一键导入
agent-teams
Orchestrate multi-agent teams in Claude Code — set up coordinated sessions with task delegation, inter-agent communication, and parallel execution
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Orchestrate multi-agent teams in Claude Code — set up coordinated sessions with task delegation, inter-agent communication, and parallel execution
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Leverage Claude Code's ultra-deep planning mode for complex architecture decisions, multi-file refactors, and system design
Deep code review using Claude Code's thorough analysis mode — security, performance, correctness, and maintainability
Safe isolated testing environment for multi-agent swarm topologies before production deployment
Automatically summarizes the current session context to prevent token window overflow.
Claude Code context pruner: slash command to summarize session and reset token bloat
Produces a campaign performance summary from session-log.md entries. Reports open rates, reply rates, and meeting conversion rates per sequence. Flags top performers (reply rate >10%) and underperformers (reply rate <3%). Recommends one concrete change per underperforming sequence.
| name | agent-teams |
| description | Orchestrate multi-agent teams in Claude Code — set up coordinated sessions with task delegation, inter-agent communication, and parallel execution |
Assess team viability. Is the work parallelizable? Do tasks touch different files? If yes to both, proceed.
Define roles. Each teammate needs a clear role, scope, and boundary:
Teammate 1: "Frontend Engineer" — React components, styling, client-side logic
Teammate 2: "Backend Engineer" — API endpoints, database queries, validation
Teammate 3: "QA Engineer" — Integration tests, edge cases, error handling
Set up the lead session. The lead coordinates — it does NOT write code:
Create a task list with clear ownership. Assign each task to a specific teammate.
Use the task list as the shared communication channel.
Define communication protocol. Teammates share findings via task updates, not direct messages:
Set conflict boundaries. Prevent file collisions:
Teammate 1 owns: src/components/**, src/styles/**
Teammate 2 owns: src/api/**, src/models/**
Teammate 3 owns: tests/**, src/utils/validation/**
Shared (read-only): src/config/**, package.json
Synthesize results. After teammates complete their work, the lead:
Common team patterns:
| Pattern | Teammates | Best for |
|---|---|---|
| Layer Split | Frontend + Backend + Tests | Feature development |
| Hypothesis Race | 2-3 investigators | Debugging complex issues |
| Research Panel | 3-4 researchers + synthesizer | Architecture decisions |
| Review Panel | Security + Performance + DX | Code review |
| Module Factory | N builders (1 per module) | Large scaffolding |
Lead: "Build a user authentication module. Teammate assignments:
- @frontend: Login form, auth context, protected routes (src/components/auth/**)
- @backend: Auth API, JWT generation, session management (src/api/auth/**)
- @qa: Auth integration tests, token expiry tests (tests/auth/**)
Rules:
1. Don't touch files outside your scope
2. Update your task when done
3. If you need something from another teammate, add a comment — don't block
Let's go."
[Teammates work in parallel]
Lead (after all done):
"Reviewing changes... frontend and backend agree on the /auth/login contract.
Tests cover both success and failure paths. All integration tests pass.
Merging into single commit: feat(auth): add JWT authentication module"
Built with Claudient · Claude Code