anthropics / single-agent-loop-kit

Why you might care

Useful if you run an orchestrator with specialist subagents and want to test whether a single loop with on-demand skills and tools performs as well. Its key feature is measurement: every change is scored on the same eval, and a subagent is removed only if its replacement holds up. It includes a runnable support-desk example and a Claude Code skill that audits your system and works through changes one at a time. The repo is unmaintained and not accepting contributions.