Agent loops, release automation, dry-run simulations, multi-model harnesses, skills vs MCPs, and enterprise permissions for agentic workflows. Rss