AgentGuards — Extension Scanner

Extensions run with your full user permissions: they can read your SSH keys, cloud credentials and source code, and talk to any server. Malicious extensions have been published to both the VS Code Marketplace and Open VSX.

AgentGuards checks every extension you have installed, and every one you install or update, and warns you before a risky one can do harm.

What it checks

• Malware behaviour in the extension's code — reading credential files (~/.ssh, ~/.aws, browser profiles, crypto wallets), sending data to exfiltration endpoints (raw IPs, paste sites, Discord/Telegram webhooks), downloading and running code, obfuscated code, and bundled native programs. Signals are judged in combination, so a language server that legitimately runs programs isn't flagged for that alone.
• Known-malware packages bundled inside the extension.
• Known vulnerabilities (CVEs) in bundled dependencies.
• Publisher reputation — unverified publishers, very new extensions, and names that imitate a popular extension.

What you see

• Malicious or high risk: a warning with a one-click Uninstall.
• Medium risk (for example, a bundled dependency with a known vulnerability): a notification. You can turn these off in settings.
• The AgentGuards view in the activity bar lists every extension with its verdict and findings.

Connect your coding agents

The Coding Agents view shows whether the Claude Code and Codex command-line tools on this computer are protected by AgentGuards, and connects them in one click. Connected, AgentGuards screens their prompts, shell commands and fetched web content.

Connecting runs the official AgentGuards installer in a terminal, so you see exactly what it does: it installs the AgentGuards plugin through each agent's own plugin command and saves your API key to ~/.agentguards/credentials.json, readable only by you. The installer is downloaded from the AgentGuards GitHub release and checked against the release's SHA-256 checksums before it runs. Checking the status tests the key each agent sends with a read-only request to AgentGuards, which uses no quota. An agent shows Protected only when its key works.

Privacy

The extension sends AgentGuards only each extension's ID, version and platform, plus your editor's name. It never sends code, files, or anything from your workspace. AgentGuards downloads the extension from the public store and checks it on its own servers.

Extensions installed from a file (.vsix) or a private gallery aren't in a public store, so they're listed as not checked.

Requirements

An AgentGuards account. Each extension version checked uses one request from your plan's quota (re-checks of the same version within 30 days are free). Create a free account.

Getting started

1. Install the extension.
2. Click Sign In in the AgentGuards view, and approve the editor in your browser — or reuse the key the AgentGuards installer already saved, or paste an API key.
3. Your installed extensions are checked within a few minutes. New installs and updates are checked automatically.

Settings

License

FSL-1.1-MIT. © AlphaOps Ltd.