portage-ucp-decision

https://github.com/tomtom87/Portage

The judgment calls between the agent loop and the Adapter/client layer, as typed, inspectable decisions instead of control flow scattered across skill instructions and the CLI: which offer to pick (OfferRanking), hand off vs. keep going (EscalationPolicy), whether a result is confident enough to act on unattended (ConfidenceGate), and a typed wrapper around Portage::Ucp::PolicyGuard (PolicyCheck). See docs/plans/system-one-decision-layer.md.