prompt-security

A full-stack AI Red Teaming platform securing AI ecosystems via Agent Scan, Skills Scan, MCP scan, AI Infra scan and LLM jailbreak evaluation.

The fastest Trust Layer for AI Agents

Leaderboard Comparing LLM Agent Security on System Prompt Leakage and Attack Probes

Lokale Werkbank zum Prüfen, Vergleichen und Härten von Skills für KI-Coding-Agenten.

Open-source prompt injection detector — 5 layers, 91.7% F1, ~27ms, offline, Apache 2.0

Extensible Go microkernel for LLM guardrails, quota enforcement and policy plugins.

A Pi extension that replaces sensitive data with realistic placeholders before it reaches the LLM provider, then restores it locally for users and tools.

Universal Prompt Security Standard (UPSS): A framework for externalizing, securing, and managing LLM prompts and genAI systems, inspired by and extending OWASP OPSS concepts for any organization or project.

LLM Penetration Testing Framework - Discover vulnerabilities in AI applications before attackers do. 100attacks + AI-powered adaptive mode.

🚀 Unofficial Node.js SDK for Prompt Security's Protection API.

Mithra Scanner is an interactive API testing tool for prompt injection, refusal detection, and LLM security benchmarking. It supports YAML-based rule definitions, custom refusal lists, REST API integration, and provides detailed CLI output for security testing of language model endpoints.

The enterprise AI governance and safety control plane for AI agent prompts

CloakPrompt is a CLI tool that redacts secrets (passwords, API keys, credentials, etc.) before sending data to AI models.

Privacy-first prompt sanitization. Fully local. Zero cloud calls. Fast, smart, and built for real-world AI workflows.

Local Rust proxy that redacts secrets from prompts sent to AI coding agents (week-1 risk-validation scaffold)

Lightweight AI security framework for prompt validation, output scanning, risk scoring, sensitive data detection, and Zero Trust policy enforcement.

Local-first sanitizer for scripts, logs, prompts, and support text — clean secrets, hostnames, and org-specific terms on your device before sharing. PowerShell-aware Portfolio-code mode keeps sanitized code readable. Windows, Linux, and a live browser demo.

Zero-LLM deterministic jailbreak regression benchmark: runs a repeatable battery of known-pattern attacks against an LLM endpoint and scores refusal/partial/compliance across runs. Catch when a model or prompt update silently got weaker. Single-turn, responsible-use. pip install hermes-jailbench

Fuller 7-Layer Agentic AI Security Framework (F7-LAS)

R package for LLM safety guardrails across prompts, outputs, RAG context, PII, secrets, and local Ollama/NLP workflows.