AgentGuards

What's supported

Which coding tools AgentGuards protects, on which systems, and what it checks. Windows results were tested on a real Windows 11 machine on 25 September 2026.

ToolHow to set it upWorks onWhat it checks
macOSLinuxWindowsPromptsShell commandsWeb contentFile edits
Claude Code
in a terminal
One command✓✓✓✓✓✓✓
OpenAI Codex
CLI
One command✓✓✓✓✓◐4✓
Claude Desktop app
Code tab
Plugin, a few clicks✓—✓✓✓✓✓
Claude Code on the web
cloud sessions at claude.ai/code
Manual, one-time setup———✓✓✓✓
Gemini CLI
Extension, manual✓✓✗2✓✓✓—
GitHub Copilot CLI
Plugin, manual✓✓✗2✓✓◐1—
OpenCode
Plugin, manual✓✓✓✓✓✓—
ChatGPT
AgentGuards app, on request
Plugin directory, a few clicks———✗◐5◐6—
VS Code Copilot, Cursor
and other MCP clients
MCP server✓✓✓◐3◐3◐3—
Your own app
any language
REST API / Gateway✓✓✓✓———
Claude Desktop — Cowork tab
tested, the hooks don't run there
Not supported✗
Claude chat
claude.ai / the app's Chat tab
Not supported✗
✓ supported◐ partly✗ not supported— not applicable
  1. 1 Pages fetched with shell commands (curl, wget) or with the built-in web fetch and web search tools are checked. Files a command saves to disk and then reads aren't covered yet.
  2. 2 Tested: the hooks don't start on Windows yet. A Windows version is planned; use macOS, Linux or WSL for now.
  3. 3 The AI decides whether to call AgentGuards, so a check can be skipped. Hooks and plugins can't be.
  4. 4 Pages fetched with shell commands (curl, wget, Python or Node scripts, and on Windows curl.exe, Invoke-WebRequest and Invoke-RestMethod), the files those commands download, and MCP fetch and search tools are checked. Codex's built-in web search runs on OpenAI's servers, so it isn't.
  5. 5 On request: ChatGPT calls AgentGuards when it decides a command or piece of code needs checking, and ChatGPT has no hooks, so nothing is checked automatically. Tested in ChatGPT on the web: rm -rf / is flagged critical, curl | sudo bash needs approval, and code containing an AWS secret key is blocked.
  6. 6 ChatGPT can send page, email or document text to AgentGuards to check for prompt injection. Tested in ChatGPT on the web: an email with injected instructions is blocked. ChatGPT's own browsing isn't checked automatically.

Extension scanning in your editor

The AgentGuards VS Code extension checks your editor's extensions for malware and vulnerable bundled code. Tested in VS Code on Linux and Windows 11, and on macOS (Apple silicon) for starting up and the installer download. Not yet tested in Cursor, Windsurf or VSCodium.

One command for Claude Code and Codex

macOS / Linux:  curl -fsSL https://agentguards.co/install.sh | sh
Windows:        irm https://agentguards.co/install.ps1 | iex

It installs AgentGuards and signs you in through the browser — no API key to copy. Open the guided setup.

Need this as an image? Download the grid (PNG). Using something not listed? Tell us.