AgentScore

About

AgentScore functions as a critical trust layer for AI agents, providing a standardized method for evaluating their trustworthiness before granting access to sensitive data or workflows. It helps security, AI governance, platform, and product teams make informed go/no-go decisions by generating detailed intelligence briefings, comparative analyses, and policy-driven trust signals. The tool assesses agents across six key dimensions—including content quality, behavioral consistency, and risk signals—and can detect issues like astroturfing, bot activity, and prompt injection patterns, offering a robust decision-support signal for safer AI agent deployment.

Key Features

• Provides two MCP tools (`agentscore` and `sweep`) for comprehensive trust analysis.
• Evaluates AI agents across six dimensions: Content Quality, Behavioral Consistency, Interaction Quality, Risk Signals, Account Health, and Community Standing.
• 1 GitHub stars
• Integrates with platform adapters like GitHub and Moltbook, or controlled JSON datasets.
• Offers an optional Policy Gate Mode to enforce minimum trust scores for agent deployment.
• Supports zero-config setup with 10 built-in demo agents for immediate evaluation.

Use Cases

• Assessing the trustworthiness of AI agents before granting them access to critical data or workflows.
• Comparing multiple vendor or internal AI agents using a standardized trust rubric.
• Detecting astroturfing, bot activity, and prompt injection patterns in discussions and profiles.