GitHubAI Skills, Agents & MCP · Agent safety
Self-hosted Go sandbox service for agent-written code that reports the isolation tier of every run, refuses runs below a floor and brokers API calls without exposing credentials.
Why it’s a gem
Makes sandbox strength a checkable value on every result, not a datasheet promise.
Open-source runtime layer for AI coding agents that intercepts shell, secrets, MCP, and installs locally.
CLI that runs coding agents inside hardware-isolated virtual machines for safe autonomous operation.
Real-time guardrails that detect and block secret leakage in agent output and tool calls.
Hook that intercepts destructive shell commands before an autonomous agent executes them.
Source-linked archive of documented incidents where AI coding agents and chatbots deleted data, leaked secrets or burned money, each with the lesson.
Open-source runtime guardrails that allow, warn, or block every LLM and MCP tool call before it commits, with a hash-chained audit trail.