A runtime-agnostic orchestration skill that routes complex work to parallel agents, sequential workflows, or a hybrid team—and verifies the combined result.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install mu-multi-agent
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install mu-multi-agent using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Multi-Agent Collaboration Commander?
mu-multi-agent is a practical coordination playbook for turning work with independent subtasks into a structured agent team. It helps the main agent decide when delegation is worthwhile, choose between commander, pipeline, and hybrid modes, provide subagents with the context they need, and combine their findings into a coherent deliverable. This Openclaw Skills entry is a methodology skill, not an agent runtime or a tool-specific integration.
Its central safeguards keep orchestration useful: do not parallelize dependent steps, keep assignments concise, and require the main agent to validate and synthesize subagent results rather than forwarding them unchanged. For important or high-risk outputs, a separate reviewer can assess the work against a rubric before delivery.
Multi-Agent Collaboration Commander Use Cases
- Parallel research: Investigate multiple topics, sources, or dimensions at the same time, then consolidate evidence and findings.
- Batch processing: Screen multiple résumés or translate, summarize, or convert several documents as independent tasks.
- Multi-version creation: Generate alternatives in different styles or from different angles, then compare and recommend an option.
- Research followed by writing: Gather information in parallel first, then use it to produce multiple drafts or formats.
- Quality-sensitive deliverables: Add an independent reviewer for formal reports, external documents, fact-heavy analysis, or tasks where the user requests a review.
- Code and content pipelines: Organize dependent work as a sequence, such as planning, implementation, and review.
- When not to use it: Skip orchestration for simple requests, tasks a specialized skill can complete on its own, and straightforward sequential work with no parallel benefit.
How Multi-Agent Collaboration Commander Works
- Acknowledge the task: Announce that multi-agent work has started before reading files or dispatching agents, so the user knows the request is being handled.
- Check whether delegation is warranted: Look for at least two independent subtasks, multiple research sources, batch work, research followed by multiple outputs, or a need for independent review. Prefer a single agent or a specialized skill when that is sufficient.
- Select a workflow: Use commander mode for tool-enabled parallel work, pipeline mode for text-focused sequential or multi-perspective work, and hybrid mode when research should precede multiple outputs.
- Plan and dispatch: Show a concise task plan. Give each subagent a clear role and task, with necessary context included in the assignment. Launch independent work together; dispatch dependent work only after its prerequisites finish.
- Monitor execution: Track dispatched agents, intervene or stop work when needed, and use available history to inspect outcomes. If the runtime lacks subagent support, perform the workflow sequentially while preserving review separation where possible.
- Integrate and verify: The main agent checks the returned results, organizes them by topic rather than by agent, and writes the final response or report. Do not pass raw subagent output directly to the user.
- Review and deliver: Run an independent rubric-based review for suitable high-risk outputs, revise and re-review if needed, then provide the deliverable and an execution statistics card.
Multi-Agent Collaboration Commander Setup
This skill does not prescribe a package installer or a specific agent runtime. Add the SKILL.md file to the skill directory recognized by your agent host, then configure or provide the runtime capabilities that correspond to the skill's abstract contracts: spawn, monitor, history, and notify. Actual tool names vary by runtime; map them by capability rather than assuming a particular implementation.
# Example only: replace these placeholders with paths used by your agent host.
mkdir -p "$AGENT_SKILLS_DIR/mu-multi-agent"
cp /path/to/SKILL.md "$AGENT_SKILLS_DIR/mu-multi-agent/SKILL.md"
After installation, confirm the host can discover the skill and that its runtime exposes the required capabilities. If subagent execution is unavailable, use the documented sequential fallback. No additional dependencies, environment variables, or CLI configuration are specified in the source skill.
Multi-Agent Collaboration Commander Data Schema & Taxonomy
The skill defines an orchestration method rather than a required database or machine-readable project schema. Its working information is organized around tasks, agents, workflow phases, context, review results, and final deliverables.
| Artifact or metadata | Organization and purpose |
|---|---|
| Agent assignment | A concise role-plus-task instruction. Include background or prerequisite results when the subagent needs them. |
| Context | Supply relevant background in the task, pass structured results between dependent phases, or point to a context file for complex background. |
| Agent label | A human-readable role label, optionally paired with an emoji, to identify each worker in plans and status summaries. |
| Workflow | One of commander (tool-enabled parallel execution), pipeline (text-focused workflow), or hybrid (research followed by multiple outputs). |
| Review rubric | For independent review, check factual accuracy, structural completeness, logical consistency, evidence, and formatting; record scores and issues. |
| Execution statistics | Report the mode, agent roles, elapsed time, token input when available, status, parallel versus estimated serial time, and review outcome. |
| Long-form output | For content over 500 characters, save a Markdown report such as ./workspace/projects/{topic}/report.md and send the file separately from its text summary. |
The main agent owns the final synthesis. The skill does not require a fixed directory structure beyond its example report and context-file paths, nor does it define a persistent metadata format.
Multi-Agent Collaboration Commander Advanced Features
- Automatic mode routing: Selects commander, pipeline, or hybrid orchestration based on tool needs, dependencies, and whether multiple outputs are required.
- Runtime abstraction: Uses capability contracts for spawning, monitoring, reviewing history, and notifying users instead of binding the method to one agent platform.
- Parallel and dependency-aware execution: Runs independent tasks together while reserving sequential dispatch for tasks that depend on earlier results.
- Independent quality review: Adds a separate reviewer for appropriate high-risk deliverables, with explicit scoring criteria and a revise-and-review loop.
- Sequential fallback: Preserves the workflow, review separation, and execution statistics when the host cannot run subagents.
- Context handoff patterns: Supports background injection, prerequisite-result handoff, or reading a context file for complex assignments.
- Cost-aware delegation: Recommends lightweight models for simpler research or review roles and allows model upgrades when quality requirements justify them.
- Execution transparency: Requires a task plan before dispatch and a statistics card after multi-agent work.
- Delivery safeguards: Requires main-agent synthesis, completion checks, and separate file and text notifications when sending a long report.
SKILL.md
Loading
Related Openclaw Skills
PhotoFresco
PhotoFresco lets an AI agent edit images in a user-visible, permission-controlled editor and export the results as PNG, PSD, or native PFD files.
Long-Task Dependency and Acceptance Planner
Turn multi-stage work into a budget-aware dependency graph with verifiable results, safe recovery checkpoints, and clear next steps.
Dining Demand Research
Research Resy and OpenTable listings to create a dated restaurant shortlist and compare time-specific availability without booking.
Mundane
Mundane connects AI agents with ID-verified local workers for real-world tasks, holding payment in escrow until the user reviews the proof.
EIN.LLC Activation Continuation
A Node.js CLI skill that acquires a pinned EIN.LLC continuation archive and safely resumes caller-owned business activation tasks without claiming, paying, or filing.
Counterparty Debt Quick Check
Turn a company name into a concise, source-linked snapshot of publicly disclosed debt clues for pre-deal research.
Featured*
