Havuz

A collection of public writeups on attacks against AI agents: prompt injection, tool abuse, memory poisoning, and data exfiltration.

Agent-ergonomic CLI for TypeSafe's Jev: fast calibrated judgments (pick, rate, check, rank, triage, guard) from the shell

Self-hosted LLM security proxy. PII redaction, prompt injection defense, KVKK/GDPR/PCI-DSS compliance. Sub-millisecond latency

Every way an LLM can break one SDK to stop it. Injection. Hallucination. Leakage. Handled.

A modular threat detection library for AI and LLM pipelines. It provides language-agnostic guardrails through a configurable pipeline of loa

Trust nothing. Ship safely. — Skeptical-reading and prompt-injection defense skill for AI agents. Provenance tagging, red-flag patterns, ref

🛡 The reference playbook for pentesting AI chatbots & LLM-powered apps in one place. Ready-to-use payloads covering the full OWASP LLM Top 1

Local-first security scanner, MCP protocol inspector, dynamic fuzzer, Docker sandbox, and report generator for Model Context Protocol server

Free local guardrail for AI coding agents: every tool_call judged on-device (regex + Laya), allow/review/block

A terminal-based, multi-domain support-ticket triage agent grounded in a local support corpus.

Security scanner for MCP servers, source trees, and packages, with explicit findings and coverage reporting.

Concurrent Go CLI that detects hidden prompt-injection content in documents — invisible Unicode, homoglyphs, hidden HTML, and known trigger

Open-source test harness for AI agents. Stress-test production agents with adversarial multi-turn scenarios in CI

OpenAI Codex 自定义系统提示词注入工具 | System Prompt Injection | CTF Security Research Red Team | Auto Install | Cross-Platform

Myrtia — agent-native red teaming for LLMs and AI agents: a deliberating council of attacker models (propose, cross-critique, Borda vote, ch

⚡ Millisecond System-1 judgement for every tool call in DeepSeek Harness — Jev-powered risk classification & evidence-gated auto-approval. F

Open-source AI agent framework for PHP. Tools, memory, security guards and a ReAct loop built in. Works with Laravel, Symfony, WordPress, Dr

An open database of real-world AI agent incidents.

🛡️ Regex catches 0%, Meta's Prompt Guard 2 catches 1% of 629 realistic AgentDojo injection attacks when they're buried in tool output. Repro

An OWASP-aligned intentionally vulnerable platform for learning and testing AI, LLM, RAG, MCP, and Agentic AI security.

Red-team your RAG pipeline for prompt injection, source-document leakage, and cross-document smuggling, in CI. The gap between RAG eval (RAG

Python SDK for LLM guardrails with safety classification, PII detection, prompt injection defense, and grounding checks. Protect AI applicat

Runtime security for AI agents. In-process, zero dependencies, Apache 2.0.

Soup Wall: open-source security for AI agents and model traffic. Local Agent, Gateway, self-hosted Console. Apache-2.0.

Security-First Agentic IDE Coding Harness

Prompt-injection, jailbreak and secret-leak defense for DeepSeek Harness: Aho-Corasick detection with allow/ask/block interception and sanit

Claude skill that catches hidden prompt-injection text in documents before acting on them

AVE - Agentic Vulnerability Enumeration. The open behavioral classification standard for agentic AI components.

Governed Postgres access for AI assistants: an MCP gateway with OAuth 2.1 and OIDC login, one Postgres role per person, database-enforced re

Command-line tool for TypeSafe's Jev AI model

Privacy and security guardrails for Spring AI: request-scoped PII tokenization, least-privilege disclosure to tools, and tool authorization

Open-source runtime security and execution control for autonomous AI agents. ALLOW, BLOCK, ESCALATE at the execution boundary with auditable

Screens an AI agent's tool calls before they run and tool results before the agent reads them. An MCP proxy plus a hooks adapter for a clien

Crash-test insurance claim AI agents before production.

Enterprise Zero-Trust AI Security & Governance Gateway: Real-Time Prompt Injection Defense, Action Scoping, HITL Approvals, Post-Quantum Mer

Security-testing methodology as portable, tool-agnostic agent skills for Claude Code and any agent runtime: white-box bug hunting, AI-agent

Adversarial security testing for AI agents — prompt injection, tool misuse, memory poisoning, and MCP scanning, with CI-ready regression tes

Open-source AI security platform providing perimeter defense for LLMs and AI agents through swarm analysis, policy enforcement, adversarial

Security scanner for AI agent skills. Flags known prompt-injection, data-exfiltration and malicious-payload patterns before you install

Python SDK for accurate and verifiable agent tool use. Agents verify that answers came from the right source and were not changed. Downstrea

Evidence-grounded evaluation for AI agents — verifies each claim against the agent's real tool outputs (constrained, evidence-grounded model

Safety benchmark for personal AI agents under realistic prompt injection. 120 test cases across 5 professional domains, 3 injection vectors,

Find known secrets in your AI coding agent sessions (Claude Code, Codex, Cursor), then guard the next one with local hooks that trace risky

Agents propose. People approve. Change control for AI agents: everyday work runs on its own, anything an agent can't take back waits for a t

Skill router and prompt-injection firewall for AI coding agents: gives Claude Code, Codex and Cursor the few skills, MCP servers and tools e

Free security skills for AI-built apps — run /lictor-security-check in Claude Code before you ship. Plain English, read-only, no telemetry.

Evidence-grounded AI secretary for Claude Code and Codex, with frozen context, deterministic retrieval, bounded quality review, and explicit

Open-source guardrails for AI agents powered by decision models like Jev; checks prompts, retrieved content, tool calls and responses for pr

Default-deny MCP/JSON-RPC security gateway for agent tool-call authorization, prompt-injection detection, exfiltration controls, and audit t

BloodHound-style risk analysis for multi-agent AI architectures. Parses CrewAI, Dify, LangGraph, or generic YAML into a typed capability gra