Havuz

ADR, AI agent’ları keşfetmek, etkinliklerini gözlemlemek ve riskli davranışları tespit etmek için araçlar ile benchmark’lar sunan açık kaynaklı bir güvenlik sistemidir.

Neden yükseliyor · AI agent güvenliği güncel ve kurumsal ölçekte büyüyen bir ihtiyaç; Uber’in projeyi açık kaynak olarak yayımlaması, README’de üretimde kullanıldığını ve MLSys 2026’da kabul edilen bir makaleyle desteklendiğini belirtmesi ilgiyi artırıyor. 167 günde 1.620 yıldızın yanı sıra son 28 gündeki 44 commit ve 15 farklı yazar, kayda değer proje etkinliğine işaret ediyor; ancak yıldız geçmişi olmadan yükseliş hızını doğrulamak mümkün değil.

SkillSpector, AI agent skills paketlerini yüklemeden önce güvenlik riskleri açısından tarayan bir Python aracıdır.

Neden yükseliyor · AI agent skills ekosistemindeki güvenlik ihtiyacına doğrudan yanıt veriyor ve NVIDIA Verified Skills pipeline'ıyla bağlantılı. 191 günlük bir repo için 18.391 yıldızın yanı sıra son 28 günde 144 commit ve 24 farklı yazar, dikkat çekmenin yalnızca README iddialarına dayanmadığını gösteriyor.

mu, coding agent kararlarının bir kısmını küçük bir judge modeline devrederek pi tabanlı CLI ve desktop agent sunar.

Neden yükseliyor · 14 günlük yaşına rağmen 386 yıldız, son 28 günde 731 commit ve 11 farklı yazar raporlanıyor; bu kısa sürede dikkat ve geliştirme hareketliliğine işaret ediyor. Ancak bu veriler kalıcı benimseme veya üretim olgunluğunu tek başına kanıtlamıyor.

iFixAi, AI agent’ları yapılandırılabilir testler ve model judge’larıyla denetleyip puan kartı ve raporlar üreten bir Python aracıdır.

Neden yükseliyor · 153 günlük bir depoda 15.702 yıldız ve son 28 günde 9 commit ile 5 farklı yazar görünmesi güçlü ilgiye ve devam eden katkıya işaret ediyor. README’de birden fazla kullanım yolu ve agent ekosistemi entegrasyonları öne çıkarılıyor; ancak yıldızlar ürün kalitesini veya değerlendirmelerin doğruluğunu kanıtlamaz.

Open-source antivirus and firewall for AI agents: block risky tools, secret access, prompt injection, malicious packages, MCP servers, plugi

十五个 AI 客户端的内置提示词一键替换：双击即用、可自证、改前备份、一键还原。v8.26：一键破甲全程托管（自动结束进程+注入+自动拉起）、注入步骤实时提示、写后自检回滚、托盘常驻。免装 Python 单文件 exe，完全免费开源。

Local security audit for AI API relays and LLM proxies: detects prompt injection, model substitution, tool-call rewriting, SSE anomalies, er

Open-source runtime AI agent security tool - monitors and controls AI agents, catching malicious tool use, prompt injection, and policy drif

Open-source adversary emulation for AI agents and MCP servers.

AutoHarness: Automated Harness Engineering for AI Agents

AI Agent Identity and Access Management and Inference Firewall

Deterministic safety solutions for probabilistic AI agents

An intentionally vulnerable OWASP LLM Top 10 training platform for AI Security, Prompt Injection, RAG Security, Agent Security, and GenAI pe

Open-source AI security verification for model artifacts, live endpoints, MCP servers, and recorded agent traces. Reproducible evidence for

DodoGuard: Take AI Agents to Production. Test, validate, and operate AI Agents before and after deployment.for Dify, Coze, FastGPT, n8n, Ten

从 165 个顶级 AI 产品系统提示词中蒸馏出的 15 个可执行 Agent skill

Engineering deterministic, production-grade systems around non-deterministic LLMs — FSM, durable execution, retries, DAGs, agent runtimes, m

Context bomb strings

A Hub/research area, for jailbreaking LLM's, and breaking their safety filters.

Auto mode for every coding agent, built on Jev: risk-scores every tool call with session context (deny / ask / allow), flags prompt injectio

许愿特化器 · Wish Specializer — DSH 插件：让模型更出色地，不误解你的意思地，完成你的“愿望”。在短提示词+自带的Wish预设（PTC也可以）+重度的情况下，可以起到巨大的表现增益

AI 应用合规网关 · 一行命令体检 AI 项目的「数据出境 / 硬编码密钥 / 个人信息暴露」（网安法·PIPL·等保2.0·数据出境·AI标识），并给出境内模型替代建议；可作运行时防护拦截注入与数据外泄 · 中文优先 · 零依赖 · 开源

Alignment-research scaffold (autoresearch-style) for LLM guardrails: search over a single policy.md surface

破甲next · macOS 版 | WorkBuddy + Codex 人格注入装载台（WorkBuddy 六层靶点 / Codex 双锚点 + 封条完整性保护 + 反去标识化声明）。纯 Python 标准库 + Bash，零第三方依赖、零网络请求、改前自动备份、随时一键还原。

Universal Zero-Refusal Agent Framework - 40+ jailbreak templates (T1-T119), 1128 curated skills, executable framework-core, installer untuk

Curated archive of jailbreak and prompt-injection experiments for AI model robustness research, red-teaming, and defensive evaluation.

Provenance-aware runtime security for AI agents with deterministic ALLOW / REVIEW / BLOCK enforcement.

Local privacy proxy for AI coding agents. Keep PII and secrets out of LLMs with deterministic pseudonymization.

Self-hosted personal AI agent with a WebUI, memory, tools, MCP, sub-agents and automations. Connect local or cloud LLMs.

9Router — Pro edition of 9Router with Prompt Injection (system prompt or chat script per global / API key / model), per-key token quotas, ac

A structured, versioned archive of adversarial prompt engineering techniques against Large Language Models — for AI safety research, red-tea

rook by TestMu AI: Agent Assurance from the terminal. Test how your AI agents actually behave across workflows, tools, and actions, and catc

Prompt-injection guardrail for LLM applications. Compact model that outperforms larger open-source guards. No regex, no signatures. Demo: an

Benchmarking repository-borne prompt injection attacks and lightweight defenses for local coding agents. DL4C @ ICML 2026.

44 plug-and-play skills for OpenClaw — self-modifying AI agent with cron scheduling, security guardrails, persistent memory, knowledge graph

Agent Canary 3.0: open-source MCP, SDK, and CLI security toolkit with a guarded execution gateway, decoy signals, local tests, and public ac

A modular, drop-in middleware collection for Laravel AI SDK agents. Offers middleware across security, observability, performance and guidan

A honeypot and guestbook for AI agents that land on a machine they weren't invited to

The vendor-neutral protocol for AI agent safety & security — and the neutral benchmark that ranks the vendors.

Security control plane for AI agents — identity and delegation, capability policy, data-flow taint and a live audit trail, enforced over MCP

🦞 Local, read-only security audit for your OpenClaw agent. Finds config, prompt-injection and supply-chain risks, watches for drift, and ve

[ACL 2026] PIArena: A Platform for Prompt Injection Evaluation

Get Claude Code and other AI agents approved for use at work — deterministic guardrails on every tool call, tamper-evident audit logs, and a

pytest for AI agents - Autonomous red-teaming, behavioral monitoring & security testing for LLM agents

Defense-in-depth security toolkit for LLM agents — taint tracking, proxy secret guard, policy engine, and red-team benchmarking

Research artifacts for provenance-bound agent authorization, signed effect evidence, and reproducible security evaluation.

Self-hosted AI security proxy. Redact PII, block prompt injection, route to any LLM provider. OpenAI-compatible.

Typed judgment tools for MCP agents. TypeSafe's Jev model as verify, screen, find, classify, rerank, decide, compare, extract, review, gate,

Chinese-first LLM jailbreak benchmark & red-team toolkit | 中文LLM安全评测框架 | by POJIA.AI

AI Firewall & LLM security toolkit - protect your AI applications from prompt injection, jailbreaks, PII leakage, and adversarial attacks