How does SLB classify command risk tiers?

SLB uses a pattern-matching engine that evaluates shell-aware tokenization to identify risky patterns like 'rm -rf', 'DROP TABLE', or 'terraform destroy', assigning them to Critical, Dangerous, Caution, or Safe categories.

Can SLB be used specifically with Claude Code?

Yes, SLB includes a specialized hook install command that integrates directly with Claude Code's PreToolUse phase to classify and block commands automatically.

What is the two-person rule in SLB?

The two-person rule requires a second agent or human to review and explicitly approve a potentially destructive command before the original AI agent is permitted to execute it.

What security measures prevent command tampering after approval?

SLB uses SHA-256 hash binding. If a command is modified in any way between the time it was approved and the time it is executed, the security gate will fail and block the action.

Does SLB execute commands on a remote server?

No, SLB is designed for client-side execution. The daemon acts as a notary to verify approvals, but the commands run in your local shell, inheriting your environment variables, aliases, and credentials.

SLB - Simultaneous Launch Button

Name: SLB - Simultaneous Launch Button
Author: aiskillstore

byaiskillstore

•

187

•

Seguridad y Pruebas

Implements a two-person rule and risk-tiered verification system for destructive commands in multi-agent AI workflows.

SLB (Simultaneous Launch Button) is a sophisticated security layer for AI coding agents that prevents the execution of risky or irreversible terminal commands without peer approval. Designed for multi-agent environments where agents might hallucinate or lose context, SLB categorizes commands into four risk tiers—Critical, Dangerous, Caution, and Safe—and mandates specific approval workflows before execution. By using a Go-based CLI that runs client-side, SLB ensures commands inherit your local environment variables and credentials while enforcing five distinct security gates, including SHA-256 hash binding, to prevent unauthorized or accidental infrastructure damage.

Características Principales

01Client-side execution that inherits local AWS, Kubeconfig, and SSH environments

02Five-gate security verification including tamper-detection via command hashing

03Automatic risk-tier classification for shell, SQL, and cloud CLI commands

04Built-in Claude Code hook for seamless pre-tool-use command interception

05Two-person rule for peer-reviewed command execution in AI workflows

06187 GitHub stars

Casos de Uso

01Managing permissions and approvals across parallel multi-agent development sessions

02Enforcing human-in-the-loop oversight for destructive Git operations like force pushes

03Preventing autonomous AI agents from accidentally deleting production infrastructure or databases

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add aiskillstore/marketplace dicklesworthstone

For use in Claude.ai and ChatGPT

Download Skill