agent-safety

Deterministic safety solutions for probabilistic AI agents

白箱AGI架构探索：元认知（自我认知循环）、持续学习（知识飞轮）、世界模型（条件空间+语义时空图）、自我改进（自举纪律）、零LLM白箱管线与可审计信任护栏。

OpenART is an open-source framework designed to evaluate the safety and robustness of autonomous AI agents in dynamic, long-horizon, and stateful environments. It stress-tests agent runtimes against multi-step state poisoning, privilege escalation, and tool-use vulnerabilities across 10,000+ benchmark scenarios.

The open standard for runtime agent control — declarative hooks, policy enforcement, and observability across AI agent frameworks.

Introducing XSafeClaw: The Open-Source Agent Safety Platform from Fudan University

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.

Safety in Embodied AI: A Survey of Risks, Attacks, and Defenses | 500+ Papers | Perception, Cognition, Planning, Interaction, Agentic System

Agent Execution Partnership AEE is an open-source control plane that ensures every AI agent action is authorized before it runs, observable while it runs, and verifiable after it completes.

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.

Ethicore Engine® is an AI safety, ethics, and compliance platform. This repo consists of the open-source components of Ethicore Engine™ - Guardian SDK; designed to protect your AI applications from prompt injection, jailbreaks, role hijacking, system-prompt extraction, and 100+ additional threat categories through a multi-layer analysis pipeline

An open taxonomy and scoring framework for evaluating AI agent sandboxes: 7 defense layers, 7 threat categories, 3 evaluation dimensions, 27 "sandboxes" scored.

🛡️ A curated list of tools, frameworks, standards, and resources for AI agent governance, safety, and compliance

Trust nothing. Ship safely. — Skeptical-reading and prompt-injection defense skill for AI agents. Provenance tagging, red-flag patterns, refusal templates, and a read-only injection auditor. MIT.

Undo for AI agent shell commands. Snapshots files before Claude Code runs destructive bash (rm -rf, git reset, rsync), so agent mistakes are reversible, even for files git never tracked.

Open-source runtime for grounded, model-agnostic computer-use agents with verifiable action receipts.

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

Human-in-the-loop execution for LLM agents

The agent's prompt shows you the command; Termaxa shows you the consequence. Previews, backups, policy, an audit record. Claude Code, Codex, Cursor, Copilot. A windshield, not a sandbox.

OpenClaw-compatible MASL safety gate with public RAG packs for memory-aware AI agents