The Skill Creator is a comprehensive framework designed to help developers build, test, and refine high-performance AI agent skills through data-driven iteration and benchmarking.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install rex-skill-creator
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 rex-skill-creator 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?
The Skill Creator provides a structured environment for developing Openclaw Skills from scratch or improving existing ones. It bridges the gap between raw prompts and production-ready agent capabilities by implementing a rigorous human-in-the-loop evaluation cycle. By using this skill, developers can ensure their AI agents trigger reliably and execute complex, multi-step workflows with precision.
At its core, the skill manages the entire lifecycle of skill development: from initial intent capture and drafting instructions to running parallel benchmarks and analyzing token efficiency. It focuses on creating lean, effective prompts that utilize theory of mind to guide LLMs, rather than relying on rigid constraints, ensuring the resulting Openclaw Skills are both flexible and powerful.
Skill Creator Use Cases
- Creating a new automated workflow for specialized data processing tasks.
- Optimizing the trigger descriptions of existing Openclaw Skills to prevent under-triggering.
- Benchmarking the performance of an AI agent against a baseline without specific skill instructions.
- Debugging complex agent behaviors by analyzing parallel execution transcripts and grading assertions.
- Packaging custom logic and helper scripts into a portable format for distribution.
How Skill Creator Works
- Define the skill's purpose and trigger conditions through a structured interview process.
- Draft the SKILL.md file including YAML frontmatter and imperative instructions.
- Generate realistic test prompts and save them to a standardized evals.json configuration.
- Execute parallel runs using subagents to compare results with the skill versus a baseline configuration.
- Capture timing, token usage, and qualitative outputs for each test case.
- Use the evaluation viewer to grade results and collect human feedback for the next iteration.
- Run the description optimizer to maximize triggering accuracy based on common user queries.
Skill Creator Setup
To begin developing Openclaw Skills with this tool, ensure you have the necessary environment scripts available. You can initiate a benchmark aggregation or launch the evaluation viewer using the following commands:
# Aggregate benchmark results for a specific iteration
python -m scripts.aggregate_benchmark <workspace>/iteration-N --skill-name <name>
# Launch the local evaluation viewer to review test outputs
python eval-viewer/generate_review.py <workspace>/iteration-N --skill-name "your-skill-name"
Skill Creator Data Schema & Taxonomy
The skill organizes its development data using a strict directory and file taxonomy to ensure compatibility with the evaluation viewer:
| File/Folder | Description |
|---|---|
SKILL.md |
The core definition file containing metadata and markdown instructions. |
evals/evals.json |
Stores the test prompts, expected outputs, and quantitative assertions. |
scripts/ |
Contains bundled executable code for deterministic tasks. |
workspace/iteration-N/ |
Holds the outputs, timing data, and grading results for a specific test run. |
benchmark.json |
An aggregated summary of pass rates, token costs, and duration deltas. |
Skill Creator Advanced Features
- Multi-agent parallel execution for rapid benchmarking of with-skill vs. baseline performance.
- Blind comparison system to evaluate output quality without reviewer bias.
- Description optimization loop that automatically iterates on frontmatter to improve triggering rates.
- Integrated grading subagents that evaluate complex assertions against task outputs.
- Automated packaging system to bundle instructions, scripts, and assets into a single .skill file.
SKILL.md
Loading
Related Openclaw Skills
Memesio Meme Generator
A powerful MCP skill for searching meme templates, captioning images, and generating AI-driven memes directly within your AI agent environment.
GORM Expert Skill
An advanced AI agent skill for GORM v2 that automates model generation, performance auditing, and database architectural best practices.
Baidu OCR
A high-precision optical character recognition skill for AI agents that extracts text, formulas, and table data from images using the Baidu AI platform.
FORGE Cross-Chain Swap Agent
FORGE is a non-custodial cross-chain swap agent that routes quotes and builds transactions for 44+ crypto assets using the THORChain protocol.
Automation Workflows
A strategic framework for solopreneurs to identify, design, and implement high-ROI no-code automation workflows that reclaim valuable time.
Bybit V5 OpenAPI Skill
A specialized interface for querying Bybit V5 public market data using uxc and curated OpenAPI schemas.
Featured*
