gogcli sandbox

Warning

This project is very early in development and not fully tested. Use at your own risk.

A small Go broker that exposes a strict, policy-enforced API to gogcli for AI agents. The broker is meant to run as a separate OS user and is the only interface available to the agent.

The idea is to remove direct access to gog from the agent, and provide a layer of protection against hallucinations and prompt injections. Running the broker as a separate user (or on a separate machine) prevents agents from accessing OAuth credentials.

Docs

• Installation & setup: docs/install.md

Key behaviors

• The broker refuses to start if the policy file is missing or invalid.
• Denied actions return ok: false with a structured error.
• All responses are redacted according to policy.
• Policies are defined per account; the client can pass --account.
• gmail.send can be forced into draft-only mode, with allowlisted recipients.
• Gmail label filtering happens after the query to avoid false negatives.

Repo layout

• cmd/broker: broker entry point
• internal/policy: policy parsing and validation
• internal/gog: gogcli command runner
• internal/redact: response filtering and redaction
• internal/server: Unix socket HTTP server
• deploy/systemd: example systemd unit

Testing

go test ./...

Disclaimer

⚠️ THIS SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND.

This tool is meant to be a security mitigation only, and not a guarantee of safety.

• Do not rely solely on this tool for security decisions in production environments.
• The authors are not responsible for any damage caused.
• Use at your own risk.