← Back to Skills Marketplace
219
Downloads
1
Stars
0
Active Installs
10
Versions
Install in OpenClaw
/install openclaw-multi-agents
Description
别再用一个 AI 单打独斗。这个 Skill 帮你在 OpenClaw 里组建专属 AI 团队：Manager 规划派活，Worker 各自执行，每个交付物强制过质量门。61 个历史人物人格可选，你的团队做完任务不会直接交卷。 Stop relying on a single AI. This skill hel...
Usage Guidance
This skill appears coherent for building persistent OpenClaw sub‑agent teams, but take these precautions before enabling or running scripts: 1) Review the bundled scripts (scripts/*.sh) yourself to confirm they perform only local file ops and no unexpected network calls; 2) Back up ~/.openclaw/openclaw.json before allowing any automated writes; 3) Be aware the skill will read your USER.md, workspace memory (wisdom) and session history — remove or redact very sensitive content first if you prefer; 4) The skill states it will update agents.list and agentToAgent.allow in openclaw.json — confirm each change before approving; 5) Run openclaw gateway restart manually in your terminal when instructed (do not let an automated agent perform the restart). If you want extra assurance, test in a disposable OpenClaw profile or VM first.
Capability Analysis
Type: OpenClaw Skill
Name: openclaw-multi-agents
Version: 1.1.0
The bundle is a comprehensive multi-agent orchestration framework for OpenClaw, implementing a three-tier hierarchy (Main, Manager, and Workers). It includes well-documented bash scripts (e.g., setup_team.sh, setup_agent.sh, and run_planning_interview.sh) for managing agent workspaces and configurations. The SKILL.md file provides detailed instructions for agent behavior, including 'Iron Rules' for quality gates and reporting, which act as reliability and safety measures. While the skill requires broad permissions to read and write to ~/.openclaw/openclaw.json and various workspace directories, these actions are strictly aligned with its stated purpose of managing agent registries and personas. No evidence of data exfiltration, malicious persistence, or obfuscation was found; an external network call to prompts.chat is used legitimately to fetch community-sourced persona templates.
Capability Assessment
Purpose & Capability
Name/description claim to build/manage sub‑agents; required binaries (bash, openclaw), required config paths (~/.openclaw/openclaw.json and workspace*/ SOUL.md/AGENTS.md, memory/) and the included scripts/templates are all appropriate and expected for creating and managing persistent sub‑agent workspaces and updating OpenClaw config.
Instruction Scope
SKILL.md explicitly instructs the agent to read USER.md, workspace memory (wisdom, logs) and sessions history to run the planning interview and design teams — that is reasonable for a planning/inspection step, but it does mean the skill will read potentially sensitive conversation history and user files. The skill documents that it will write SOUL.md/AGENTS.md and update openclaw.json (agents.list and agentToAgent.allow) and requests explicit user approval before making changes. The README/INSTALL also warns the AI should not restart the gateway itself and instructs the user to run restart commands locally.
Install Mechanism
No install spec (instruction-only skill) and scripts are bundled in-repo. No remote download/extraction directives in the registry metadata. That minimizes supply‑chain risk; user-run installation (clawhub or GitHub copy) is normal for this type of skill.
Credentials
No environment variables or external credentials are required. The skill asks for read/write access to OpenClaw config and workspace paths — these are proportionate to its function of creating and registering sub‑agents. It also declares reading session history (documented in metadata), which is relevant but sensitive; user consent/awareness is important.
Persistence & Privilege
always is false (not force‑included), and the skill modifies only OpenClaw configuration and workspace files related to sub‑agents (SOUL.md, AGENTS.md, openclaw.json) which is expected behavior. The skill does not request system‑wide persistent privileges beyond its own workspace/config updates. It documents that updates happen with user approval and warns the agent should not execute gateway restart itself.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install openclaw-multi-agents - After installation, invoke the skill by name or use
/openclaw-multi-agents - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.1.0
openclaw-multi-agents v1.1.0
- Documentation (SKILL.md) expanded and reorganized for clarity.
- Added detailed process breakdowns for team creation, incremental team updates, and health checks.
- Included step-by-step guidance for the planning interview, emphasizing minimal user friction.
- Clarified configuration file roles, team existence detection, and communication protocol.
- No implementation or code changes—documentation only.
v1.0.9
fix: correct Worker→Manager session key (agent:manager:main); Pure Relay Iron Rule must notify user via message tool; Quality Gate requires sessions_send verification and detailed report format; Wisdom uses absolute path for team sharing; Worker/Manager templates translated to English
v1.0.8
fix: correct Worker→Manager session keys in scripts (agent:manager:main); prevent AI exec of gateway restart in run_planning_interview.sh
v1.0.7
fix: expand metadata permissions declarations to reduce Suspicious flag; all file access scopes now explicitly documented
v1.0.6
fix: Quality Gate now verifies Worker sent sessions_send before checking content; upgrade report format with verification detail table and worker status table
v1.0.5
feat: sub-agents read/write team Wisdom; fix absolute path for shared Wisdom directory (~/.openclaw/workspace/memory/wisdom/); add Wisdom recording standard to Worker/Manager templates
v1.0.4
fix: Main Agent Pure Relay Iron Rule - must notify user via message tool immediately after receiving Manager report; not sessions_send, not waiting for user to ask
v1.0.3
fix: correct Worker→Manager session key to agent:manager:main; add Iron Rule for Worker/Manager reporting; add Worker SOUL/AGENTS templates; rewrite examples; translate templates to English
v1.0.1
fix: correct package structure (remove nested skills/ directory in bundle)
v1.0.0
Initial release. Three-tier architecture (Main → Manager → Workers), built-in QA gate, 61+ persona library, planning interview. Fix: agentToAgent.allow must include "main".
Metadata
Frequently Asked Questions
What is Openclaw Multi Agents?
别再用一个 AI 单打独斗。这个 Skill 帮你在 OpenClaw 里组建专属 AI 团队：Manager 规划派活，Worker 各自执行，每个交付物强制过质量门。61 个历史人物人格可选，你的团队做完任务不会直接交卷。 Stop relying on a single AI. This skill hel... It is an AI Agent Skill for Claude Code / OpenClaw, with 219 downloads so far.
How do I install Openclaw Multi Agents?
Run "/install openclaw-multi-agents" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Openclaw Multi Agents free?
Yes, Openclaw Multi Agents is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Openclaw Multi Agents support?
Openclaw Multi Agents is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Openclaw Multi Agents?
It is built and maintained by Ryan (@porkapple); the current version is v1.1.0.
More Skills