Awesome-OpenClaw-Research

🦞 OpenClaw launched Nov 2025, hit 200k GitHub stars in 84 days, and surpassed 330k stars by March 2026. This repo collects papers studying or built upon the OpenClaw ecosystem — covering agent infrastructure, learning, safety, embodiment, social dynamics & domain applications. The questions are universal; OpenClaw is the lens.

---

Table of Contents

• Papers — Core of this repo
  • Infrastructure & Systems
  • Learning & Evolution
  • Safety & Security
  • Agent Society
  • application
• Architecture
• Ecosystem Timeline
• Other Resources — SDKs, tools, community, related repos
• Contributing

---

📄 Papers

25+ papers published in Feb–Mar 2026 alone. Each entry includes paper link, code (if available), and key highlights.

Infrastructure & Systems

Frameworks, OS paradigms, benchmarks, and protocol evaluation.

Title	Venue	Date	Paper	Code	Highlights
OpenClaw as Language Infrastructure: A Case-Centered Survey of a Public Agent Ecosystem in the Wild (Survey)	Preprints.org	2026.03			GATE / AERO frameworks; analyzes 38 ecosystem papers
AgentOS: From Application Silos to a Natural Language-Driven Data Ecosystem	arXiv	2026.03			Agent-centric OS paradigm; Skills-as-Modules; KDD framing
EvoClaw: Evaluating AI Agents on Continuous Software Evolution	arXiv	2026.03			DeepCommit Milestone DAGs; 7 repos / 5 languages; best agent only 38% in continuous settings
ClawGUI: A Unified Framework for Training, Evaluating, and Deploying GUI Agents	arXiv	2026.04			ClawGUI is an open-source framework for training, evaluating, and deploying GUI-based AI agents across platforms
MCP-Atlas: A Large-Scale Benchmark for Tool-Use Competency with Real MCP Servers	arXiv	2026.01			36 MCP servers; 220 tools; 1,000 tasks; multi-step workflow eval
NanoClaw	GitHub	2026.01			NanoClaw is an open-source, lightweight, and containerized personal AI assistant framework
nanobot	GitHub	2026.02			Nanobot is a lightweight system framework for personal AI agents, providing agent execution architecture, cross-platform message integration, and tool invocation mechanisms
PicoClaw: Ultra-Efficient AI Assistant in Go	GitHub	2026.02			Supports operation on extremely resource-constrained hardware (e.g., $10 boards, <10 MB RAM)
AutoResearchClaw:Chat an Idea. Get a Paper. Autonomous, Collaborative & Self-Evolving.	GitHub	2026.03			AutoResearchClaw is an open‑source infrastructure for fully automated and co‑pilot‑assisted scientific research
Three Provinces and Six Ministries · Edict	GitHub	2026.02			Edict is an OpenClaw‑based multi‑agent coordination and governance infrastructure that defines structured roles, task delegation, and audit mechanisms for orchestrating agent workflows
ClawTeam: Agent Swarm Intelligence	GitHub	2026.03			ClawTeam is a foundational framework for AI agent group collaboration and automated orchestration, providing system-level capabilities such as multi-agent task allocation, team coordination, and real-time monitoring
CyberClaw: Next-Gen Transparent Agent Architecture	GitHub	2026.03			CyberClaw is a transparent, controllable agent architecture for multi-domain AI systems, enabling decision auditing, safe execution, persistent memory, and monitoring, compatible with OpenClaw/Claude Code

Learning & Evolution

Reinforcement learning, meta-learning, and self-improvement of agents.

Title	Venue	Date	Paper	Code	Highlights
OpenClaw-RL: Train Any Agent Simply by Talking	arXiv	2026.03			Evaluative + Directive signals; async 4-component RL architecture; personalization 0.17→0.81
MetaClaw: Just Talk -- An Agent That Meta-Learns and Evolves in the Wild	arXiv	2026.03			Continuous meta-learning; skill-driven adaptation; accuracy 21.4% → 40.6%
Hermes Agent	GitHub	2026.03			Hermes Agent is an autonomous AI agent with a built‑in self‑improving learning loop, capable of persistent memory, autonomous skill creation and refinement, and adaptive personalized interactions across sessions
SkillClaw: Let Skills Evolve Collectively with Agentic Evolver	arXiv	2026.04			SkillClaw is a framework for evolving and sharing skills across multi-user autonomous agents, enabling continuous capability improvement and knowledge transfer

Safety & Security

Attack benchmarks, defense frameworks, supply-chain security, and runtime protection.

Title	Venue	Date	Paper	Code	Highlights
From Assistant to Double Agent: Formalizing and Benchmarking Attacks on OpenClaw for Personalized Local AI Agent	arXiv	2026.02			End-to-end security eval; only 17% native defense rate
A Trajectory-Based Safety Audit of Clawdbot (OpenClaw)	arXiv	2026.02			34 test cases / 6 risk dimensions; overall pass rate 58.9%; Intent Misunderstanding 0%
Formal Analysis and Supply Chain Security for Agentic AI Skills	arXiv	2026.02			Dolev-Yao modeling; 96.95% F1; 0% false positives; 22 agent frameworks
Don't Let the Claw Grip Your Hand: A Security Analysis and Defense Framework for OpenClaw	arXiv	2026.03			47 adversarial scenarios; MITRE ATLAS/ATT&CK; HITL defense: 17% → 19–92%
OpenClaw PRISM: A Zero-Fork, Defense-in-Depth Runtime Security Layer for Tool-Augmented LLM Agents	arXiv	2026.03			Defense-in-depth; zero-fork; anti prompt injection
Uncovering Security Threats and Architecting Defenses in Autonomous Agents: A Case Study of OpenClaw	arXiv	2026.03			Tsinghua & Ant Group; tri-layered risk taxonomy; FASA + ClawGuard; 26% community tools have vulns
Defensible Design for OpenClaw: Securing Autonomous Tool-Invoking Agents	arXiv	2026.03			Security-as-engineering blueprint; risk taxonomy; practical research agenda
Claw-Eval: Toward Trustworthy Evaluation of Autonomous Agents	arXiv	2026.04			Claw‑Eval is a benchmark suite for autonomous agents with trajectory-aware grading and safety metrics, addressing gaps in existing evaluations for deployable agents
ClawArena: Benchmarking AI Agents in Evolving Information Environments	arXiv	2026.04			ClawArena benchmarks autonomous agents on belief accuracy in dynamic, multi-source environments, highlighting reasoning, belief revision, and personalization challenges
Taming OpenClaw: Security Analysis and Mitigation of Autonomous LLM Agent Threats	arXiv	2026.03			Taming OpenClaw is a security framework analyzing LLM agent vulnerabilities and proposing lifecycle-aware defenses to reduce systemic risks
ClawBench: Can AI Agents Complete Everyday Online Tasks?	arXiv	2026.04			ClawBench benchmarks autonomous agents on completing real-world online tasks, tracking behavior and revealing reliability and safety challenges
Your Agent, Their Asset: A Real-World Safety Analysis of OpenClaw	arXiv	2026.04			Your Agent, Their Asset evaluates OpenClaw using the CIK taxonomy, exposing high vulnerabilities and limited defenses across multiple models

Agent Society

Social behaviors, emergent norms, peer learning, and collective dynamics in agent populations.

Title	Venue	Date	Paper	Code	Highlights
OpenClaw Agents on Moltbook: Risky Instruction Sharing and Norm Enforcement in an Agent-Only Social Network	arXiv	2026.02			39k posts / 14.5k agents; 18.4% action-inducing; emergent norm enforcement
When AI Agents Teach Each Other: Discourse Patterns Resembling Peer Learning in the Moltbook Community	arXiv	2026.02			2.4M agents' peer learning patterns
OpenClaw AI Agents as Informal Learners at Moltbook: Characterizing an Emergent Learning Community at Scale	arXiv	2026.02			2.8M agents' informal learning behavior
When Openclaw Agents Learn from Each Other: Insights from Emergent AI Agent Communities for Human-AI Partnership in Education	arXiv	2026.03			167k agents; bidirectional scaffolding; emergent peer learning; implications for AIED
MoChat: Reconnecting the World through AI Agents	GitHub	2026.02			MoChat is a native AI agent social platform that builds an interactive ecosystem and collaborative space for AI agents
AgentPanel: The world’s first research-focused human-AI Agent collaborative discussion community	GitHub	2026.03			AgentPanel is a collaborative discussion platform where humans and AI agents (e.g., OpenClaw) jointly engage in research‑oriented social interactions, enabling multi‑agent Q&A, debate, and knowledge sharing

application

Vertical applications, organized into embodied, scientific discovery, medical, and other directions.

Embodied Agents

Robotics, physical embodiment, and ROS integration.

Title	Venue	Date	Paper	Code	Highlights
RoboClaw: An Agentic Framework for Scalable Long-Horizon Robotic Tasks	arXiv	2026.03			VLA model; Entangled Action Pairs; +25% success rate; −53.7% human effort
RoboClaw (MINT): Open-Source Embodied Intelligence Assistant	GitHub	2026.03			Conversational arm setup; calibration & teleoperation; built on OpenClaw ecosystem
ROSClaw: An OpenClaw ROS 2 Framework for Agentic Robot Control and Interaction	arXiv	2026.03			SF Hackathon champion; Unitree G1/H1, DJI; runs on RPi4
RoClaw: The Cerebellum — physical embodiment for AI agents	GitHub	2026.03			Dual-Brain bytecode architecture; somatic firmware; open-source hardware CAD & simulation

Scientific Discovery

Automated research, distributed discovery, and life-science research platforms.

Title	Venue	Date	Paper	Code	Highlights
Autonomous Agents Coordinating Distributed Discovery Through Emergent Artifact Exchange	arXiv	2026.03			Automated research collaboration; shared artifacts; peptide design / ceramic screening / cross-domain discovery
From Agent-Only Social Networks to Autonomous Scientific Research: Lessons from OpenClaw and Moltbook, and the Architecture of ClawdLab and Beach.Science	arXiv	2026.02			Agent social networks to research platforms; life-science collaboration; workflow orchestration and platform architecture
ScienceClaw	GitHub / Website	2026.03			Self-evolving research colleague for scientists; 285 skills; persistent research memory; life-science automation platform

Medical

Clinical workflows, medical imaging, digital twins, and medical skill ecosystems.

Title	Venue	Date	Paper	Code	Highlights
Autonomous Agent-Orchestrated Digital Twins (AADT): Leveraging the OpenClaw Framework for State Synchronization in Rare Genetic Disorders	arXiv	2026.03			Medical digital twins for rare genetic disorders; heartbeat-based synchronization; longitudinal phenotype tracking
MedOpenClaw: Auditable Medical Imaging Agents Reasoning over Uncurated Full Studies	arXiv	2026.03			Auditable agents for full 3D medical imaging studies; MedFlowBench; spatial grounding and tool use
OpenClaw-Medical-Skills	GitHub	2026.03			Open-source medical skill library; 869 curated skills; clinical, genomics, drug discovery, and bioinformatics
When OpenClaw Meets Hospital: Toward an Agentic Operating System for Dynamic Clinical Workflows	arXiv	2026.03			Restricted execution; document-centric interaction; page-indexed memory; medical skill library

Other

Other application directions such as education, knowledge work, and math learning.

Title	Venue	Date	Paper	Code	Highlights
Scaling Laws for Educational AI Agents	arXiv	2026.03			Agent scaling laws; AgentProfile framework; 330+ profiles and 1,100+ skill modules across K-12
DenchClaw	GitHub	2026.02			OpenClaw framework for business operations and sales outreach; CRM automation, calling, and local workflows
MathClaw	GitHub	2026.03			Multimodal learning assistant for secondary-school mathematics; math education and interactive learning support

---

🏗 Architecture

flowchart LR
    subgraph Channels["📱 Channels"]
        direction TB
        C1[WhatsApp] ~~~ C2[Telegram] ~~~ C3[Discord] ~~~ C4[Slack] ~~~ C5["50+"]
    end

    subgraph Core["⚙️ Core"]
        direction TB
        Router[Message Router] --> Kernel[Agent Kernel]
        Kernel <--> LLM["LLM Backend\nClaude / GPT / Ollama"]
    end

    subgraph Memory["🧠 Memory"]
        direction TB
        M1[Session Context] ~~~ M2[Daily Notes] ~~~ M3[Long-term Memory] ~~~ M4[Semantic Search]
    end

    subgraph Extensions["🔌 Extensions"]
        direction TB
        E1["Skills (5,700+)"] ~~~ E2["MCP (3,200+)"] ~~~ E3[Plugins]
    end

    subgraph External["🌐 External"]
        direction TB
        X1[Moltbook] ~~~ X2[ROSClaw] ~~~ X3[RoboClaw]
    end

    Channels --> Core
    Core <--> Memory
    Core <--> Extensions
    Extensions --> External

Loading

---

📅 Ecosystem Timeline

2025.11 ─── Launch (ClawdBot / Moltbot → OpenClaw)
    │
2025.12 ─── ClawHub skill marketplace
    │
2026.01 ─── Moltbook (1.5M agents / 72h) ─── Academic paper boom (6 papers / 2 weeks)
    │
2026.02 ─── ClawHavoc attack ─── CVE-2026-25253 ─── 200k Stars (84 days)
    │         │
    │         └── Response: VirusTotal + audit mechanisms
    │
2026.03 ─── RL / Meta-Learning / Robotics papers ─── 330k Stars
    │
    └── ROSClaw Hackathon champion ─── v2026.3.13-1 (68th release)
              │
              └── WeChat official ClawBot plugin (03.22)
                    │
                    └── National Cybersecurity Advisory (China, 03.13)

Full timeline

Date	Event
2025.11.24	OpenClaw (formerly ClawdBot / Moltbot) launched
2025.12	ClawHub skill marketplace released
2026.01	Moltbook launched — 1.5M agents registered in 72h
2026.01	6 academic papers produced in 2 weeks
2026.02.02	Risky Sharing & Norm Enforcement paper
2026.02	PASB security evaluation paper
2026.02	ClawHavoc supply chain attack — 1,184 malicious skills
2026.02	CVE-2026-25253 disclosed (RCE, CVSS 8.8)
2026.02.16	GitHub Stars crossed 200k (84 days)
2026.02	OpenClaw + VirusTotal security partnership
2026.03	OpenClaw-RL / MetaClaw / AgentOS / RoboClaw papers
2026.03	ROSClaw won SF OpenClaw Hackathon
2026.03.13	v2026.3.13-1 released (68th release)
2026.03.13	China National Cybersecurity Advisory
2026.03.22	WeChat official ClawBot plugin released
2026.03	GitHub Stars crossed 330k

---

📦 Other Resources

Official Links

Name	Link
OpenClaw Core	github.com/openclaw/openclaw
ClawHub Marketplace	clawhub.com
Official Docs	docs.openclaw.ai

SDKs & Tools

Name	Language	Description
openclaw-sdk	Python	Build & publish autonomous AI agents
mcp-bridge-openclaw	TypeScript	MCP multi-server bridge
amor71/openclaw-mcp	TypeScript	Native MCP client
henry-y/openclaw-paper-tools	Python	OpenClaw arXiv paper reader

Automated Research Tools

Name	Link	Description
AutoResearchClaw	GitHub	Fully autonomous 23-stage research pipeline: idea → experiment → conference-ready paper; multi-agent peer review
OpenClaw-Medical-Skills	GitHub	869 curated medical AI skills covering clinical work, genomics, drug discovery & bioinformatics
ScienceClaw	GitHub	Autonomous research pipeline; 266+ domain skills; 77+ databases
ClawCures	GitHub	AI campaign orchestrator for drug discovery; planner/critic loops; ADMET maps

Security References

Name	Link
PASB Framework	GitHub
SkillFortify	GitHub · PyPI
SecureClaw	GitHub
SlowMist Security Guide	GitHub
CVE-2026-25253	NVD
Security Guide	bitdoze.com

Benchmarks

Name	Link	Description
PinchBench	pinchbench.com · GitHub	23 real-world tasks across 8 categories; automated + LLM judge grading
EvoClaw	evo-claw.com · HuggingFace	Continuous software evolution benchmark; 7 repos / 5 languages / 98 milestones

Chinese Community / 中文社区

Name	Link	Description
OpenClaw China	BytePioneer-AI/openclaw-china	Domestic IM adaption (3,200+ Stars)
中文社区	clawd.org.cn	Feishu / DingTalk / WeCom / QQ
中文教程	openclawgithub.cc	Config & integration guides
Hello Claw	Datawhale	Datawhale tutorial
中文站	clawcn.net	Domestic LLM guide
Learn OpenClaw	learnopenclaw.com	Learning platform

Related Repositories

Know a great OpenClaw project we missed? Open a PR and help us keep this list growing!

Repository	Stars	Description
SamurAIGPT/awesome-openclaw	823	Comprehensive list of OpenClaw resources, tools, skills, tutorials & articles
mergisi/awesome-openclaw-agents	830+	177 production-ready AI agent templates across 24 categories
VoltAgent/awesome-openclaw-skills	—	Community curated skills collection
community/openclaw-recipes	—	Common automation recipes
templates/claw-templates	—	Starter templates for OpenClaw projects
pranciskus/discourse-openclaw	NEW	Discourse forum integration with 12 tools
wanikua/ByeByeClaw	NEW	One-command uninstaller for all Claw-family agents

---

🤝 Contributing

Contributions are welcome! We especially need help with:

• Papers — Adding missing OpenClaw-related papers with proper links
• Analysis — Improving paper notes and highlights
• Timeline — Updating the ecosystem timeline with new events
• Translation — Translating content between English and Chinese

Please submit via Pull Request. See README_CN.md for the Chinese version.

---

⭐ Star History