← All repos

agentshield

AI agent security scanner. Detect vulnerabilities in agent configurations, MCP servers, and tool permissions. Available as CLI, GitHub Action, ECC plugin, and GitHub App integration. 🛡️

GitHubcerebralvalley.ai/e/claude-code-hackathon
ai-agentanthropicclaude-codehackathonmcpopussecurity
Browse cluster: Claude AI Agent Frameworks & MCP Tools →
192commits
4contributors
4languages

Tech stack & purpose

AgentShield is a security auditor for AI agent configurations that detects vulnerabilities in Claude Code setups, MCP servers, hooks, and agent prompts. Built in TypeScript with a CLI entry point using Commander, it scans configuration files (CLAUDE.md, settings.json, mcp.json) and applies 102 rules across five categories—secrets detection, permissions, hooks, MCP servers, and agent config—assigning findings a severity level and producing graded reports (A–F, 0–100 score). The project also includes an Opus 4.6 adversarial analysis pipeline and MiniClaw, a sandboxed secure agent server, and is distributed as an npm package, GitHub Action, and GitHub App integration. Created at the Claude Code Hackathon by participants in the Everything Claude Code ecosystem, it addresses a gap in automated security tooling for the rapidly growing AI agent ecosystem.

Languages

TypeScript
96.0%
JavaScript
2.7%
HTML
0.9%
Shell
0.3%

Contributors