← Day 04: Assemble DAY 04 · BRIEFING AND GATE

Does one slice work, is it tested, and is it safe?

Today you build one vertical slice with Claude Code, in small loops, and prove it works and refuses the wrong people.

Plan modeOne vertical sliceTestsRead the diff/rewindAccess testspip-audit
Present slides Full screen, one slide at a time · 7 slides
TODAY 01

What you will leave with today

  • A walking-skeleton plan approved in plan mode
  • One vertical slice that runs end to end
  • Tests for the happy path, bad input and “not allowed”, all passing
  • Logged-out and cross-user access tests, and a pip-audit scan, with findings dealt with

OUTPUT · A working slice committed + product/04-build.md

RECAP · DAY 1 02

The build loop, one slice at a time

  1. PLANPlan mode: Claude proposes, you approve; no files change before that
  2. BUILDOne small step of the slice, not the whole product
  3. VERIFYRun the tests and click through it; compare Claude’s report with the real output
  4. REVIEWRead the diff with the five questions; /rewind anything you don’t want
  5. COMMITA named save point, after every step that works

Go round the loop several times today. Commit every time it works.

WORKED EXAMPLE · TASKBRIDGE 03

TaskBridge: the test that matters most

Starting point

“It works: I logged in, logged a job and saw it on the list.”

Reframed

Logged out, the list redirects to sign-in. A lead from Team B opening a Team A job’s address gets “not found”, and cannot accept it. Both are tests that run every time. Illustrative.

Ask

  • What does the weak version prove, and what does it not?
  • Why test the address directly, not just the list?
  • Why “not found” rather than “not allowed”?