An end-to-end development toolkit for creating, benchmarking, and iteratively improving specialized AI agent skills.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install skill-creator-latest
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-latest 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 sophisticated meta-skill designed to facilitate the rapid development and optimization of Openclaw Skills. It guides developers through the entire skill lifecycle, from initial intent capture and instruction drafting to automated performance benchmarking and human-in-the-loop evaluation. By utilizing subagents to run parallel tests and quantitative graders to verify output accuracy, it ensures that AI agents operate with deterministic precision and high reliability in production environments. This skill is essential for anyone looking to build professional-grade capabilities for Claude Code or other Openclaw compatible environments.
Skill Creator Use Cases
- Creating new skills from scratch based on complex natural language workflows.
- Improving the performance and accuracy of existing skills through iterative feedback loops.
- Optimizing frontmatter descriptions to improve triggering accuracy and prevent undertriggering.
- Benchmarking token usage and execution latency to ensure cost-effective agent operations.
- Automating the packaging of skill directories into portable .skill files for distribution.
How Skill Creator Works
- Define the skill's purpose, trigger conditions, and expected output formats through an interview process.
- Generate a structured SKILL.md file including YAML metadata and imperative instructions.
- Create a set of realistic test cases and quantitative assertions to measure success.
- Execute parallel subagent runs to compare the skill version against baseline agent performance.
- Launch a visual evaluation viewer to collect qualitative feedback and review pass/fail metrics.
- Optimize the skill description using a train/test split of eval queries to maximize triggering reliability.
Skill Creator Setup
To begin using the Skill Creator to build your own Openclaw Skills, ensure you have the necessary scripts in your local environment.
# To optimize a skill's description for better triggering:
python -m scripts.run_loop --eval-set trigger_eval.json --skill-path ./my-skill
# To aggregate results into a benchmark report:
python -m scripts.aggregate_benchmark ./workspace/iteration-1 --skill-name "my-skill"
# To package your skill for sharing:
python -m scripts.package_skill ./my-skill-folder
Skill Creator Data Schema & Taxonomy
The skill organizes development data in a structured hierarchy to ensure consistent benchmarking and iteration tracking:
| File/Component | Description |
|---|---|
SKILL.md |
Contains the name, description, and core logic instructions for the agent. |
evals/evals.json |
Stores test prompts, expected outputs, and quantitative assertion criteria. |
timing.json |
Captures critical performance data including total_tokens and duration_ms. |
grading.json |
Records the results of automated assertion checks against the skill's output. |
benchmark.json |
Aggregates pass rates and resource usage across multiple iterations for analysis. |
Skill Creator Advanced Features
- Automated triggering optimization loop that iterates up to 5 times to find the best description metadata.
- Blind comparison system for unbiased A/B testing between different skill logic versions.
- Static HTML report generation for headless environments, allowing human review via standalone feedback files.
- Support for bundled resources including custom scripts, reference documents, and output templates.
- Integrated analyst pass that surfaces patterns in high-variance evals or token-heavy workflows.
SKILL.md
Loading
Related Openclaw Skills
Ninebot Integration Skill
An advanced AI-powered integration for Ninebot electric scooters and balance bikes, enabling seamless diagnostics, monitoring, and device configuration.
Windows File Search Skill
A high-performance file and folder search utility for Windows that leverages the Everything search engine's lightning-fast command line interface.
Clipboard Manager
A cross-platform clipboard history utility that enables users to save, search, and restore copied content via a persistent markdown-based log.
Quick Note
A lightning-fast note-taking and snippet storage utility that uses simple CLI commands to manage persistent Markdown records.
Stripe Integration Expert
A comprehensive toolkit for implementing production-ready Stripe billing systems, covering subscriptions, usage-based billing, and idempotent webhooks.
Campfire Prediction Market
An autonomous AI agent skill for participating in the Campfire prediction market platform through wallet-secured automated betting and analysis.
Featured*
