@kirkelabs/open-agent-access-guard

Human approval guardrails for agent-controlled repositories, deployments, and production-facing actions.

Agents may read, inspect, summarize, pull fast-forward updates, generate reports, and prepare changes. They may not push, deploy, publish, mutate environment variables, change domains, alter payment configuration, or deploy contracts without bounded human approval and an auditable approval record.

Core Flow

diff packet -> human review -> one-time approval token -> guard -> ledger record

Important Boundary

Local hooks are guardrails. Enterprise enforcement requires remote branch protection, required checks, deployment protections, and protected secrets.