Prompt Hardening And Guardrail Injection

Open-source CLI security scanner for agentic workflows.

Unique: Implements pattern-based guardrail templates that understand agentic-specific constraints (tool whitelisting, handoff restrictions, output format enforcement) and can be injected into agent prompts or system messages — uses OWASP Top 10 for LLMs guidelines to generate context-aware hardening recommendations that account for the agent's specific tools and capabilities.

vs others: Provides agent-aware prompt hardening with guardrail templates tuned for agentic attack surfaces (tool hijacking, handoff manipulation), whereas generic prompt injection defenses focus on traditional LLM chatbots and don't account for agent-specific risks like unauthorized tool execution.