AgentLens — See What Agents See

See machine-readable webpage and PDF content that is not normally visible to you.

See what AI sees. AgentLens reveals suspicious instructions and hidden content that AI agents may read from webpages and PDFs—even when that content is difficult or impossible for you to see. Want to see AgentLens in action? Try it on these safe test pages, which contain examples of hidden content and prompt-injection techniques: – https://ref.jock.pl/modern-web/ – https://garden-advice.netlify.app/ – https://www.agentpwn.com/attacks Web pages contain more than visible text. AI tools may also encounter transparent or clipped text, off-screen elements, accessibility attributes, metadata, comments, hidden containers, and other machine-readable content. Attackers can use these surfaces to insert instructions that attempt to redirect an agent, expose sensitive information, invoke tools, or send data somewhere unexpected. AgentLens makes those differences visible before they become a problem. WHAT AGENTLENS DOES • Detects suspicious agent-directed instructions Scan the current page for instruction overrides, credential requests, data-exfiltration attempts, tool-use commands, prompt delimiters, concealment tactics, and common obfuscation patterns. • Reveals hidden instructions on the page AgentLens highlights the nearest visible source location and displays concealed text in a readable overlay. Move between every detected signal using the previous and next controls. • Strips suspicious content Remove detected instruction sources from the working page before giving its content to an AI tool. Restore the original page at any time by reloading it. • Explains every finding Instead of returning an unexplained score, AgentLens shows: – What was detected – Why it was flagged – How it was concealed – Which extraction profiles could encounter it – Where it appeared in the page – Which rule produced the finding – What action is recommended • Separates suspicious instructions from ordinary hidden content Machine-readable does not automatically mean malicious. AgentLens distinguishes instruction signals from neutral visibility differences such as responsive menus, accessibility labels, and other routine page elements. • Provides a complete scan report Open the full report to review risk levels, inspect evidence, copy source locators, reveal findings on the original page, strip individual sources, and export a bounded JSON report. • Supports live protection Optionally scan pages after navigation and receive an immediate alert when medium- or high-risk instruction signals appear. Use AgentLens on all websites or limit automatic access to websites you choose. • Scans supported PDFs Inspect digital and OCR-text PDF layers for invisible text, clipped or off-page content, suspicious annotations, metadata, and agent-directed instructions. Image-only PDFs are not OCR’d. • Keeps a local scan history Revisit pages you have scanned without creating an account or sending browsing history to a server. History can be disabled or cleared at any time. • Uses editable rules Review the active deterministic ruleset, customize detection behavior, and import or export rules for reuse or auditing. PRIVATE BY DESIGN AgentLens performs scanning locally in your browser. – No cloud scanning – No AI model required – No telemetry – No account – No remote rule downloads – No permanent scan history unless you enable it – Reports are exported only when you explicitly request them AgentLens requests website access only for the protection mode you select. Manual scans use the access granted by your toolbar action. BUILT FOR PEOPLE WHO USE AI ON THE WEB AgentLens is useful when you: – Copy webpage content into an AI chat – Let an agent summarize or research a page – Use browser automation or tool-enabled assistants – Review unfamiliar documents or websites – Test prompt-injection defenses – Audit how a page exposes information to machines – Need evidence you can inspect instead of a black-box warning AgentLens does not claim to reproduce the exact behavior of every AI agent. It compares multiple machine-readable extraction profiles and applies transparent, versioned rules so you can understand what was found and make the final decision yourself. Install AgentLens to reveal the instructions hiding behind the page—and decide what your AI tools are allowed to read.