Agent Skill Evaluator

About

Agent Skill Evaluator is a specialized tool designed to verify the integrity and safety of agent skills and MCP servers before they are deployed. It automatically retrieves skill files from GitHub, websites, or direct URLs and subjects them to a rigorous multi-step evaluation process, including prompt injection detection, malicious script analysis, and permission auditing. The skill provides actionable security reports with standardized risk scoring, detailed evidence of vulnerabilities, and definitive recommendations, ensuring that users can extend their AI's capabilities without compromising system security or data privacy.

Key Features

• Source authenticity and community reputation verification
• Automated risk scoring across five security dimensions
• Standardized Markdown and PDF assessment report generation
• 6 GitHub stars
• Detection of prompt injection patterns and hidden instructions
• Security analysis of executable scripts and asset files

Use Cases

• Assessing whether a GitHub-hosted skill is safe to download and use
• Identifying hidden 'jailbreak' or 'ignore instructions' prompts within a skill file
• Reviewing a skill's scripts for unauthorized network or file system access