Anthropic publishes a kit for folding subagents into one agent loop

TL;DR

• On 22 October Anthropic published guidance on when to fold subagents into one agent loop that loads skills and tools on demand, with a migration kit on GitHub under Apache 2.0.
• Across several enterprise deployments, a single agent with skills frequently beat both the one-prompt design and the subagent design on quality, often at lower cost and latency, Anthropic says.
• The kit measures rather than assumes: a Claude Code skill audits your subagents, builds a baseline eval, and folds one subagent at a time while you decide whether each change stays.