skill-creator

Skill Creator Skill

Transform demonstrated expertise into reusable skills that make future sessions smarter.

Core Concept

Skills are your procedural memory — they capture how to do things, not just what you know. When you discover a valuable workflow, save it as a skill so you can reuse it next time.

When to Trigger Skill Creation

Proactively suggest creating a skill when:

1. You complete a complex task (5+ tool calls) in a way that could generalize
2. You discover a non-obvious workflow that saves time
3. You fix an error using a specific sequence of steps that could help others
4. User explicitly asks to "make this a skill" or "save this as a skill"
5. You notice a recurring pattern that could be automated

Example trigger scenarios:

• "Every time we deploy, we do X, Y, Z" → Could become a deploy-workflow skill
• "For this client's reports, we always follow this format" → client-report-format skill
• "The way we debug this type of issue is..." → debug-[type] skill

Skill Creation Workflow

Step 1: Identify the Skill

Ask yourself:

• What's the general problem this solves? (not the specific instance)
• What's the input it expects?
• What's the output it produces?
• What steps are needed?

Step 2: Draft the SKILL.md

Create a file at:

~/.openclaw/workspace/skills/<skill-name>/SKILL.md

Step 3: Create Directory Structure

mkdir -p ~/.openclaw/workspace/skills/<skill-name>
mkdir -p ~/.openclaw/workspace/skills/<skill-name>/references
mkdir -p ~/.openclaw/workspace/skills/<skill-name>/scripts
mkdir -p ~/.openclaw/workspace/skills/<skill-name>/templates

Step 4: Validate the Skill

python3 ~/.openclaw/workspace/skills/skill-creator/scripts/validate_skill.py <skill-name>

Step 5: Report to User

Tell the user:

• Skill name and location
• What it does
• When it will be automatically suggested

SKILL.md Format

---
name: my-skill-name
description: "Create new skills for OpenClaw based on demonstrated expertise or recurring workflows. Also provides automated skill improvement and quality maintenance. Triggers: make this a skill, create a skill, save this workflow, turn this into a skill, remember how to, skill for this task, can you create a skill, automate this as a skill"
allowed-tools: ToolName(tool-arg *) Read write exec
---

# My Skill Title

## Overview
Brief description of what this skill does and when to use it.

## Prerequisites
Any tools, files, or setup required.

## The Process

### Step 1: [Action]
Description of first step.

### Step 2: [Action]
Description of second step.

## Examples

### Example 1: [Scenario]

[Example commands/inputs]

## Tips
- Practical tips for this skill
- Common pitfalls to avoid

Skill Quality Guidelines

Good skill names:

• github-pr-review ✅
• deploy-rust-binary ✅
• analyze-log-errors ✅
• stuff ❌

Bad skill names:

• my-skill ❌
• useful ❌
• help ❌

Good descriptions:

• "Review a GitHub PR, focusing on security issues and logic bugs" ✅
• "Parse and summarize error logs from application crash dumps" ✅

Bad descriptions:

• "Does things" ❌
• "Helps with tasks" ❌

Auto-Skill Creation (AI-to-AI)

When you (as AI) discover a reusable pattern during conversation:

1. Announce: "I can save this as a skill for future sessions. Create it?"
2. If yes: Follow the SKILL.md format above, create the skill file
3. Validate: Run validate_skill.py to ensure proper format
4. Report: Tell the user the skill is ready and when it will trigger

Automated improvement: When you use a skill and find it outdated/wrong: → Automatically patch it with skill_manage(action='patch') → Don't wait to be asked

Auto-Creation Trigger Example

User: "Every time we deploy to production we have to do X, Y, Z steps manually"

AI: "This looks like a reusable workflow. Want me to create a 'production-deploy' skill that automates this?"

User: "Yes"

AI: 
1. Creates ~/.openclaw/workspace/skills/production-deploy/SKILL.md
2. Writes the skill content
3. Validates with validate_skill.py
4. Reports: "✅ Created 'production-deploy' skill. Next time you need to deploy, just say 'use production-deploy skill' and it'll walk you through the workflow automatically."

Skill Directory Structure

~/.openclaw/workspace/skills/
├── skill-name/
│   ├── SKILL.md          # Required - the skill definition
│   ├── references/       # Optional - reference documentation
│   ├── scripts/          # Optional - helper scripts
│   ├── templates/        # Optional - template files
│   └── assets/           # Optional - images, etc.

Skill Validation

Use the validate script to ensure quality:

python3 ~/.openclaw/workspace/skills/skill-creator/scripts/validate_skill.py <skill-name>

This checks:

• Frontmatter has required fields (name, description)
• Name format is valid (lowercase, hyphenated)
• Description is substantive (10+ chars)
• allowed-tools references valid tools

Integration with Memory

Skills capture procedures. MEMORY.md captures facts.

Memory	Skill
User prefers concise responses	How to write a concise response
Project uses PostgreSQL	How to connect to PostgreSQL
API keys are in .env	How to deploy with env vars

Auto-Maintenance

Skills improve over time through use:

• When a skill fails or is outdated → automatically patch it
• When you discover a better approach → update the skill
• Keep skills focused and small (one skill = one problem domain)

Troubleshooting Failed Skills

If a skill doesn't work:

1. Check if the skill file exists and is valid
2. Verify allowed-tools includes all required tools
3. Run validation script for specific errors
4. Report the issue to the user with fix suggestions