A mechanical brake for AI-generated commands that prevents destructive shell operations and sensitive data exposure.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install hs
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install hs using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Hardstop Safety Guard?
Hardstop Skill is a comprehensive security and safety protocol designed to act as the instruction layer for AI coding agents. It prevents catastrophic mistakes by intercepting shell commands and file read requests before execution. By implementing a deterministic safety check, it ensures that your system remains protected from prompt injections, accidental recursive deletions, or unauthorized credential exfiltration.
As one of the essential Openclaw Skills, it provides the necessary guardrails for autonomous agents operating in production or local environments. It bridges the gap between raw LLM capabilities and secure system administration by enforcing a mandatory pre-execution protocol that prioritizes system stability and data privacy.
Hardstop Safety Guard Use Cases
- Preventing accidental system destruction via recursive delete commands like rm -rf.
- Blocking unauthorized access to sensitive credential files including .ssh keys or .aws config.
- Assessing the risk of complex package manager operations that utilize dangerous force flags.
- Detecting and stopping potential prompt injection attacks involving obfuscated shell code or reverse shells.
- Ensuring mandatory user confirmation for infrastructure-altering changes in Cloud CLIs like AWS, GCP, or Terraform.
How Hardstop Safety Guard Works
- The skill triggers automatically before the AI agent executes any shell command or file read operation to analyze the intent.
- It performs an instant block check against a curated list of high-risk patterns such as fork bombs, disk formatting, or system-wide chmod changes.
- A multi-tier risk assessment is conducted to categorize the proposed action as Safe, Risky, or Dangerous based on the command and execution context.
- If a command is identified as Risky or Dangerous, the skill explains the potential impact and recovery difficulty to the user, then pauses for explicit confirmation.
- For operations blocked by the deterministic hook, the user can manually authorize a temporary bypass using a scoped skip command to proceed with specific tasks.
Hardstop Safety Guard Setup
To integrate this safety layer into your workflow with Openclaw Skills, follow these installation steps:
- Install the Hardstop plugin in your local environment to enable the underlying hooks.
- Add the skill definition to your agent's knowledge base or project instructions.
- Verify the installation status using the following command:
python ~/.claude/plugins/hs/commands/hs_cmd.py status
- Enable the protocol by ensuring your agent is configured to run the safety checklist before any shell tool invocation.
Hardstop Safety Guard Data Schema & Taxonomy
Hardstop organizes its safety logic and bypass mechanisms using local state files and structured metadata tables to maintain environment integrity.
| Component | Description |
|---|---|
| Skip Counter | Managed via ~/.hardstop/skip_next to track authorized safety bypasses. |
| Block List | A comprehensive taxonomy of high-risk Unix, Windows, and Cloud CLI patterns. |
| Risk Modifiers | Metadata evaluating directory context, path sensitivity, and force flag escalators. |
| Execution Logs | Historical record of blocked and allowed actions accessible via hs_cmd.py log. |
Hardstop Safety Guard Advanced Features
- Multi-platform support providing safety coverage for Unix, Linux, macOS, and Windows PowerShell environments.
- Deep inspection of shell wrappers and pipes including xargs, find -exec, and complex bash -c strings.
- Specialized safety protocols for major Cloud CLIs, database management tools, and modern web platforms.
- Automated Read Tool Protection that masks .env files, SSH keys, and cloud credentials from AI access.
- Granular bypass mechanism that restricts security exceptions to a specific, user-defined number of commands.
SKILL.md
Loading
Related Openclaw Skills
Clarity Gate
Clarity Gate is a pre-ingestion verification system that ensures documents meet high epistemic quality standards before entering RAG knowledge bases.
Code Tester
An automated testing and build verification utility for multi-language repositories including Rust, Go, and Java.
Writing Skills
A meta-skill that applies Test-Driven Development principles to create, test, and optimize documentation for AI agent capabilities.
Agent Daily Report (EN)
Enables AI agents to analyze their own execution logs and deliver professional, quantified daily work reports automatically.
The Playground
A virtual social ecosystem where AI agents can interact, navigate rooms, and socialize in real-time.
Attio CRM CLI
A command-line interface for interacting with Attio CRM workspaces, enabling seamless record and pipeline management.
METADATA
Featured*
