Zeroclawed: Secure Agent Gateway
Credential proxy plus Starlark policy engine for AI agent tool calls.

Firewall proxy for devcontainers, letting Claude Code use OAuth and Chrome safely.
Developers using Claude Code or similar AI coding agents in teams
Continue · Cursor · Cline
With 45 years of development experience, what can I change?
I wanted to create all my ideas, and for that to go full --yolo in a secure way, so I created human as a firewall for devcontainers that keeps MCP and OAuth flows working. And devcontainers that are easy to setup and just work. You can get some of it with other tools, but it's a pain to install them all and make them work individually.
This evolved into all I needed for development: Context via Notion, Sentry and Linear/JIRA integration, a control plane that shows running Claude instances, assigning tickets to Claude instances, all the prompts you need for development tuned to use human as a tool and much more.
https://gethuman.sh - Open Source, MIT, Free.
This is in it's early stages, works for my use cases, gas town light.
Credential proxy plus Starlark policy engine for AI agent tool calls.
Credential proxying keeps keys out of sandboxes, unlike Ramp Inspect.
Agent credential proxy using OS keychain injection—credentials never touch agent memory.
Agents never see credentials — brokered access beats retrieval for prompt injection safety.
Transparent proxy blocks agent leaks to LLM APIs without touching agent code.
Scoped runtime credentials for AI agents replace insecure .env API keys.