| name | vertical-slice |
| description | Break a complex build into thin, end-to-end slices — each independently runnable and verifiable — instead of horizontal layers, so you always have something working and learn early. Part of the "keep the agent on track" suite. Use when a task is big enough that "build it all, then test" risks weeks of invisible drift. |
Vertical slices
Why
A thin slice that works end-to-end beats three half-built layers. Slicing vertically gives you something runnable from day one, surfaces integration problems early, and de-risks the whole build.
How
- Find the smallest path that exercises the whole system — input → process → output.
- Ship slice 1 working, end to end, even if trivially narrow.
- Add slices that broaden or deepen, each with its own definition of done.
Example
Building "pull from an API, store it, report on it"? Don't build all ingestion, then all storage, then all reporting. Build one record: API → SQLite → one-line report first — then widen.
Anti-patterns
- Horizontal layering (all of X, then all of Y) that only integrates at the very end.
- "Slices" that aren't actually end-to-end (a layer wearing a slice's name).
- A first slice too big to finish in one sitting — thin it further.
