learning-scaffold

What problem does it solve? When learning a new framework or language, AI agents tend to write complete implementations, which removes the actual learning. This Skill runs a project as a guided learning exercise where the agent scaffolds files with stubs and intent comments while you implement the logic and defend every architectural decision. ## Core Features & Use Cases - Stub-only scaffolding: Writes project layout, signatures, type hints, and plumbing files fully, but leaves function bodies as owner-tagged TODOs with raise NotImplementedError. - Socratic architecture interview: Before any plan is written, the agent grills you 2-3 rounds per load-bearing decision (storage engine, sync vs async, module boundaries) and records locked decisions as ADRs. - Learning-mode plan workflow: Layers conventions on top of brainstorming and writing-plans flows, producing plans with USER IMPLEMENTS markers, one complete test plus stubbed tests, and three execution modes. - Use Case: You want to learn FastAPI by building a real project. Trigger /learn-build, brainstorm the spec, defend your choice of Postgres and async SQLAlchemy under questioning, then receive scaffolded files you fill in yourself while the agent reviews your code. ## Quick Start Ask the agent to run /learn-build to start your project as a guided learning exercise where you write the implementation and defend the architecture decisions.