What is the primary purpose of the HardStop skill?

HardStop acts as a safety barrier between the AI and your terminal, catching dangerous commands—whether from AI mistakes, hallucinations, or prompt injection—before they can execute.

Is HardStop compatible with Windows?

Yes, HardStop features comprehensive protection for Windows-specific threats, including dangerous PowerShell patterns, registry deletions, and system directory modifications.

Can I bypass a command that HardStop blocks?

Yes, if a command is blocked or flagged, you can manually review it and use the 'skip' argument to authorize the execution if you are certain it is safe.

Does HardStop support cloud infrastructure tools?

Yes, it includes specific safety patterns for AWS, GCP, kubectl, Terraform, and Firebase to prevent accidental resource destruction.

How does HardStop handle sensitive files like SSH keys?

The skill includes an instant block list for any command that attempts to read or exfiltrate sensitive directories such as .ssh, .aws, or .env files.

HardStop Shell Safety

Name: HardStop Shell Safety
Author: frmoretto

byfrmoretto

•

安全与测试

Protects your system by intercepting and validating AI-generated shell commands and file access before execution.

HardStop provides a critical safety layer for agentic AI environments like Claude Code, acting as a deterministic seatbelt against accidental or malicious system damage. It implements a multi-tier risk assessment protocol that automatically blocks high-severity destructive commands (such as root-level recursive deletes or disk formatting), flags risky operations for user confirmation, and prevents the unauthorized reading of sensitive credential files like .env or SSH keys. This skill is essential for developers running AI agents locally, ensuring that hallucinations or prompt injections cannot result in irreparable data loss or security breaches.

主要功能

01Credential protection blocking access to .env, .ssh, and .aws folders

0219 GitHub stars

03Multi-tier risk assessment protocol (Safe, Risky, Dangerous, Instant Block)

04Real-time command interception and heuristic safety validation

05Cross-platform support covering Linux, macOS, Windows, and Cloud CLIs

06Detection of hidden risks in shell wrappers like xargs, find -exec, and piped bash scripts

使用场景

01Preventing accidental system-wide file deletion during autonomous coding or cleanup tasks

02Securing local environments against prompt injection attacks targeting sensitive developer credentials

03Implementing human-in-the-loop verification for destructive cloud infrastructure changes

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add frmoretto/hardstop hs

For use in Claude.ai and ChatGPT

Download Skill