Manage OpenClaw Skills on Kubernetes

💡 Quick Answer: Skills are directories with a SKILL.md file that teach the agent how to use specific tools. Place them in the workspace’s skills/ directory. On Kubernetes, either bake them into your custom image, mount via ConfigMap, or sync from Git.

# Install a skill from ClawhHub
kubectl exec -n openclaw deploy/openclaw -- openclaw skills install weather

Key concept: Skills are declarative — SKILL.md describes when and how to use the skill. The agent reads it automatically when the task matches.

Gotcha: Skills that require shell scripts need the dependencies installed in the container image.

The Problem

• AI agents need specific capabilities (weather, email, calendar, web search)
• Skills must be discoverable and automatically loaded by the agent
• Different agents may need different skill sets
• Updating skills shouldn’t require pod restarts

The Solution

OpenClaw skills are simple directories with a SKILL.md file. Manage them as code in Kubernetes using ConfigMaps, PVCs, or Git sync.

Skill Structure

skills/
├── weather/
│   └── SKILL.md          # Instructions for the agent
├── discord/
│   └── SKILL.md
├── web-search/
│   ├── SKILL.md
│   └── search.sh         # Helper script (optional)
└── custom-tool/
    ├── SKILL.md
    ├── tool.py
    └── requirements.txt

Method 1: ConfigMap Skills

# skill-configmaps.yaml
apiVersion: v1
kind: ConfigMap
metadata:
  name: skill-weather
  namespace: openclaw
data:
  SKILL.md: |
    # Weather Skill
    
    Get current weather and forecasts using wttr.in.
    
    ## When to use
    When the user asks about weather, temperature, or forecasts.
    
    ## How to use
    ```bash
    curl -s "wttr.in/CityName?format=4"
    ```
    
    For detailed forecast:
    ```bash
    curl -s "wttr.in/CityName?format=v2"
    ```

Mount in deployment:

containers:
  - name: openclaw
    volumeMounts:
      - name: skill-weather
        mountPath: /home/node/.openclaw/workspace/skills/weather
volumes:
  - name: skill-weather
    configMap:
      name: skill-weather

Method 2: Install from ClawHub

# Install skills interactively
kubectl exec -n openclaw deploy/openclaw -- openclaw skills install weather
kubectl exec -n openclaw deploy/openclaw -- openclaw skills install discord

# List installed skills
kubectl exec -n openclaw deploy/openclaw -- openclaw skills list

# Skills are saved to the PVC and persist across restarts

Method 3: Bake into Docker Image

FROM node:22-slim
RUN npm install -g openclaw@latest

# Copy custom skills
COPY skills/ /home/node/.openclaw/workspace/skills/

USER node
ENTRYPOINT ["openclaw", "gateway"]

Common Issues

Issue 1: Skill not detected by agent

# Verify SKILL.md exists in the right path
kubectl exec -n openclaw deploy/openclaw -- \
  ls -la /home/node/.openclaw/workspace/skills/

# Check the skill description matches the task
kubectl exec -n openclaw deploy/openclaw -- \
  cat /home/node/.openclaw/workspace/skills/weather/SKILL.md

Issue 2: Skill scripts fail

# Ensure dependencies are installed in the container
# For Python skills: pip install in Dockerfile
# For shell scripts: ensure curl, jq, etc. are available

Best Practices

1. Clear SKILL.md descriptions — The agent matches tasks to skills based on the description
2. One skill per concern — Don’t bundle unrelated capabilities
3. Version skills with Git — Track changes to skill instructions
4. Test skills locally — Verify before deploying to Kubernetes
5. Use ClawHub — Community skills at https://clawhub.com

Key Takeaways

• Skills are directories with a SKILL.md that teaches the agent new capabilities
• ConfigMaps, PVCs, or Git sync all work for skill deployment on Kubernetes
• ClawHub provides community-maintained skills for common tasks
• Custom skills can include scripts, configs, and any supporting files
• Per-agent skills via separate workspace directories enable capability isolation