agent-guard

Description

**The safety check an AI agent runs *before* it acts.** Every agent that installs packages, runs shell commands, or commits/outputs text can do real damage — install malware, wipe a disk, or leak a credential. agent-guard is the cheap "look before you leap" gate an agent calls before the irreversible step. | tool | call it before… | |---|---| | `check_package` | `pip/npm install` — is it real, malware, a typosquat, or does it run code on install? | | `check_command` | running a shell command — destructive / RCE (rm -rf, curl\|bash, dd, force-push)? | | `scan_secrets` | committing / pasting /logging — does the text leak an API key, token, or private key? || `scan_project` | shipping a web backend — fail-open auth, unsigned payment webhooks, SQL injection, SSRF? | Dependency-free checks. If a result is high/critical, stop and get human confirmation before proceeding.

Community Metrics

Categories

• mcp

Tags

• mcp
• smithery

Capabilities

• Remote MCP server

Information