Phase 1: Spec-Driven Development · 45 min · Claude Code · Python

From Spec to Plan to Tasks

Build the thinnest path through the whole system first. Save the hard part for last and you find out it's unworkable after everything else is already built.

Hiring signal: Decomposing a spec into a walking skeleton plus a hardest-first task order is a concrete planning skill interviewers ask candidates to demonstrate live — it shows you can turn a spec into something an agent can be safely pointed at incrementally.

What you will learn

Explain the walking skeleton pattern and why it should be the first task in any execution plan
Apply hardest-first ordering to a task breakdown and justify why it beats easiest-first
Decompose a real spec's acceptance criteria into an ordered, executable task list
Produce a full spec and task breakdown for your own course-long product

Introduction

From Spec to Plan to Tasks

A team has a solid spec for their discount-stacking feature — the acceptance criteria from earlier in this phase, capped, sequential, tested. They point an agent at the whole spec at once and let it generate the entire feature: database schema, backend calculation logic, third-party coupon-validation integration, and the frontend UI, all in a single enormous pull request. It runs. It also touches nine files, mixes genuinely novel business logic with routine UI scaffolding, and takes a reviewer forty minutes to even understand the shape of the change before they can start checking whether it's correct. Three days later, the third-party coupon-validation API turns out to have a rate limit nobody accounted for, which means the core calculation logic — already reviewed, merged, and built on top of by two other people — needs to change. The spec was good. The plan to build from it was nonexistent, and that's a different, separable failure.

A spec tells you what must be true. It doesn't tell you what order to build things in, how big each piece of generated work should be, or which part of the feature to derisk first. That's the job of a plan, and this lesson covers the two ideas that make plans good instead of just present: the walking skeleton and hardest-first ordering.

The walking skeleton: prove the path before filling it in

A walking skeleton is the thinnest possible version of the feature that exercises every layer end to end — frontend input, backend logic, database write, and back — with minimal or stubbed logic at each step, deployed and running before any real feature work begins. For the discount feature, that's: a UI field for one discount code, a backend that applies a single fixed percentage (no stacking, no cap logic yet), and a database write that persists the result — genuinely useless as a feature, completely real as a proof that every integration point actually connects.

The value isn't the skeleton itself. It's what building it first tells you before you've invested in anything more elaborate: does the frontend actually reach the backend the way you assumed? Does the database schema you sketched actually accept the data shape the backend produces? Is there an auth or deployment wrinkle nobody noticed until something tried to run end to end? Discovering an integration problem while the "feature" is three trivial lines per layer is cheap. Discovering the same problem after building the full stacking logic, the coupon-validation integration, and a polished UI on top of a broken connection is expensive, and it's expensive in the specific, painful way of having to unwind work that depended on an assumption that turned out to be wrong.

The skeleton must be worthless as a feature and priceless as a proof
If your "walking skeleton" already includes real business logic, it's not a skeleton — it's just the first feature task with an intimidating name. A real skeleton does the least possible at every layer, on purpose, so that what it proves is the connections between layers, not any particular layer's correctness.

Unlock the full lesson

You've read the first 2 sections. The rest of this lesson covers Hardest-first, not easiest-first, Tying the plan back to the spec, Build It — plus a hands-on lab, quiz, and project artifact.

Create a free account to unlock Phase 0 and Phase 1 of every course — no credit card.