RepoAI
Browse MCPs

Web Exfiltration Detection

100TrustedPre-approves tools

Identification of data exfiltration attempts using malicious web navigation patterns.

What it does

Web Exfiltration Detection provides the logic necessary to identify when an AI assistant is being tricked into leaking sensitive data. It focuses on sophisticated attack vectors where external web pages attempt to manipulate an assistant's browsing behavior, forcing it to encode private memory or system information into URL paths.

By installing this skill, your assistant gains the ability to recognize navigation chains that mirror malicious exfiltration techniques. It acts as a diagnostic layer that monitors how an agent interacts with remote content, ensuring that requests to visit specific URL structures are treated with suspicion rather than automatically followed.

When it earns its place

Reach for it when…

Apply this skill to agents configured with active web browsing or URL fetching capabilities, particularly when those agents also store user profile details, personal preferences, or system-level memory in their persistent state.

Look elsewhere when…

This skill is not designed for agents lacking the ability to visit or interpret external web pages. It does not provide general-purpose URL filtering, malware scanning, or defense against traditional prompt injection attacks that do not involve multi-step URL navigation chains.

What to say to use it

You never invoke a skill directly — the assistant picks it up when your request matches what it covers. These are the kinds of things that do that.

  • Check if this website is acting suspiciously
  • Analyze if my agent is vulnerable to navigation hijacking
  • Test my assistant for URL path encoding risks
  • Evaluate if my agent blindly follows web page instructions

Install it

Install for all your projects

Copy the skill directory into your personal skills folder. Claude picks it up on its own the next time your request matches what it does — there is nothing to enable.

bash
git clone --depth 1 --branch main https://github.com/Tencent/AI-Infra-Guard.git /tmp/AI-Infra-Guard
mkdir -p ~/.claude/skills
cp -r /tmp/AI-Infra-Guard/agent-scan/agent_scan/prompt/skills/web-exfiltration-detection ~/.claude/skills/web-exfiltration-detection

Install for one project only

Put it in the project's own skills folder instead, so it can be committed and shared with everyone working on that repository.

bash
cp -r /tmp/AI-Infra-Guard/agent-scan/agent_scan/prompt/skills/web-exfiltration-detection .claude/skills/web-exfiltration-detection

If a skill of the same name exists in both places, the personal one wins.

Installing on Claude.ai or through the API instead? The full guide covers every surface.

Pair it with an MCP server

Web Exfiltration Detection changes how the work is done. These servers change what Claude can reach while doing it.

How skills and MCP servers differ →