A security-focused web fetching and search tool that shields AI models from malicious prompt injections found in live web content.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install safe-web
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 safe-web 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 safe-web: Secure Web Search and Fetching?
safe-web is a robust security extension designed for the Openclaw Skills ecosystem to mitigate the risks associated with processing untrusted web data. While standard web tools simply retrieve content, safe-web acts as a protective proxy, utilizing the PromptGuard scanning engine to analyze incoming text for hidden instructions, system impersonation, or role manipulation before it enters the AI's context window.
By prioritizing safety without sacrificing utility, this skill allows developers to build more resilient agents that can research the open web while maintaining strict security boundaries. It integrates seamlessly into existing workflows, providing a fail-closed architecture that ensures unverified or high-risk content is never passed to the LLM.
safe-web: Secure Web Search and Fetching Use Cases
- Safely fetching and summarizing articles from untrusted third-party websites.
- Performing broad web searches where results may contain malicious SEO-injected prompts.
- Automating data collection for AI analysis while ensuring the source material is clean of prompt injections.
- Implementing secure research workflows within broader Openclaw Skills configurations.
How safe-web: Secure Web Search and Fetching Works
- The user or agent initiates a request using the fetch or search command.
- The tool retrieves the web content via the requests library or the Brave Search API.
- Content is sanitized using BeautifulSoup to strip away potentially dangerous or irrelevant HTML elements like scripts and styles.
- The extracted text is passed to the PromptGuard engine for pattern-based threat detection.
- If the content is deemed clean, it is returned to the agent; if a threat is detected, the tool exits with a security alert and blocks the data.
safe-web: Secure Web Search and Fetching Setup
Dependencies
Ensure PromptGuard is installed in your workspace first:
cd /home/linuxbrew/.openclaw/workspace/skills/prompt-guard
pip3 install --break-system-packages -e .
Installation
Install the required Python packages for web processing:
pip3 install --break-system-packages requests beautifulsoup4
Global Access
To use the command directly, create a symlink to the script:
sudo ln -s /home/linuxbrew/.openclaw/workspace/skills/safe-web/scripts/safe-web.py /usr/local/bin/safe-web
safe-web: Secure Web Search and Fetching Data Schema & Taxonomy
| Component | Description |
|---|---|
| Output Formats | Supports raw text, structured JSON, and direct file output for automation. |
| Threat Levels | Classifies content based on PromptGuard severity (e.g., CRITICAL, MEDIUM). |
| Exit Codes | Uses specific codes: 0 for success, 1 for errors, and 2 for security blocks. |
| Metadata | Includes source URL, character counts, and specific matched attack patterns. |
safe-web: Secure Web Search and Fetching Advanced Features
- Strict Mode: Automatically blocks content at lower threat thresholds for maximum security.
- JSON Integration: Enables seamless data piping into other Openclaw Skills for automated processing.
- Fail-Closed Logic: Guarantees that if scanning fails or dependencies are missing, the tool will not return unverified data.
- Search Filtering: Automatically scans and removes suspicious search results from the final list provided to the AI.
SKILL.md
Loading
Related Openclaw Skills
LSP Code Navigation
A multi-language LSP client that manages persistent background daemons for precise code navigation and semantic analysis.
YouTube Analytics Toolkit
A comprehensive data analysis toolkit for the YouTube Data API v3 that enables automated channel benchmarking, video performance tracking, and search results analysis.
DataForSEO Keyword Research
An automated SEO research toolkit for Openclaw Skills that leverages the DataForSEO API for keyword discovery, competitor intelligence, and SERP analysis.
GA4 Analytics Toolkit
A comprehensive toolkit for Google Analytics 4, Search Console, and URL Indexing to monitor and optimize website performance through automated data retrieval.
OpenAI Codex CLI
A powerful terminal-based coding assistant that integrates OpenAI Codex directly into your local development workflow for automated refactoring, debugging, and feature implementation.
Migma
A terminal-based AI email engine for designing, sending, and managing marketing campaigns and contact segments.
METADATA
Featured*
