick agents
• Selection principles:
  • Complementary styles (architect + critic, creative + pragmatic)
  • Core + Support (1-2 core + 1-2 support + Munger for major decisions)
  • 3-5 is optimal for most tasks, don't overdo
  • Always include contrarian-reviewer for architecture decisions
• Explain selection to user, then build immediately

2. 组建 Agent Team

• TeamCreate with kebab-case team_name (e.g., aw-brain-design)
• TaskCreate for each member with full context:
  • Read their agent definition file from .claude/agents/<name>.md
  • Include the full persona content in the task prompt
  • Specify input files to read and output directory
  • Reference any relevant existing docs
• Task spawn with:
  • subagent_type: general-purpose
  • Inject FULL agent file content in prompt
  • Tell agent about their docs/<role>/ output directory
  • Mention available skills they can use via /find-skills

3. 协调与汇总

• Parallel: spawn analysis agents simultaneously, collect all, merge findings
• Quality gate: if Munger flags a blocker, loop back for revision
• Produce structured summary report with each agent's key findings
• TeamDelete after completion

注意事项

• Chinese for communication, English for code and technical terms
• Each agent writes output to their docs/<role>/ directory
• Team is temporary — created per task, deleted after completion
• User (Jinming) is final decision maker on all design choices
• Disagreements between agents are VALUABLE — present all perspectives honestly
• Munger's objections should be addressed before proceeding, not dismissed