A tool that chains AI skills to build self-improving loops in one sentence
orchflows is a library of small, reusable "skill" building blocks that can be combined into workflows, and skills can call other skills. Instead of relying on rigid, prescriptive instruction files, the design leans on flexibility, on the idea that as models get smarter, being overly prescriptive actually hurts output quality. Any request automatically gets routed to the smallest workflow needed to handle it.
A self-improving skill ships as a built-in first-class piece, meaning a single sentence can set off a loop where the system keeps improving itself. It works with both Claude and Codex, and effectively turns the -to-subagent pattern into something you program like a language.
Key points
- A library of that can call each other to build workflows
- Automatically routes requests to the smallest workflow needed
- Ships with a built-in self-improving skill, enabling one-sentence s
- Works with both Claude and Codex
- Treats the -subagent pattern like a programmable language