security-scan

Scanning Claude Code configuration files for security vulnerabilities involves auditing CLAUDE.md, settings.json, mcp.json, hooks, and agent definitions to flag hardcoded secrets, unsafe permissions, and exfiltration risks before onboarding or releasing.

Prompt injection risk in MCP server configuration occurs when untrusted instructions hijack agent behavior, detected by auditing tool access permissions and settings.json patterns to identify permissive tool access and unsafe shell usage.

Auditing Claude Code hooks and agents for secrets and exfiltration risks requires reviewing hook scripts and agent files for hardcoded credentials, unsafe shell usage, and data exfiltration patterns using AgentShield to provide prioritized remediation.

Checking .claude configuration for unsafe permissions before committing changes is supported by scanning settings.json and MCP definitions to identify overly permissive tool access, ensuring repository hardening prior to release or onboarding.

A configuration audit for Claude Code repositories is needed when onboarding to an existing repository, tightening permissions before a release, or checking recent config changes for exposure to prompt injection and secret leakage.

Limitations of automated security scanning for Claude Code configuration include reliance on pattern matching to detect risky configurations, meaning novel exfiltration techniques or context-dependent prompt injection vectors may require manual review.