Does Hardstop interfere with my normal coding workflow?

No. Safe, read-only commands like 'ls', 'git status', and project-scoped actions are allowed to run silently. The skill only triggers warnings for high-risk or system-altering operations.

How does Hardstop protect against accidental data loss?

Hardstop scans every command for destructive patterns like 'rm -rf' or 'format' and categorizes them as Dangerous. It then pauses execution and requires the user to provide explicit confirmation or choose a safer alternative.

Does this skill cover cloud provider commands?

Yes, Hardstop includes safety rules for destructive operations in AWS, Google Cloud, Azure, and Kubernetes, preventing the accidental termination of instances or deletion of namespaces.

Can it detect malicious code hidden in shell wrappers?

Yes, Hardstop is specifically designed to catch dangerous commands hidden inside wrappers like 'bash -c', 'xargs', or 'find -exec', which are often used to bypass basic filters.

Hardstop Safety Protocol

Name: Hardstop Safety Protocol
Author: frmoretto

byfrmoretto

•

Security & Testing

Protects your system by intercepting and validating AI-generated shell commands before execution to prevent data loss and security breaches.

Hardstop acts as a rock-solid safety layer for Claude Code, functioning as a mechanical brake for agentic AI actions. It implements a mandatory pre-execution protocol that evaluates every command against a comprehensive database of dangerous patterns, including recursive deletions, system formatting, and reverse shells. By categorizing actions into Safe, Risky, or Dangerous tiers, it ensures that high-stakes operations—such as cloud infrastructure destruction or credential exfiltration—are blocked or require explicit human confirmation, effectively neutralizing risks from AI hallucinations and prompt injection attacks.

Key Features

01Support for identifying dangerous shell wrappers and package manager overrides

02Mandatory pre-execution safety protocol for all shell and file operations

03Context-aware risk assessment (Safe, Risky, Dangerous) to minimize friction

04Instant block list for destructive Unix, Windows, and Cloud CLI commands

056 GitHub stars

06Detection of data exfiltration attempts and unauthorized credential access

Use Cases

01Preventing accidental system-wide file deletions during autonomous cleanup tasks

02Auditing AI-generated infrastructure changes in AWS, GCP, or Kubernetes

03Defending against prompt injection attacks that attempt to run malicious scripts

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