An expert development framework for building, iteratively testing, and optimizing specialized skills for AI coding agents.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install skill-creator-4
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install skill-creator-4 using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Skill Creator?
Skill Creator is a high-performance development tool designed to architect and refine Openclaw Skills. It facilitates a rigorous lifecycle for AI agent behaviors, moving beyond simple prompting into a structured engineering process. By leveraging subagents and automated evaluation loops, it allows developers to define intent, capture complex workflows, and ensure reliability through both qualitative human review and quantitative benchmarking.
This skill is essential for users who want to transform repetitive manual sequences into automated, reliable agent capabilities. It ensures that every skill is lean, performant, and correctly triggered by optimizing the frontmatter descriptions and bundling necessary execution scripts. Whether you are building from scratch or improving an existing skill, Skill Creator provides the harness needed to reach production-grade AI performance.
Skill Creator Use Cases
- Transforming a successful manual multi-step workflow into a reusable agent skill.
- Benchmarking an AI agent's performance against a baseline to verify improvements.
- Optimizing skill triggering descriptions to prevent undertriggering in complex environments.
- Iteratively refining output formats for high-stakes data extraction or code generation tasks.
- Creating specialized reference libraries for agents to consult during domain-specific tasks.
How Skill Creator Works
- Capture Intent: The user defines the skill's purpose, trigger contexts, and expected output formats.
- Draft SKILL.md: A structured markdown file is created containing YAML metadata and imperative instructions.
- Test Case Generation: Define 2-3 realistic test prompts and save them to a standardized eval schema.
- Parallel Execution: Spawn subagents to run the task both with and without the skill to establish a performance baseline.
- Evaluation & Review: Use the evaluation viewer to compare outputs qualitatively and run quantitative assertions for grading.
- Iterative Refinement: Modify the skill instructions based on feedback and rerun tests until the desired pass rate is achieved.
- Trigger Optimization: Run an automated loop to refine the skill description for maximum triggering accuracy.
Skill Creator Setup
To begin creating Openclaw Skills, ensure you have the core toolkit installed and initialize a new skill directory:
# Create a directory for your new skill
mkdir -p my-new-skill
touch my-new-skill/SKILL.md
# Run the skill description improver for existing skills
python -m scripts.run_loop --skill-path ./my-new-skill --max-iterations 5
# Package the finished skill for installation
python -m scripts.package_skill ./my-new-skill
Skill Creator Data Schema & Taxonomy
Skill Creator organizes data within a dedicated workspace to maintain iteration history and benchmarking integrity:
| Component | File/Path | Description |
|---|---|---|
| Metadata | SKILL.md |
Contains YAML frontmatter (name, description) and instructions. |
| Test Set | evals/evals.json |
Stores test prompts, expected outputs, and assertions. |
| Workspaces | iteration-N/ |
Contains subdirectories for each evaluation run (with/without skill). |
| Analytics | benchmark.json |
Aggregated stats including pass rates, token usage, and duration. |
| Resources | scripts/ |
Bundled executable code for repetitive or deterministic sub-tasks. |
Skill Creator Advanced Features
- Multi-agent benchmarking: Automatically spawns parallel subagents to compare 'with-skill' vs 'baseline' performance.
- Automated Assertion Grading: Uses a grader subagent to verify specific technical requirements in outputs.
- Triggering Optimization Loop: Conducts a 20-query evaluation set to maximize skill invocation accuracy via held-out test scores.
- Blind Comparison System: Employs an independent agent to judge quality between two versions without bias.
- Resource Bundling: Supports external reference files and scripts for complex, multi-domain skill organization.
SKILL.md
Loading
Related Openclaw Skills
Banker
An expert financial services advisor designed to provide clarity on loans, mortgages, and banking product comparisons.
Social Worker
A specialized AI assistant designed to guide users through the complexities of social services, welfare benefits, and community advocacy.
Advisor
A sophisticated decision-support skill designed to provide objective, multi-angle analysis and stress-testing for high-stakes business and personal choices.
Form CRO
An expert AI skill designed to analyze and optimize non-signup forms to maximize completion rates and capture high-quality data.
Oil Price Query
A zero-config CLI tool for retrieving localized fuel prices across China without requiring API keys.
add.tools
A dedicated search engine for AI agents to discover, rank, and integrate free and x402-paid tools and APIs.
Featured*
