Skip to content

Plan the change

Goal: an explicit plan for making spent honest, before touching code. This is the spine of the whole workshop: the plan is the difference between a change that holds and one that silently breaks.

You’ll build the plan two different ways and compare what each surfaces.

Run it against the same goal, but specific enough that Kiro can find what to fix — name the symptom, not the fix:

/grill-me-with-docs the spending total in /summary over-counts — make it report the honest number

It reads the code and the project docs you generated with /code summary, then interrogates you, often 10 to 20 questions, until it’s sure it understands, and writes the answers into a decision doc. Naming the symptom only points it at the right code — it still forces the decisions the prompt left open:

  • Does a transfer between your own accounts count as spending? (no)
  • Does a refund reduce spending, or get ignored? (reduce)
  • Do pending charges count before they clear? (no)

That’s the whole point: the bug was born from unanswered questions. This drags them into the light before a single line changes.

However you got here, capture the decisions and the touch points:

  1. Decisions: transfers excluded, refunds netted, pending excluded, for an honest total of $850.
  2. Touch points (this is what vibing misses): the transactions schema + seed (add a kind so a transfer is distinguishable), the /summary query, the /transactions output, and the tests.