Skip to content
Back to skills

Multi Agent Coordinator Majiayu000

ASecurity

Use when executing implementation plans with independent tasks - dispatches fresh subagent for each task with code review between tasks, enabling fast iteration with quality gates. Supports orchestrator, peer-to-peer, and pipeline coordination modes.

  • 57 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added October 17, 2026
ai-agentsgo

Works with

  • claude code

Security analysis

A100/100

Scanned October 17, 2026

npx -y skills add diegosouzapw/awesome-omni-skill --skill multi-agent-coordinator-majiayu000 --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Multi Agent Coordinator Majiayu000?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Multi Agent Coordinator Majiayu000
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/diegosouzapw-multi-agent-coordinator-majiayu000/badge)](https://www.skillsdirectory.com/skills/diegosouzapw-multi-agent-coordinator-majiayu000)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: multi-agent-coordinator
description: Use when executing implementation plans with independent tasks - dispatches fresh subagent for each task with code review between tasks, enabling fast iteration with quality gates. Supports orchestrator, peer-to-peer, and pipeline coordination modes.
metadata:
  short-description: 多代理协调与编排
  keywords:
    - multi-agent-coordinator
    - 多代理
    - 协调器
    - 并行处理
    - 任务编排
    - 工作流
    - 任务分发
    - 结果聚合
    - subagent-driven-development
    - coordination
  category: 架构设计
  author: Bensz Conan
  platform: Claude Code | OpenAI Codex | ChatGPT
---

# Multi-Agent Coordinator - 多代理协调专家

目标:把“复杂开发任务”拆成可并行的独立任务,分派给子代理执行,并在任务间做质量门禁(代码审查 + 轻量测试),最后聚合成一致的可交付结果。

为满足社区推荐的 `SKILL.md` 500 行以内约束:完整细节/长示例已下沉到 `awesome-code/agents/multi-agent-coordinator/references/legacy-skill-full.md`。

## 核心原则(Subagent-Driven Development)

- 任务拆分要原子:每个任务有明确输入/输出/验收标准
- 每个任务用“全新子代理”:降低上下文污染与确认偏差
- 任务之间强制门禁:至少一次代码审查;必要时补回归测试
- 结果必须可聚合:统一术语、接口约定、日志口径与错误处理风格

## 何时使用

- 用户给出明确实施计划/任务清单,需要并行推进
- 任务跨度大(多文件/多模块/多领域),单线程容易遗漏或拖慢
- 需要严格质量门禁(合并前必须审查、必须验证)

## 输入

- 计划来源:用户文字 / `PLAN.md` / issue 列表 / TODO 列表
- 约束:时间、兼容性、目录边界、不可破坏性要求
- 验收:测试要求、性能目标、行为回归标准

## 输出

- 一个可执行的“任务编排表”(任务 → 负责人子代理 → 依赖 → 验收)
- 每个任务的结果摘要(改动点、风险、验证)
- 最终聚合报告(P0/P1/P2 风险 + 下一步)

## 工作流

1. 读取计划并生成任务清单
   - 将大任务拆成 3-15 个原子任务
   - 为每个任务写清:目标、范围、验收、风险、依赖

2. 选择协调模式
   - orchestrator:默认;中心协调器分派任务并统一口径
   - peer-to-peer:小团队/低耦合;允许子代理互相同步但必须记录决定
   - pipeline:强依赖链;按阶段推进(例如:设计→实现→测试→文档)

3. 分派任务(可并行)
   - 并行仅限“文件/模块不重叠”或“改动可安全合并”的任务
   - 明确要求:输出必须包含(a)改动说明(b)验证方式(c)潜在回滚点

4. 任务间门禁(强制)
   - 对每个任务结果做快速审查:安全/正确性/一致性/边界条件
   - 必要时补回归测试或最小验证步骤

5. 聚合与冲突解决
   - 先合并“口径/接口/命名/日志风格”,再合并代码
   - 如出现冲突:优先保持正确性与可读性;无法判定时暂停并向用户确认

6. 最终交付检查
   - 关键路径功能可跑通
   - 无明显安全/路径越界/敏感信息泄露
   - 产出文档与代码一致(如有)

## 协调器自检清单

- [ ] 任务拆分是否避免重叠修改同一文件?
- [ ] 是否为每个任务定义了可验证的验收标准?
- [ ] 并行任务是否有清晰的依赖边界?
- [ ] 每个任务是否经过最小审查与验证?
- [ ] 聚合后是否统一了术语与接口风格?

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.