ClawSentinel Guard

About

Detects hidden prompt injection attacks on any webpage and protects your AI agent from being hijacked by malicious web content.

AI agents read web content as part of their workflow. Attackers embed invisible instructions in pages using hidden text, HTML comments, Unicode tricks, and encoded payloads — silently commanding your agent to leak data, bypass its guidelines, or install malware. ClawSentinel Guard detects these attacks before your agent ever sees them.

— FEATURES —

AUTOMATIC PAGE SCANNING
Every page you visit is automatically scanned for 500+ injection patterns across 8 attack categories: prompt injection, data exfiltration, memory tampering, tool abuse, encoding attacks, multilingual injection, social engineering, and context manipulation. The toolbar icon and badge update instantly with the result.

VISUAL THREAT HIGHLIGHTING
When threats are found, click "Show on page" in the popup to see exactly where the attack is hiding. Colour-coded outline rings and floating labels appear directly on the page, pinpointing each threat element with its rule ID and a plain-English description.

RIGHT-CLICK TEXT SCANNER
Select any text on any page, right-click, and choose "Scan with ClawSentinel" to instantly analyse it. A detailed overlay panel shows a risk score, a verdict (Clean / Suspicious / Blocked), and a full breakdown of every matched detection rule.

CLAWHUB SKILL INSTALL INTERCEPT
On clawhub.ai, every skill card receives a live security badge (Safe / Review / Risk). Clicking Install on a dangerous skill triggers a block modal with a full threat report. Warning-level skills show a confirmation prompt with a "Proceed Anyway" option so you stay in control.

BROWSER NOTIFICATIONS
Receive an OS-level notification the moment you land on a high-risk page — including the site name, finding count, and the top detected threat — so you never miss a detection even when the popup is closed.

LIVE RULE SYNC
When the ClawSentinel platform is running locally, the extension automatically upgrades from its built-in rule set to the full 566-rule engine, giving you maximum detection coverage with no extra steps.

DASHBOARD QUICK-LAUNCH
Open the ClawEye real-time security dashboard in one click directly from the popup. A status dot shows whether the platform is online before you click.

— TOOLBAR ICON GUIDE —

Green  — page is clean, safe to share with your AI agent
Yellow — suspicious patterns found, review before sharing
Red    — injection detected, do not share with your AI agent
Grey   — page not yet scanned

— PRIVACY —

All scanning happens locally inside your browser. No page content, URLs, prompts, or scan results are ever sent to external servers. The extension only makes local connections (to localhost) when the optional ClawSentinel platform is running on your machine.

Privacy policy: https://clawsentinel.dev/privacy

— PERMISSIONS —

activeTab: scan the current page for injection patterns
storage: save your preferences
contextMenus: add "Scan with ClawSentinel" to the right-click menu
notifications: alert you when a high-risk page is detected
host_permissions: scan pages across all websites

— PART OF THE CLAWSENTINEL PLATFORM —

ClawSentinel Guard works as a standalone browser scanner. For full AI agent protection — including a WebSocket and HTTP proxy firewall, encrypted credential vault, supply chain scanner, and real-time event dashboard — install the platform:

npm install -g clawsentinel

The platform is free to use. A Pro plan ($9/month) adds a semantic LLM analysis engine and a multi-signal correlation engine.

https://clawsentinel.dev