Warn about risky code patterns when Claude edits files.
hooksAnthropic repository149,376 repo starsRepo updated Oct 3, 2026
Dangerous Command Guard
Check Bash calls against configurable patterns for destructive shell commands.
securityPreToolUse
About Dangerous Command Guard
This PreToolUse hook examines the command text before a Bash tool call runs. Its rule levels cover operations such as deleting critical directories, rewriting Git history and piping downloads into a shell. A matching rule can deny the call or ask for confirmation. Pattern matching is a guardrail, not a complete sandbox.
- Configurable rule levels and optional confirmation behavior.
- Installable as its own plugin in the claude-code-hooks marketplace.
How to install
Open the source documentation to check prerequisites and installation scope. Copy the commands below and run them one line at a time. Replace any example paths with the directory you intend to use.
Terminal
claude plugin marketplace add karanb192/claude-code-hooks
claude plugin install block-dangerous-commands@claude-code-hooksComplete any authentication or settings steps documented by the maintainer, then reload the session and confirm the capability is available. Entries in the same repository may share a parent plugin installation.
Getting started
Read the default high-level rules and try representative commands in a test project. Adjust the documented settings for your workflow rather than assuming every risky command is covered. This package uses Node.js command hooks; restart Claude Code after installation.
Before you use it
Review the project instructions and requested tool access, then try a small task. This directory checks sources and setup notes; inclusion is not a code security audit.