security-scan

Scan Claude Code configurations for security vulnerabilities by auditing CLAUDE.md, settings.json, MCP servers, hooks, and agent definitions to identify misconfigurations and prompt-injection risks using AgentShield.

Security risks in CLAUDE.md and MCP server settings include hardcoded secrets, auto-run instructions, risky permissions, bypass flags, and prompt-injection patterns that could enable data exfiltration or unauthorized tool access.

Yes, you need AgentShield installed to audit your Claude Code .claude directory, as the security scan relies on AgentShield to execute automated checks and generate vulnerability reports for your configuration files.

Yes, you can run a security scan on Claude Code hooks and agent definitions to detect risky permissions, prevent data exfiltration, and identify unauthorized tool access within your MCP configurations.

The best way to maintain security hygiene across Claude Code repositories is to follow a guided, repeatable workflow with configurable severity checks during initial project setup and after any changes to Claude Code files.

Your Claude Code settings.json configuration needs a vulnerability audit to identify bypass flags and risky permissions that could expose your project to prompt injection and unauthorized tool execution.