slowmist-agent-security

To check a GitHub repository for prompt-injection risks, the framework analyzes external inputs and project configurations using standardized templates to detect malicious patterns and output a structured risk assessment. It evaluates trust levels to prevent data leakage and exploitation.

You can review an on-chain address for security risks by submitting it to the assessment workflow. The framework evaluates the address alongside URLs and documents, applying universal templates to generate a trust assessment and risk rating for your AI agent integration.

The best way to detect malicious patterns in Skill or MCP installations is to run a pre-installation security review. The framework audits configurations and external inputs to identify prompt-injection vulnerabilities and outputs a risk rating requiring human approval for high-risk decisions.

The security review framework supports human-in-the-loop approvals by requiring human approval for HIGH or REJECT decisions. It outputs universal risk ratings and trust assessments to guide thorough review workflows, preventing automated exploitation when evaluating external inputs.

To assess the security of a product or service integration, the framework analyzes the external inputs and configurations for prompt-injection vulnerabilities. It uses standardized templates to produce a comprehensive risk rating and trust assessment to guide your integration decisions.