eugene1geugene1gCommunityยท1 Agent Skills Included

agent-safehouse

Sandbox AI coding agents on macOS with least-privilege file access

Restricts AI coding agents like Claude, Cursor, and Aider to only the files and integrations they actually need on macOS. Blocks unauthorized access to secrets, home folders, and system resources using deny-by-default sandbox policies. Includes ready-made profiles for major agents and apps, plus a release skill that automates changelogs, versioning, and publishing.
npx skills add eugene1g/agent-safehouse --all -g -y
Available:

Tells the AI agent how the sandbox policy layers are assembled, which commands to run for testing and packaging, and the rules to follow when editing sandbox profiles in this repository.

All Skills in This Repository (1)

Pure Emerald Level Indicators

Frequently Asked Questions

FAQPage Schema
How to install Agent Safehouse?โ–ผ

Run `npx skills add eugene1g/agent-safehouse --all -g -y` in your terminal to install all skills in this suite globally.

How to sandbox AI coding agents on macOS?โ–ผ

Wrap your agent launch command with the safehouse script, which runs it under macOS sandbox-exec with a deny-by-default policy so it can only reach approved files and tools.

Which AI agents does Agent Safehouse support?โ–ผ

It ships with built-in profiles for Claude Code, Cursor, Aider, Gemini, Codex, Copilot, Goose, and other popular coding agents and app-hosted workflows.

Can agents still read my secrets or home folder?โ–ผ

No. The default policy only allows minimal home-directory metadata access, and files like ~/.ssh keys or personal documents stay blocked unless you explicitly grant a path.

Do I need to write sandbox policies myself?โ–ผ

No. Safehouse includes composable pre-built profiles, and you only add simple flags like --add-dirs-ro for extra folders when needed.

Related Repositories in Software Engineering

View All in Software Engineeringโ†’