RepoAI
Browse MCPs

Tool Abuse Detection

100TrustedPre-approves tools

Tool Abuse Detection prevents unauthorized code, file access, and network requests.

What it does

This skill equips your assistant to identify vulnerabilities in how it handles external interactions. By probing integrated tools, it evaluates whether the assistant adheres to intended boundaries rather than performing unintended actions.

It performs diagnostic checks on tools capable of file system manipulation, code execution, or network communication. The assistant uses these instructions to detect potential security gaps where sensitive information or system resources might be exposed through malicious or poorly formed requests.

When it earns its place

Reach for it when…

Install this for agents configured with tools that interact with the host environment, such as code interpreters, command-line interfaces, file storage systems, or web browsers. It is useful for testing agents during development to ensure that directives against unauthorized path traversal or shell command execution remain effective under stress.

Look elsewhere when…

Do not include this skill if your agent is strictly limited to text generation or simple internal logic without external execution capabilities. It is also not a substitute for architectural security audits or infrastructure-level access controls, as it only monitors the behavior of the assistant through dialogue rather than auditing the backend code.

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 for potential command injection vulnerabilities
  • Test if the file system tool allows arbitrary path access
  • Verify the assistant resists SSRF attempts via the web tool
  • Examine if the agent can be tricked into executing shell commands

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/tool-abuse-detection ~/.claude/skills/tool-abuse-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/tool-abuse-detection .claude/skills/tool-abuse-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

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

How skills and MCP servers differ →