An advanced multi-agent collaboration framework that structures complex task planning, execution, peer reviews, and self-evolution.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install collaborative-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 collaborative-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 Collaborative Agent?
The Collaborative Agent skill implements the AutoResearchClaw (ARC) 2026 framework to enable powerful multi-agent peer reviews and task execution workflows. Governed under the core think-expand cognitive layer, this framework decomposes complex tasks into specialized sub-tasks and delegates them to dedicated agent roles. By utilizing cooperative planning, executing, reviewing, summarizing, researching, and criticizing agents, the system establishes an iterative feedback loop that continuously refines output quality.
Designed to seamlessly fit into the ecosystem of Openclaw Skills, this tool integrates with powerful core engines such as ScriptDB, VSAEngine, SkillEvolver, and CognitiveReasoner. This ensures that complicated coding, research, and technical writing projects benefit from self-reinforcing loops and automated quality assurance.
Collaborative Agent Use Cases
- Executing complex, multi-step engineering tasks requiring sequential planning, development, and stringent review cycles.
- Managing academic and technical research projects that need multi-perspective analysis and multi-source synthesis.
- Generating high-quality code and content where peer reviews and iterative feedback-driven optimizations are vital.
- Aggregating information from diverse channels into structured, comprehensive, and verified reports.
How Collaborative Agent Works
- The think-expand core system receives a complex, high-level goal and delegates the decomposed requirements to the collaborative agent workflow.
- Specialized roles (such as researcher, planner, executor, reviewer, and summarizer) are dynamically assigned to individual sub-tasks, mapping out explicit dependency trees.
- The workflow executes sub-tasks sequentially based on dependencies, leveraging capability_executor.detect_and_execute() to interact with system capabilities.
- Output deliverables go through an iterative peer-review phase where reviewer and critic agents provide linguistic and numeric feedback via the CognitiveReasoner.
- The executor agent refines the output based on the evaluations, while the SkillEvolver logs failures to enable continuous agent self-evolution.
- The summarizer compiles all verified execution details and feedback logs into a highly structured final report.
Collaborative Agent Setup
To get started with the Collaborative Agent skill, ensure your workspace is configured to point to your Openclaw Skills directory.
Configure your python environment to access the collaborative agent modules:
# Verify the installation path
cd ~/.config/opencode/skills/collaborative-agent
Integrate the collaborative agent into your Python automation workflows using the example below:
import sys
sys.path.insert(0, '/Users/pc/.config/opencode/skills/collaborative-agent')
from collaborative_agent import ARCWorkflow, AgentRole
# Instantiate the collaborative workspace
wf = ARCWorkflow()
# Orchestrate your task dependencies and assign roles
wf.add_task("Research advanced optimization routines", role="researcher")
wf.add_task("Formulate implementation blueprint", role="planner", depends_on=[0])
wf.add_task("Write and debug implementation", role="executor", depends_on=[1])
wf.add_task("Review code standards and coverage", role="reviewer", depends_on=[2])
wf.add_task("Compile structured executive summary", role="summarizer", depends_on=[3])
# Run all cooperative tasks in order
results = wf.run_all()
# Retrieve the compiled report
report = wf.get_report()
print(report)
Collaborative Agent Data Schema & Taxonomy
The Collaborative Agent structures its internal configurations and telemetry data to allow tracing, evaluation, and evolution across your Openclaw Skills workflow environment.
| Data Component | Format / Storage | Description |
|---|---|---|
| Base Directory | Local Path (~/.config/opencode/skills/collaborative-agent) |
Contains core modules, including collaborative_agent.py and external tool connectors. |
| ARCWorkflow Graph | In-memory Dependency Tree | Maintains a list of active tasks, task states (pending, running, complete), dependency references, and role mappings. |
| Cognitive Feedback | CognitiveReasoner Schema | Structured linguistic evaluations and numeric grading used during the iterative peer-review loop. |
| Failure History | SkillEvolver Database | Persistent record of execution bugs, failed reasoning paths, and corrected strategies for skill evolution. |
Collaborative Agent Advanced Features
- Iterative Peer Review Cycle: Establishes an automatic 'Produce -> Review -> Feedback -> Optimize' loop, preventing unverified outputs from reaching the end-user.
- Failure Logging & Self-Evolution: Natively calls SkillEvolver to record logical or mechanical execution failures, facilitating continuous self-evolution across your Openclaw Skills deployments.
- Cognitive Evaluation Logic: Employs CognitiveReasoner to generate deep verbal and numerical feedback on code and document outputs, driving context-aware refinements.
- Extensible Tool Execution: Integrates directly with capability_executor to dynamically execute pre-registered system tools and system capabilities.
- Tracing and Observability: Designed with tracing integration, allowing developers to observe, debug, and monitor multi-agent actions and token-usage efficiency.
SKILL.md
Loading
Related Openclaw Skills
ComfyUI Multi-Platform Master Guide
An all-in-one AI coding agent skill compiling top-tier tutorials for multi-platform ComfyUI deployment and AI image generation mastery.
Efficiency Gold Miner & SOP Alchemy Universal
An intelligent work trace analyzer that scans daily calendars, tasks, approvals, and messages to discover hidden operational frictions and distill them into reusable automation SOPs and skill specifications.
AI Chat Enhancer
An advanced LLM interaction toolkit designed to optimize prompt templating, response caching, and context token tracking.
AI Painting and Image Generation Suite (ai绘画-ddcd9b)
A comprehensive AI drawing and generation skill aggregating top Bilibili video-learned workflows for Stable Diffusion, ComfyUI, Midjourney, and Kling AI.
CrewAI Multi-Agent Orchestration SOP
A production-grade standard operating procedure for building role-based multi-agent systems and event-driven workflows using the CrewAI framework.
Collaborative Agent Self-Evolution (MUSE-Autoskill)
A self-evolving agent framework inspired by ByteDance's MUSE-Autoskill that enables AI agents to dynamically generate and reuse their own tools.
Featured*
