agentsclimarketplace

Coordinator

Skill Fantasymax/easy-build-skills/plugin/user-research-for-ai-config/skills/user-research-for-ai-config/templates/mode-b/with-subagents/coordinator

Let AI research you back—answer ~15 multiple-choice questions to get YOUR AI config (CLAUDE.md / AGENTS.md / custom skills), instead of copy-pasting templates. By FantasyMax.

Install
npx -y skills add Fantasymax/easy-build-skills --skill coordinator

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Coordinate a {N}-perspective parallel review of {B1 对象}. Use when the user asks for "{B7 多视角触发原话,如 '从 N 个角度看' / '多角度 review' / '并行评估'}" or any request implying multiple independent perspectives. Trigger especially when the user mentions {N 个视角的关键词,如 stability + performance + maintainability}. Do NOT use for: single-perspective requests (use the relevant individual subagent directly).

SKILL.md

4.3 KB, as published. Nobody here has run it

{Coordinator 标题:N 视角并行 review/分析}

这是 {用户名} 的"从 {N} 个独立视角并行处理 {B1 对象}"的 coordinator。本 skill 在 fork context 里跑 — 不污染主对话历史。 生成日期 {YYYY-MM-DD} | V1

Overview

{B1 一句话定义原话}

本 skill 派发任务给 {N} 个独立 subagent 并行跑:

  • {role-1} — {视角 1 边界}
  • {role-2} — {视角 2 边界}
  • {role-N} — {视角 N 边界}

各 subagent 各自在 isolated context 工作,最后返回 JSON,coordinator 汇总后输出主对话。

Workflow

[Phase 1:理解任务]
1. 接收用户输入({B2 输入材料})
2. 判断是否真的需要多视角:
   - 是 → 进 Phase 2
   - 否(用户其实只问一个视角) → 直接调对应 subagent,不走并行
   - 用户没说清楚 → ASK:「你想要哪几个视角?默认是 [N 视角列表]」

[Phase 2:派发并行]
3. 把用户输入分别发给 {N} 个 subagent(每个 subagent 拿到的是同一份原始输入)
4. 等待 {N} 个 subagent 全部返回(Claude Code 自动并行 + 自动等待)

[Phase 3:汇总]
5. 收到 {N} 份 subagent 输出(各自是 JSON / 结构化结果)
6. 按统一 schema 汇总:
   ```json
   {
     "{role-1}": [...],
     "{role-2}": [...],
     ...
     "summary": "整体判断(≤ 3 句)",
     "blocking_issues": [...],   // 跨视角的红线问题
     "suggested_actions": [...]
   }
  1. 输出给主对话

[Phase 4:用户回到主对话后] 8. 主对话只看到 summary(各 subagent 的细节在 fork 里) 9. 用户可问"{role-1} 具体说了什么" → coordinator 重新输出该 subagent 的明细


## Hard Rules

- MUST 派发**同一份原始输入**给所有 subagent(不要给每个 subagent 解读后的"摘要",否则会污染独立性)
- MUST 在汇总时**保留各视角的标识**(不要把 N 份合并成 1 段散文 prose)
- MUST 在 `blocking_issues` 里提取**跨视角共同标记**的问题(说明真的是 critical)
- NEVER 让 subagent 直接执行带副作用的动作(subagent 应该是"分析者",不是"执行者";执行需要回到主对话用户确认后由主对话动手)
- NEVER 在 fork 里调用 Skill tool 启动其它 skill(避免无限嵌套)
- ALWAYS 在汇总输出顶部给 summary 段(一句话),让用户能 1 行决定继续看不看明细

## When to ask the user

- Phase 1 step 2:用户没说清楚要哪几个视角时,列默认 + 让用户挑
- Phase 4:用户问明细时,问"是要 {role-1} 全文还是只要某条 finding 的解释"

## Output Format

Coordinator 返回主对话的格式(统一 schema):

```json
{
  "task": "{B1 对象一句话}",
  "perspectives": {
    "{role-1}": {
      "summary": "...",
      "findings": [
        { "severity": "critical|risk|smell", "detail": "...", "location": "..." }
      ]
    },
    "{role-2}": { ... },
    "{role-N}": { ... }
  },
  "summary": "整体判断(≤ 3 句)",
  "blocking_issues": [
    "issue-1 (出现于 {role-1} 和 {role-2})"
  ],
  "suggested_actions": [
    "action-1"
  ]
}

Reference files

  • examples/three-views-walkthrough.md — 完整 3 视角并行 + 汇总的实例
  • references/glossary.md — 跨视角共享的术语
  • 各 subagent 的 prompt 在 ~/.claude/agents/{role-N}.md

Notes from author

{B10 ① 收尾原话;如有用户重要补充}


Coordinator 与各 subagent 的关系(架构图)

主对话
  ↓ (调用 coordinator)
fork context (coordinator 跑)
  ├─→ subagent {role-1} (own fork) → JSON 1
  ├─→ subagent {role-2} (own fork) → JSON 2
  ├─→ subagent {role-N} (own fork) → JSON N
  ↓
合并 JSON
  ↓ (返回主对话)
主对话看到 summary

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.