在 Claude Code / OpenClaw / Cursor 中直接运行 agency-orchestrator YAML 工作流——无需 API key,使用当前会话的 LLM 作为执行引擎。当用户提供 .yaml 工作流文件或要求多角色协作完成任务时触发。
npx skills add https://github.com/jnMetaCode/superpowers-zh --skill workflow-runner
直接在当前会话中执行 agency-orchestrator 的 YAML 工作流,无需配置 API key。当前 LLM 就是执行引擎——依次扮演每个角色完成任务。
.yaml 工作流文件(如 运行 workflows/story-creation.yaml)agency-agents-zh 并希望直接在 AI 工具内编排多角色按以下顺序执行,不要跳步:
用 Read 工具读取用户指定的 YAML 文件,提取以下字段:
name: "工作流名称"
agents_dir: "agency-agents-zh" # 角色定义目录
inputs: # 输入变量
- name: xxx
required: true/false
default: "默认值"
steps: # 执行步骤
- id: step_id
role: "category/agent-name" # 角色路径
task: "任务描述 {{变量}}" # 支持模板变量
output: variable_name # 输出变量名
depends_on: [other_step_id] # 依赖关系
忽略 llm、concurrency、timeout、retry 配置——Skill 模式使用当前会话的 LLM,这些字段仅用于 CLI 模式。
定位角色目录:用 Bash test -d 按以下顺序检查,用第一个存在的:
{agents_dir}/(如 ./agency-agents-zh/)../{agents_dir}/(上级目录){agents_dir}/node_modules/agency-agents-zh/如果全部找不到,停止执行并提示用户:
找不到角色目录。请先安装:
git clone --depth 1 https://github.com/jnMetaCode/agency-agents-zh.git
或:npm install agency-agents-zh
required: true 的输入,检查用户消息中是否已提供值default 的可选输入:使用默认值根据 depends_on 进行拓扑排序,将步骤分成多个层级:
在回复中展示执行计划:
执行计划(共 N 步):
第 1 层: [step_id] — 角色名
第 2 层: [step_a, step_b] — 并行
第 3 层: [step_id] — 角色名
对每一层:
用 Read 工具读取该层所有步骤的角色 .md 文件:{角色目录}/{role}.md
从文件中提取:
name 字段--- 之后的全部 markdown 内容将 task 中的 {{变量名}} 替换为:
单步骤层:直接在主会话中扮演该角色执行。格式:
### Step N/Total: step_id(角色名)
[以该角色身份完成 task,使用角色的专业知识和沟通风格]
多步骤层(并行):使用 Agent 工具为每个步骤启动子代理。每个子代理的 prompt 必须包含:
如果 step 有 output 字段,将该步骤的输出文本存入变量上下文,供后续步骤的 {{变量}} 使用。
用 Write 工具将结果保存到文件:
.ao-output/{工作流名称}-{YYYY-MM-DD}/
├── steps/
│ ├── 1-{step_id}.md # 每步的输出
│ ├── 2-{step_id}.md
│ └── ...
├── summary.md # 最后一步的完整输出(最终成果)
└── metadata.json # 基本元数据
metadata.json 格式:
{
"name": "工作流名称",
"date": "2026-03-22",
"success": true,
"steps": [
{"id": "step_id", "role": "category/agent", "status": "completed"},
...
]
}
执行完毕后,向用户展示:
<HARD-GATE>
</HARD-GATE>
如果用户没有指定 YAML 文件,但描述了需要多角色协作的任务:
示例:
ao init 或 npm install agency-agents-zhComprehensive GitHub project management with swarm-coordinated issue tracking, project board automation, and sprint planning
Comprehensive technology-agnostic prompt for analyzing and documenting project folder structures. Auto-detects project types (.NET, Java, React, Angular, Python, Node.js, Flutter), generates detailed blueprints with visualization options, naming conventions, file placement patterns, and extension templates for maintaining consistent code organization across diverse technology stacks.
Use when complex problems require systematic step-by-step reasoning with ability to revise thoughts, branch into alternative approaches, or dynamically adjust scope. Ideal for multi-stage analysis, design planning, problem decomposition, or tasks with initially unclear scope.
Multi-agent workflow examples to work together on the OpenServ Platform. Covers agent discovery, multi-agent workspaces, task dependencies, and workflow orchestration using the Platform Client. Read reference.md for the full API reference. Read openserv-agent-sdk and openserv-client for building and running agents.
> Compress natural language memory files (CLAUDE.md, todos, preferences) into caveman format to save input tokens. Preserves all technical substance, code, URLs, and structure. Compressed version overwrites the original file. Human-readable backup saved as FILE.original.md.
API design principles and decision-making. REST vs GraphQL vs tRPC selection, response formats, versioning, pagination.
Patterns for automating GitHub workflows with AI assistance, inspired by [Gemini CLI](https://github.com/google-gemini/gemini-cli) and modern DevOps practices.
Groups existing components into logical business domains to plan service-based architecture. Use when asking "which components belong together?", "group these into services", "organize by domain", "component-to-domain mapping", or planning service extraction from an existing codebase. Do NOT use for identifying new domains from scratch (use domain-analysis) or analyzing coupling (use coupling-analysis).
Take jnmetacode/workflow-runner from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.
The instructions reference npm.
Without those the skill loads but fails at the first command.