Can I use this for skills I'm currently developing locally?

Absolutely. Safe Skill Install includes a 'Scan Local' mode designed specifically to vet skills in local directories or those already installed in your environment.

How does Safe Skill Install prevent prompt injection from bypassing security?

The tool uses a two-layer architecture where security decisions are made by a deterministic bash wrapper script rather than the LLM, ensuring that malicious instructions within a skill cannot influence the safety assessment.

Does this protect against Git-based vulnerabilities?

Yes, it specifically checks for Git version 2.45.1+ to mitigate CVE-2024-32002 and utilizes archive downloads to avoid the execution risks associated with standard git clone operations.

What scanning technology does this tool use?

It leverages Cisco's skill-scanner, which combines static YARA-based pattern matching with behavioral analysis to identify suspicious code intent in Python and other languages.

Safe Skill Install

Name: Safe Skill Install
Author: swannysec

byswannysec

0•

セキュリティとテスト

Secures your AI development environment by performing automated supply chain security scans on Claude Code skills before installation.

Safe Skill Install provides a critical security boundary for Claude Code users, preventing malicious code execution during the plugin installation process. By wrapping Cisco’s skill-scanner in a deterministic bash execution layer, the tool identifies malware, suspicious behavioral patterns, and supply chain vulnerabilities without relying on LLM interpretation for security decisions. It mitigates advanced risks like CVE-2024-32002 through hardened Git operations and archive-first downloads, ensuring that any skill added to your agent—whether from GitHub, the Claude marketplace, or local paths—is thoroughly vetted and safe for your production environment.

主な機能

01Hardened Git operations and archive downloads to prevent clone-time exploits

02Deterministic security classification using Cisco’s skill-scanner engine

03Multi-source support for GitHub repositories, local directories, and marketplace plugins

04Integrated audit logging and scan history management for security compliance

050 GitHub stars

06Static and behavioral analysis to detect malicious YARA patterns and suspicious logic

ユースケース

01Protecting enterprise development environments from AI supply chain attacks

02Performing automated security audits on locally developed plugins

03Vetting community-contributed Claude Code skills before integration

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add swannysec/robot-tools safe-skill-install

For use in Claude.ai and ChatGPT

Download Skill