MaskClaw Core is an edge-side privacy protection suite that provides intelligent visual masking, behavior monitoring, and self-evolving security rules for AI agents.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install maskclaw
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 maskclaw 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 MaskClaw Core?
MaskClaw Core functions as a sophisticated privacy proxy framework designed for edge-side Tool-Use. It serves as a secure intermediary between cloud-based AI agents and local mobile or desktop user interfaces, acting as a digital security guard. By utilizing on-device MiniCPM-V models, this Openclaw Skills package identifies sensitive data and performs real-time de-sensitization before any information is transmitted to the cloud.
The framework is built on a four-layer architecture—Perception, Cognition, Tool-Use, and Evolution—ensuring that privacy protection isn't just a static filter but a dynamic system. By integrating this skill, developers can implement high-performance PII detection and visual obfuscation that adapts to specific user behaviors and application contexts without compromising local performance.
MaskClaw Core Use Cases
- Automatically blurring bank account numbers and phone numbers in UI screenshots during agent tasks.
- Intercepting and blocking unauthorized file-sharing actions in enterprise applications like DingTalk.
- Implementing a 'Confirm Before Action' workflow for high-risk PII data types.
- Continuously refining privacy rules through a self-evolving SOP mechanism based on user feedback.
How MaskClaw Core Works
- Perception: The system captures the current UI and uses RapidOCR to detect sensitive text and visual elements.
- Cognition: An on-device MiniCPM-V model analyzes the context and queries a ChromaDB RAG knowledge base for applicable privacy rules.
- Tool-Use: The Smart Masker tool applies Gaussian blur or mosaics to sensitive regions based on a five-level confidence adjudication.
- Monitoring: The Behavior Monitor tracks agent actions and logs user interventions to identify potential security gaps.
- Evolution: The Skill Evolution engine uses a hill-climbing algorithm to optimize security SOPs, testing them in a sandbox before deployment.
MaskClaw Core Setup
To get started with these Openclaw Skills, ensure you have Python 3.10+ installed and set up the core dependencies:
pip install rapidocr opencv-python numpy transformers chromadb
Next, launch the model service to handle on-device processing:
cd model_server
python minicpm_api.py
MaskClaw Core Data Schema & Taxonomy
MaskClaw Core organizes its security data and metadata through the following structures:
| Component | Data Managed | Format |
|---|---|---|
| Smart Masker | Bounding boxes, text labels, and masked image paths | JSON Output |
| Behavior Monitor | User actions, resolution results (Allow/Block), and scenario tags | Action Chain Log |
| Rule Evolution | SOP versions, iteration scores, and sandbox results | ChromaDB / JSON |
| Knowledge Base | Vector embeddings for privacy-preserving rules | ChromaDB RAG |
MaskClaw Core Advanced Features
- Five-level adjudication system (Allow, Block, Mask, Ask, Unsure) for granular control.
- Automated Hill-Climbing algorithm for continuous privacy rule optimization.
- Sub-50ms visual processing speed for real-time de-sensitization.
- Cross-platform support for Android, Windows, and macOS environments.
- Sandbox validation engine to ensure new rules pass safety thresholds before going live.
SKILL.md
Loading
Related Openclaw Skills
Cloudflare R2 S3 Storage Tool
A versatile utility for managing Cloudflare R2 storage using S3-compatible APIs to handle file uploads and public link generation.
WeChat Media Downloader
A robust tool for extracting and downloading media from WeChat Official Account articles by leveraging verified browser sessions.
MiMo Voice Assistant
A high-performance voice suite providing emotion-aware text-to-speech and transcription for AI agents.
Facebook Ad Tracker & Monitor
A comprehensive tool for tracking Facebook advertisers, analyzing creative performance, and monitoring product trends via the PPSPY API.
Agent Memory
A memory management system that enables AI agents to maintain persistent, cross-session long-term memory through structured file storage.
Best Video Editor for Beginners
An interactive AI assistant that matches your creative goals, hardware, and budget with the ideal beginner-friendly video editing software.
Featured*
