Coordination Engineering’s Key Leap: Jiuwen Claw Introduces the New Team Skills Paradigm

AI engineering is shifting from single‑agent "drive and govern" to multi‑agent "coordinate and evolve". The openJiuwen community, supported by Huawei, released the latest JiuwenClaw featuring Coordination Engineering, which enables autonomous division of labor, efficient communication, and seamless collaboration among multiple agents.

To avoid rebuilding collaboration from scratch each time, the community introduced Team Skills —the industry’s first standardized capability package for multi‑agent cooperation. Team Skills capture the entire collaboration process—including requirement decomposition, team formation, task allocation, communication mechanisms, conflict handling, and delivery specifications—into a reusable SOP.

What Team Skills Look Like

A Team Skill is essentially a directory structure:

<team-skill-name>/
├── SKILL.md            ← team name, purpose, members
├── roles/
│   ├── <role-a>.md
│   └── <role-b>.md
├── workflow.md         ← execution order
├── bind.md             ← error handling and boundaries
├── dependencies.yaml   ← external tool dependencies
└── examples/ | templates/ | assets/ …

Even a minimal SKILL.md plus a few role definitions can form a functional team, and more complex tasks can incrementally extend the structure.

Creating Team Skills

The platform provides a Team Skills creator (teamskill‑creator) and a Team Skills Hub for sharing. To create a multidisciplinary medical expert team, follow these steps:

Download the teamskill‑creator from the Team Skills Hub.

Install the skill on JiuwenClaw.

Enter the prompt: "Help me create a medical expert consultation team skill with comprehensive specialties that dynamically loads based on user symptom descriptions."

The creator generates a skill containing 23 AI medical specialists, automatically instantiating the appropriate experts for each case.

Practical Verification

In a Feishu channel, users can switch to team mode with /mode team, start or stop sessions with /new_session, and toggle planning or performance modes via /mode agent.plan and /mode agent.fast. When a user asks for a diagnosis, the system’s triage role parses the symptoms, dynamically creates the required specialist agents, assigns analysis focus, and a chief doctor aggregates the results into a comprehensive report. The entire workflow is visible, traceable, and repeatable.

Cross‑Framework Compatibility

Team Skills extend the Agent Skills open standard and run without adaptation on platforms such as Claude Code or Cursor. Any platform supporting Agent Skills can directly reuse Team Skills.

Co‑creation and Sharing

The openJiuwen Team Skills Hub hosts ready‑to‑use skills across eight domains—development, productivity, content creation, multimodal media, data science, compliance, lifestyle, and finance. Users can generate new skills with the creator, upload them to the hub, and contribute to a growing ecosystem of reusable multi‑agent collaboration patterns.

Conclusion

JiuwenClaw’s Team Skills standardize, reuse, distribute, and make multi‑agent collaboration framework‑agnostic, completing the Coordination Engineering stack from single‑agent utility to team‑level capability, turning ad‑hoc agent groups into repeatable, evolving team operations.