Setup

---
name: setup
description: The setup interview that customises this template for one founder's company by region, domain, audience, customer size, stack, gates, team and budget, and writes project.toml. Use at the start of a new company or when any of those change.
user_only: true
---

# Setup

Run `skills/grilling/` over the tree below, in rounds, each question with a recommended answer.
Facts you can look up (the founder's git identity, an existing codebase's stack, a region's
data protection law) you look up. Write nothing until the founder confirms the summary.

## The tree

Ask as few rounds as the dependencies allow; a non-technical founder should not face 25
questions. Combine branches whose prerequisites are settled, and defer anything not needed
before the first feature to `p1-open-questions.md`.

1. **Founder:** name and email for commits; technical or not (sets how much agents explain).
2. **Product:** name, one sentence, stage (idea, prototype, beta, live).
3. **Market** (unlocked by 2):
   - regions (ISO codes, where users live, not where the founder lives);
   - domain;
   - audience in a sentence;
   - customer size: consumers, small business, mid-market, enterprise.
4. **Data** (unlocked by 2, and asked with 3 in one round): what personal data, payments, health, children, biometrics, AI
   decisions about people. Each one becomes a `sensitive` entry and switches on gates.
5. **Stack** (unlocked by 2 and 4): platform, frontend, backend, database, hosting, the test and
   build commands ("none yet" when no code exists). Recommend what the founder already knows,
   hosted, boring. If a codebase
   exists, read it and propose what it already is. Run `skills/spec-audit/` on it, and put
   its most dangerous silent fills into the next round.
6. **Gates:** recommend **strict** and say what it buys, using the table in `skills/gates/`,
   never your own description of a level. If the founder picks lighter, ask for
   the reason and record it.
7. **Team** (unlocked by 2, 4 and 6): the smallest set of agents the stage needs. Recommend, say
   what each one does and when it is first needed, let the founder trim. Time per feature is not
   measured yet, so do not quote one.
8. **Budget:** monthly AI spend ceiling for the optimizer to report against.
9. **Tools:** which AI coding tools the founder uses (adapters are generated for each). Always
   include the tool running this setup, or its generated files are removed by the sync.

## Write

- `project.toml`, every field the answers settled.
- `docs/legal/regions/<CODE>.md` for each region, copied from `docs/legal/regions/_TEMPLATE.md`,
  filled from current sources, each claim cited and dated with a short quote, unsettled points
  marked UNSETTLED, and `python tools/link_check.py` run on the packs with no DEAD links.
  Hand the packs to `legal` to review before any launch.
- `docs/requirements/p1-closed-decisions.md`: one dated line per decision taken, with the reason.
- `docs/requirements/p1-open-questions.md`: what the founder deferred, and what each blocks.
- Requirement rows in `docs/requirements/LAUNCH.md` (prefix `LAUNCH`), marked `X`, for what
  customer size and domain imply (an enterprise customer
  implies SSO and a DPA; consumers in the EU imply cookie consent and data export).
- If the product has a free tier or a trial: requirement rows marked `X` for its caps (per user,
  per device without sign-up, a global daily spend ceiling), with the numbers left for the
  founder to set once the prototype has measured the cost per action. A free tier with no caps
  is not offered as an option.
- `git config --local user.name` and `user.email` set to the founder (this repository only),
  and
  `git config core.hooksPath tools/githooks`.

## Finish

Run `python tools/sync_adapters.py`, `python tools/checks/req_check.py --fix` (the new
`LAUNCH.md` changes the index) and `python tools/run_all.py`. Show the founder the summary,
the diff and the green sweep. Commit on their go.