S

sandbox

Stratum is an AI governance layer for MCP systems that sits between model intent and tool execution. It enforces permissions, policies, context controls, approval workflows, sandboxed execution, and audit logging to ensure every action taken by an AI system is authorized, traceable, and constrained by explicit rules before it runs. https://roxanneardary.com/stratum/

AI development environment management tools - multi-org credential isolation, sandboxed environments, VPN support

Minimal Cloudflare Worker using the @cloudflare/sandbox SDK to run untrusted code in isolated containers backed by Durable Objects - execute commands, read and write files, and expose services straight from the edge. Ships a Dockerfile and wrangler config and is deployed live.

Demonstrating what you can achieve without mandatory usage of any kind custom of server side logic. You could use these on a CDN, shared web hosting without SSI or .htaccess or source code repository based hosting. Rendered: https://bkil.gitlab.io/static-wonders.js/ Some compliant with JavaScript0: https://bkil.gitlab.io/gemiweb0 Mirrored: https://github.com/bkil/static-wonders.js

Securekit is a protocol-agnostic security kernel that enforces zero-trust, sandboxed execution for AI tool use. It sits between any LLM or agent system and its tools, validating, isolating, and auditing every action to prevent unsafe execution across MCP, OpenAI tools, and custom AI protocols. https://roxanneardary.com/securekit/

Practice lab for learning Git and GitHub through commits, branches, issues, pull requests and clean professional workflow habits.

SecurePath is an open-source, enterprise-grade framework for AI evaluation, red-teaming, and regulatory compliance. It provides sandboxed testing, policy-aligned metrics, multi-modal evaluation, training data sensitivity analysis, and historical compliance tracking, all while maintaining audit-ready logs and dashboards. Designed for organizations, enterprises, and research teams, SecurePath ensures AI safety, transparency, and regulatory assurance. https://roxanneardary.com/securepath/

Isolate your big brother apps https://secure-system.gitlab.io/Insular/

Local claude session sandbox: kind cluster plus per-session pods running the published config-baked dev-sandbox image from konradodwrot/infra/oci-images.

This open-source plugin extends Brightlayer, the AI-native, stateless, privacy-first browser, with AI-powered features while respecting its sandboxed, end-to-end encrypted architecture. It enables local AI inference for tasks like summarization, translation, and privacy monitoring, all without storing or transmitting user data. Fully open-source and licensed under AGPL 3.0+, the plugin ensures transparency, reproducible builds, and strict adherence to Brightlayer’s stateless, privacy-first philosophy. https://roxanneardary.com/brightlayerplugin/

Brightlayer is an open-source, AI-native, stateless browser built to eliminate tracking, enforce end-to-end encryption, and give users full control over their data. Designed with a privacy-first architecture and an open-source-only extension system, it combines local AI intelligence, sandboxed execution, and reproducible builds to deliver a secure, transparent, and high-performance browsing experience. https://roxanneardary.com/brightlayerbrowser/

https://thibaultlemaire.gitlab.io/triambricate

Emnoxx's Mod! [ WIP Garry's Mod-like sandbox ]

Mini 2D sandbox-style game built with Python and Pyxel, featuring tile-based rendering, simple AI systems, camera mechanics, and design patterns like Flyweight, Factory, and Observer.

A place for trainings & some scripts worth keeping.

Started originally with Open Classrooms courses

A security sandbox for AI code assistants

Compiled .NET 8 PowerShell deobfuscation sandbox

The judging server of YACPS - Yet Another Competitive Programming System

Docker-based sandbox for running OpenCode with optional MCP services, including a Python CLI wrapper, persistent config/data volumes, and CI checks for tests and container startup.

The Triangle of Art is a hardened Docker-based sandbox designed for summoning and containing LLM coding agents (Aider, Claude, Qwen, Gemini, and Codex). It provides a restricted environment where these entities can execute code, run tests, and manipulate files without escaping their tether or accessing the Magician's host filesystem.