A high-performance prompt engineering and vector matching engine that clarifies vague requests and routes them to the correct AI capabilities.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install smart-refinement
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 smart-refinement 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 Smart Refinement Skill?
The Smart Refinement Skill is a sophisticated middleware for AI agents designed to bridge the gap between ambiguous user intent and executable technical prompts. By integrating a Prompt Refinement Module and a Vector Optimizer, it automatically identifies incomplete requests and reconstructs them into structured, high-clarity instructions. This ensures that Openclaw Skills operate with maximum efficiency, reducing unnecessary LLM calls by up to 90%.
Beyond simple text cleanup, the skill utilizes TF-IDF and Cosine Similarity to match user requests against six core skill databases, including code generation, data analysis, and system operations. It maintains a deep understanding of conversation history and project context, making it an essential tool for developers building complex multi-agent systems or team task allocation frameworks.
Smart Refinement Skill Use Cases
- Refining vague user commands like "process that file" into detailed, actionable instructions.
- Automatically routing user queries to specific specialized tools such as web search or data analysis modules.
- Optimizing team workflows by matching incoming tasks to the most qualified AI agents based on skill scores.
- Reducing token consumption and latency in LLM-based applications through pre-processing and caching.
How Smart Refinement Skill Works
- The system receives a raw user message and captures the current conversation context.
- It analyzes the message for ambiguity; if the confidence score is low, the Prompt Refinement Module generates a structured version.
- The Vector Optimizer calculates similarity scores against the internal skill database to find the best functional match.
- Integrated context from project history and operation logs is merged with the refined prompt.
- The skill returns a comprehensive JSON object containing the refined prompt, matched skills, and an execution guide for the agent.
Smart Refinement Skill Setup
Install the skill via the official registry for Openclaw Skills:
npx clawhub install smart-refinement
Alternatively, for manual installation:
- Copy
smart_refinement_system.pyinto your skills directory. - Ensure your Python environment has the necessary dependencies for TF-IDF and vector math.
Smart Refinement Skill Data Schema & Taxonomy
The skill produces a structured JSON output to guide agent execution. Key fields include:
| Field | Type | Description |
|---|---|---|
refined_prompt |
string | The optimized version of the user's original input |
needs_refinement |
boolean | Flag indicating if the original input was ambiguous |
skill_matches |
array | List of matched skill categories with similarity scores |
intent |
object | Extracted user intent and entity data |
system_stats |
object | Performance metrics and processing time |
execution_guide |
string | Step-by-step instructions for the following agent |
Smart Refinement Skill Advanced Features
- Multi-agent collaboration support for complex workflows involving multiple Openclaw Skills.
- Real-time performance monitoring with sub-1ms average response times.
- Customizable refinement and vector match thresholds via a local
config.jsonfile. - Intelligent caching of high-frequency keyword matches and vector calculations to further reduce latency.
- Seamless integration with Team Managers for automated task delegation based on skill scores.
SKILL.md
Loading
Related Openclaw Skills
Capability Evolver
A deterministic meta-skill for AI agents to analyze runtime logs, identify error patterns, and generate structured self-improvement proposals.
CTF Forensics & Signal Analysis
A comprehensive toolkit for AI agents to perform deep digital forensics, memory analysis, and signal decoding for capture-the-flag competitions.
ctf-misc
A versatile toolkit for solving complex miscellaneous Capture The Flag (CTF) challenges including jail escapes, esoteric encodings, and system exploitation.
Double Search
A high-performance parallel search engine aggregator combining Tavily and Kimi for comprehensive AI-driven research.
Volcengine Voice Notification
A professional AI agent skill for sending automated voice notifications and alerts via Volcengine Voice Service API.
byted-seedream-image-generate
A high-performance image generation skill leveraging Volcano Engine Seedream models to create professional-grade visual content from text and reference images.
Featured*
