Local safety layer for coding agents with workspace confinement, command allowlists and auditable task evidence.
Setup
This project documents no single launch command; follow its repository instructions.
PatchWarden
jiezeng2004-design
Related
More- CCybersec Toolkit26zlInstalls 670+ security tools and lets agents run them through an authorization-gated server in a disposable sandbox VM for CTF, pentesting and DFIR.Security77 Repository stars
- BbullaRARS-ossA hermetic, no-root sandbox that emits a signed, independently-verifiable receipt that the isolation seal held — verifiable governance for AI agents (Rust).Security3 Repository stars
- FFidaajipurnLocal MCP layer for coding agents that redacts detected secrets from file reads and command output before they reach the model.Security18 Repository stars
- SSPARDAzakariagharzouliOffline gate that compiles a backend into a behavior graph and flags when an AI edit removes a guard, exposes a route or breaks an invariant.Security8 Repository stars
- 火火山引擎云审计 CloudTrail MCP火山引擎OfficialChinaCall the Volcengine CloudTrail audit service in natural language to look up audit events.Security330 Repository stars
- EEMILIA ProtocolEMILIA ProtocolRequires a named human's approval before irreversible agent actions such as payments or deploys, then issues an offline-verifiable Ed25519 receipt.Security599 Repository stars