PromptShield

PromptShield is a security scanner for AI coding assistant configurations across IBM Bob, Claude Code, and Cursor. It audits MCP servers, skills, custom modes, and AI workflow files for known exploit classes, then returns structured findings for terminal, JSON, SARIF, HTML, or MCP-based assistant workflows. PromptShield provides deterministic security analysis and remediation guidance for AI-assistant configuration risk. Together they make AI-assisted development safer and faster to review. PromptShield currently detects five categories. PS-001 catches chained-command bypass through auto-approved shell utilities. PS-002 detects toxic skills and malicious agent instructions. PS-003 detects MCP stdio command-injection risks. PS-004 detects over-privileged custom modes without guardrails. PS-005 detects comment-and-control prompt injection in CI workflows. Together, these cover the main places where AI coding assistants move from helpful automation into unsafe execution paths.