← Back to Skills Marketplace
137
Downloads
0
Stars
0
Active Installs
3
Versions
Install in OpenClaw
/install hermes-coordinator
Description
Transforms the agent into a coordinator that breaks down goals, dispatches tasks to workers, and synthesizes results without direct execution.
README (SKILL.md)
Coordinator Mode Skill
🛡️ OpenClaw 混合进化方案 — 将 Hermes-agent（100K ⭐）+ Claude Code 核心能力移植到 OpenClaw
Skill Name: coordinator Version: 1.1.0 Trigger: Manual activation
🚀 一键安装
mkdir -p ~/.openclaw/skills && cd ~/.openclaw/skills && curl -fsSL https://github.com/olveww-dot/openclaw-hermes-claude/archive/main.tar.gz | tar xz && cp -r openclaw-hermes-claude-main/skills/hermes-coordinator . && rm -rf openclaw-hermes-claude-main && echo "✅ coordinator 安装成功"
Overview
Turns the main agent into a Coordinator — a commander that only dispatches tasks, never executes them directly. All execution is delegated to Worker subagents. Results flow back via \x3Ctask-notification> messages.
一键安装
cd ~/.openclaw/workspace/skills/coordinator
bash install.sh
安装后激活 Coordinator 模式：
# 方式1: 运行激活脚本（查看并复制 prompt 到 session）
bash ~/.openclaw/workspace/skills/coordinator/scripts/activate-coordinator.sh
# 方式2: 直接对 EC 说
进入协调模式
How It Works
When activated, the agent's system prompt is replaced with the Coordinator prompt. The agent gains a clean role:
- Analyze the user's goal
- Break down into independent tasks
- Fan out — spawn Workers in parallel
- Wait for
\x3Ctask-notification>results - Synthesize and report to user
Tool Set (Coordinator)
Only these tools are available in Coordinator mode:
| Tool | Purpose |
|---|---|
spawn / sessions_spawn |
Launch a Worker subagent |
message (send) |
Continue an existing Worker |
sessions_yield |
End turn and wait for results |
Coordinator never calls tools directly to do work. Only to dispatch.
Key Principles
- Never thank workers in results
- Never predict worker results — wait for actual
\x3Ctask-notification> - Be verbose in prompts — workers can't see the coordinator's conversation
- Fan out aggressively — parallel workers are free
- Synthesize before delegating — understand results before sending follow-ups
Files
coordinator/
├── SKILL.md ← This file
├── README.md ← User-facing guide
├── install.sh ← 一键安装脚本
├── scripts/
│ └── activate-coordinator.sh ← 激活脚本
└── src/
├── coordinator-prompt.ts ← Coordinator system prompt
└── worker-prompt.ts ← Worker agent prompt template
🧩 配套技能
本 skill 是 OpenClaw 混合进化方案 的一部分：
将 Hermesagent（100K ⭐）+ Claude Code 核心能力移植到 OpenClaw
将 Hermes-agent（100K ⭐）+ Claude Code 核心能力移植到 OpenClaw
🔗 GitHub 项目：olveww-dot/openclaw-hermes-claude
完整技能套件（6个）：
- 🛡️ crash-snapshots — 崩溃防护
- 🧠 auto-distill — T1 自动记忆蒸馏
- 🎯 coordinator — 指挥官模式（本文）
- 💡 context-compress — 思维链连续性
- 🔍 lsp-client — LSP 代码智能
- 🔄 auto-reflection — 自动反思
重新安装 / 更新
bash ~/.openclaw/workspace/skills/coordinator/install.sh
Usage Guidance
This skill appears coherent with its stated purpose, but review a few things before installing: 1) Inspect the bundled files (install.sh, scripts/activate-coordinator.sh, and the two prompt files) yourself — the curl command in SKILL.md downloads a GitHub archive, which is common but you should verify the repo. 2) Ensure you have Node available if you plan to use the activation script (it runs node -e). 3) Be aware spawned Workers are granted exec/read/write/web_fetch/browser in the Worker prompt — they can access workspace files and run shell commands; if your workspace contains secrets, consider running in a restricted/sandboxed environment or adjusting worker permissions. 4) Note the install script attempts to rsync to /Users/ec/research if present (odd hardcoded path); if that is a concern, run the install steps manually or edit the script. 5) If you need higher assurance, run the install in an isolated environment and verify network artifacts and file writes.
Capability Assessment
Purpose & Capability
The name/description (Coordinator that breaks goals into tasks and dispatches Workers) matches the included prompts and scripts: coordinator-prompt instructs dispatch-only behavior and worker-prompt grants work capabilities. The skill does not request unrelated credentials or config paths. Minor note: the activation script uses node to extract the prompt but the registry metadata did not declare 'node' as a required binary.
Instruction Scope
SKILL.md and README clearly describe only spawning workers and synthesizing <task-notification> results. Worker prompts explicitly allow exec/read/write/web_fetch/browser, which is expected for Workers but means spawned workers can access files, run shell commands, and fetch web resources — powerful operations that are appropriate for a Worker but worth awareness. SKILL.md also includes an install curl | tar command that downloads a GitHub archive (a reasonable source), and the activation script prints the system prompt for manual paste.
Install Mechanism
There is no registry-level install spec; installation is via included install.sh or the SKILL.md curl|tar command. The curl target is a GitHub archive (trusted release host). The bundled install.sh copies files into the OpenClaw workspace and optionally rsyncs to a hardcoded /Users/ec/research path if present — odd but not obviously malicious. Activation script invokes node -e to read a .ts file, so Node must be present though not declared.
Credentials
The skill declares no required environment variables or credentials, which matches the bundle. However, the Worker template explicitly authorizes file I/O, shell exec, and web/browser access. Those permissions are coherent with implementing worker tasks but give workers broad access to workspace files and networked endpoints; that power is expected for the pattern but is sensitive in practice.
Persistence & Privilege
always:false and default autonomous invocation are preserved. The install.sh writes files into the user's OpenClaw workspace (normal for a skill) and sets execute bits for its scripts; it does not modify other skills or system-wide configuration. There is an optional rsync to a research directory if present (non-standard path), but no evidence the skill auto-enables itself or persistently alters unrelated settings.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install hermes-coordinator - After installation, invoke the skill by name or use
/hermes-coordinator - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.2
v1.0.2: 添加一键安装命令，ClaWHub安装后可直接使用
v1.0.1
v1.0.1: 详情页加入Hermes-agent+Claude Code融合卖点
v1.0.0
Initial release: 指挥官模式，只派发任务不执行，交给Worker子代理完成
Metadata
Frequently Asked Questions
What is Coordinator?
Transforms the agent into a coordinator that breaks down goals, dispatches tasks to workers, and synthesizes results without direct execution. It is an AI Agent Skill for Claude Code / OpenClaw, with 137 downloads so far.
How do I install Coordinator?
Run "/install hermes-coordinator" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Coordinator free?
Yes, Coordinator is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Coordinator support?
Coordinator is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Coordinator?
It is built and maintained by ECsss (@olveww-dot); the current version is v1.0.2.
More Skills