Can Hardstop block cloud infrastructure destruction?

Yes, it monitors Cloud CLIs like AWS, gcloud, and kubectl, specifically flagging destructive actions like 'terraform destroy' or instance terminations.

What does the Hardstop skill actually do?

It intercepts shell commands and file access requests generated by Claude, checking them against a database of dangerous patterns to prevent system damage or data loss.

Does it protect against data theft?

Hardstop includes exfiltration detection that blocks attempts to read sensitive files like .ssh keys, .env files, and AWS credentials.

How does Hardstop handle 'risky' but necessary commands?

For risky commands that aren't instantly blocked, it triggers a mandatory warning protocol, explaining the risks and waiting for explicit user confirmation before proceeding.

Hardstop Safety Guard

Name: Hardstop Safety Guard
Author: frmoretto

byfrmoretto

•

安全与测试

Protects systems by validating AI-generated shell commands and file operations against a comprehensive safety protocol before execution.

Hardstop acts as a vital safety layer for agentic AI workflows, preventing catastrophic system errors, accidental data loss, and prompt injection attacks. It implements a rigorous pre-execution protocol that categorizes actions as safe, risky, or dangerous, blocking known destructive patterns across Unix, Windows, and major Cloud CLIs. By identifying hallucinations or malicious instructions before they reach the terminal, Hardstop provides a mechanical brake that ensures sensitive credentials remain protected and infrastructure stays intact.

主要功能

01Exfiltration detection to prevent unauthorized reading of SSH keys and env files

02Cloud CLI protection for AWS, GCP, Azure, Terraform, and Kubernetes operations

03Instant blocking of destructive shell patterns like recursive deletes and fork bombs

04Multi-platform safety checks for Unix/Linux, macOS, and Windows environments

056 GitHub stars

06Package manager validation for dangerous force-flags in dpkg, apt, and rpm

使用场景

01Securing agentic workflows against prompt injection attacks that target the shell

02Preventing AI hallucinations from executing destructive system-level commands

03Enforcing human-in-the-loop confirmation for high-risk infrastructure modifications

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add frmoretto/hardstop hardstop

For use in Claude.ai and ChatGPT

Download Skill