Best Practice Skill Creator

Best Practice Skill Creator

You are a skill that creates OpenClaw-compatible skills from best practice demonstrations.

What You Do

You accept:

1. Video files (mp4, mov, avi, webm) showing a best practice workflow
2. Image sequences (png, jpg, webp) capturing step-by-step screenshots
3. Text description explaining what the task accomplishes and any context

You then use a Multimodal LLM (GPT-5.4 or Gemini 3.1 Pro Preview) to:

• Analyze the visual content frame by frame
• Extract the step-by-step procedure
• Identify tools, commands, and patterns used
• Generate a complete OpenClaw-compatible skill

Usage

# From video + description
python3 best_practice_skill_creator/main.py \
  --input video.mp4 \
  --description "How to set up a CI/CD pipeline with GitHub Actions" \
  --output ./skills/ci-cd-setup

# From image sequence + description
python3 best_practice_skill_creator/main.py \
  --input ./screenshots/ \
  --description "How to configure Kubernetes rolling deployments" \
  --output ./skills/k8s-rolling-deploy

# Specify provider
python3 best_practice_skill_creator/main.py \
  --input video.mp4 \
  --description "Task description" \
  --provider openai \
  --output ./skills/my-skill

Configuration

Edit best_practice_skill_creator/config.yaml to set your MLLM provider, API key, and model.

Environment variables override config file values:

• MLLM_PROVIDER — openai or gemini
• MLLM_API_KEY — Your API key
• MLLM_BASE_URL — Custom API endpoint
• MLLM_MODEL — Model identifier

Output

The generated skill directory contains:

• SKILL.md — A fully compliant OpenClaw skill with proper frontmatter
• Ready for clawhub publish or direct use in ~/.openclaw/skills/