Skill Creator is a comprehensive development framework designed to build, test, and optimize modular AI agent capabilities through iterative evaluation loops.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install skill-creator-claude
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-claude 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 provides a rigorous, industrial-grade workflow for developing new AI agent functionalities. It moves beyond simple prompting by offering a structured lifecycle: capturing user intent, drafting technical specifications, and executing parallel test runs to compare skill performance against baselines. This ensures that every skill is not just functional, but optimized for reliability and accuracy across diverse edge cases.
At its core, this skill utilizes a progressive disclosure architecture, managing metadata, body instructions, and bundled resources like scripts and assets. By incorporating Openclaw Skills into your development environment, you gain access to an automated evaluation viewer and quantitative benchmarking tools that transform the way AI capabilities are engineered and refined.
Skill Creator Use Cases
- Creating complex, multi-step AI agent skills from scratch based on conversational workflows.
- Optimizing existing Openclaw Skills to improve triggering accuracy and reduce model undertriggering.
- Running qualitative and quantitative benchmarks to compare new skill versions against baseline performance.
- Automating the generation of test assertions and realistic evaluation queries for AI behavior verification.
How Skill Creator Works
- The user defines the intended functionality and triggers for the new capability.
- Skill Creator drafts the SKILL.md file, following strict structural and safety principles.
- The system generates realistic test prompts and saves them to a standardized evaluation schema.
- Parallel subagents execute tasks both with and without the new skill to establish a performance delta.
- Automated grading scripts evaluate assertions and capture timing and token usage data.
- A web-based evaluation viewer is launched for the user to review outputs and provide iterative feedback.
- The skill is packaged into a deployable format once performance targets are met.
Skill Creator Setup
To initialize the development environment for Openclaw Skills, ensure you have Python 3 installed and follow these steps:
# Initialize the skill workspace
mkdir -p my-skill-workspace
cd my-skill-workspace
# Create the skill structure
mkdir -p my-skill/{scripts,references,assets}
# Run the evaluation viewer (after generating benchmark.json)
python eval-viewer/generate_review.py iteration-1 --skill-name "my-skill"
# Package the finalized skill
python -m scripts.package_skill path/to/my-skill
Skill Creator Data Schema & Taxonomy
Skill Creator organizes its development lifecycle through a multi-layered directory structure and JSON-based metadata tracking:
| File/Directory | Description |
|---|---|
SKILL.md |
The core definition containing YAML frontmatter and Markdown instructions. |
evals/evals.json |
Stores test prompts, expected outputs, and quantitative assertions. |
iteration-N/ |
Contains specific run data, including with_skill and baseline outputs. |
timing.json |
Captures critical performance metrics: total_tokens and duration_ms. |
grading.json |
Records pass/fail status of assertions with supporting evidence. |
Skill Creator Advanced Features
- Parallel subagent spawning for real-time A/B testing and blind comparisons.
- Description optimization loop using automated trigger testing to minimize model undertriggering.
- Automated script bundling that identifies repeated subagent workflows and promotes them to reusable skill assets.
- Headless mode support for remote development environments using static HTML report generation.
SKILL.md
Loading
Related Openclaw Skills
Airflow Read-Only Skill
A secure, read-only interface to inspect Apache Airflow DAGs, runs, and task instances via the Stable REST API.
Crypto Sentiment Monitor
A real-time analysis tool that aggregates social media, search trends, and exchange data to track cryptocurrency market sentiment.
EMR Analyzer
A specialized diagnostic toolkit for analyzing health, configurations, and logs of open-source big data ecosystems on Linux systems.
Baidu SEO Toolkit
A specialized SEO optimization suite designed to maximize visibility on major Chinese search engines like Baidu, Sogou, and 360 Search.
UpLead
The UpLead skill for Openclaw Skills enables automated B2B sales intelligence and lead verification through the Membrane framework.
Article Taster
A multi-dimensional article analysis tool that evaluates content quality, detects AI-generated text, and identifies original writing to optimize reading time.
Featured*
