The situation
and the fourth project is set up slightly differently from the first for no reason anybody chose.
1. Capture what you actually do
2. Start the product
create_product with a name and the five "why" fields. Those are not ceremony - they are what get_product_context hands an agent later, and an agent that knows why the product exists writes different code from one that does not.
3. Apply the profile
apply_setup_profile returns everything the environment needs, resolved at the versions the profile pins. Nothing is invented here: it owns no library logic and reads through the services that do, so what you get is exactly what those tools would have returned one call at a time.
4. Seed the work
create_tasks takes a batch, so the first sprint is one call rather than fifteen. Give an agent the outline and let it write them:
Read get_product_contextfor this product. Propose the first two weeks as tasks, grouped by epic, each with what "done" looks like. Do not create anything until I say go.