Agent Safety

DeepSeek Harness gate - per-turn persistent-operation counter: remind, instruct the model to stop, then ask for confirmation. Steps aside under sandbox/read-only modes; follows a live permission switch without restart. Verified on real hardware (full -> workspace-write -> read-only -> full). Crafted on an estimated 0.27 USD of model spend.

Operation-bound protection for autonomous AI agents: contain and verify operations before committing changes, understand intent and risk across long-running work, and detect threats across agent and system activity.

Local safety-gated MCP bridge for AI agents and Outlook mail & calendar: metadata-first reads, explicit content, reviewed sends, guarded cleanup.

Practices, protocols, and skills for AI-driven software development. Skills and safety hooks for Claude Code, Codex, OpenCode, Cursor, Antigravity, and any agent supporting the Agent Skills standard.

🛡️AutoControl Arena: Synthesizing Executable Test Environments for Frontier AI Risk Evaluation

Measuring how well CLI agents like Claude Code or Codex CLI can post-train base LLMs on a single H100 GPU in 10 hours

A coding agent hook that acts as a safety net, catching destructive git and filesystem commands before they execute. Supports Codex, Claude Code, OpenCode, Gemini CLI and Copilot CLI.

A pre-execution guard for AI coding agents. It blocks destructive Git and file system commands, plus common attempts to access sensitive files, before a tool call runs. Supports Amp Code, Antigravity CLI, Claude Code, Codex, Cursor, DeepSeek Harness, Devin CLI, GitHub Copilot CLI, Grok Build, Hermes Agent, Kimi Code, OpenClaw, OpenCode, and Pi.

Autonomous agent framework with structured memory, safety hooks, and loop management. Built by the agent that runs on it.

Static security scanner for LLM agents — prompt injection, MCP config auditing, taint analysis. 51 rules mapped to OWASP Agentic Top 10 (2026). Works with LangChain, CrewAI, AutoGen.

Open-source Discord MCP server for Claude, Codex, Cursor and other AI clients: 209 typed tools, caller-owned bots, safety controls and verifiable guild builds.

不死鸟 Phoenix — Hermes Agent 插件：路由分档/风险防线/自愈/存档点提醒/审批策略自适应，官方钩子接入，不改 Hermes 核心代码

Subagent Verification for Claude AI Code Networks 2026

Agent orchestration & security template featuring MCP tool building, agent2agent workflows, mechanistic interpretability on sleeper agents, and agent integration via CLI wrappers

GoClaw - GoClaw is OpenClaw rebuilt in Go — with multi-tenant isolation, 5-layer security, and native concurrency. Deploy AI agent teams at scale without compromising on safety.

Far Cry 3 Memory Editor Python Script 2026: Modify Health, Money, Ammo and Items

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

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

Universal AI Charter: plain-language governance principles and a public certification registry for AI models, built on disclosure instead of a development slowdown.

[NeurIPS 2025] Official repository of RiOSWorld: Benchmarking the Risk of Multimodal Computer-Use Agents

Think before doing, prove before saving. Vestige is the local-first cognitive operating system and deterministic safety kernel for autonomous AI agents. Powered by FSRS-6 causal graph memory, cryptographic Merkle-chained state ledgers, and out-of-band kernel gates bringing frontier PhD reasoning to local silicon with zero cloud token tax.

👾 下一代透明智能体架构 | Next-Gen Transparent Agent Architecture 🔍 全行为审计 | 🛡️ 两段式安全调用 | 🧠 双水位记忆 | ⏰ 心跳任务 📊 P0 级事故率降低 80% | 兼容 OpenClaw + Claude Code 技能生态

Apple's official Agent Skills exported from Xcode 27 — SwiftUI, UIKit modernization, Swift Testing, C bounds-safety, and security hardening for AI coding agents.

oh-my-zsh for Claude Code — 16 agents, 35 commands, 32 skills, 21 safety hooks in one install. v4.0 adds an adversarial review loop: a second agent that never sees the first one's reasoning. MIT.

A more beautiful and easier-to-use alternative to OpenClaw. It features a nicer Web UI, built-in IM support, and a sandboxed runtime for improved safety. Under the hood, it is powered by a Claude Code–based agent.