---
status: accepted
---

# Agent capabilities split by native seam

omp-spt selects a parity seam per capability rather than forcing every agent-facing outcome through one mechanism. Public OMP `omp-plugins` skills own model-driven commune (including checkpoint mode), signoff, durable-role, and setup-assistance workflows, with packaged-skill discovery proven from the installed release; extension-native commands own ready/live activation, while the extension retains binding, listener, delivery custody, activity, recovery, reply, and shutdown. A universal skill layer was rejected because lifecycle infrastructure must not depend on model compliance, while a universal extension-command layer was rejected because authoring continuity and repair guidance genuinely require agent reasoning.