New Batch

---
name: new-batch
description: Lead a batch of 20 to 50 landmark models for one city or park, with builder sub-agents, ending in a draft pull request. Asks the user which city, then follows docs/briefs/lead.md.
argument-hint: "[city or park]"
---

# New batch

Recommended models: Astra, Claude Opus 5.5 or newer, or Claude Fable 5.1
or newer. If you are not one of these, tell the user so before starting.

Lead one batch of landmarks. The user may have named the city already:

> $ARGUMENTS

## 1. Ask

If no city or park is named above, ask which one. Then work out the rest and
confirm it with the user in one message before starting:

- **Prefix**: a short lowercase id prefix (`bos`, `atl`, `disneyland`).
  Reuse an existing one from `catalog.json` if the city already has models,
  and say how many it has.
- **Places and credits**: choose explicit `catalog.places` IDs (add/commit
  missing places before model merges), real contributor identities, optional
  GitHub usernames, and each AI assistant's required `model` field.
- **Branch**: `batch/<city>`.
- **Must-haves**: ask whether there are landmarks they definitely want, or
  any to leave out. Both are optional.
- **Scratch directory**: propose one outside the repo, e.g.
  `~/landmarks-scratch/<prefix>`.

## 2. Lead

Read `docs/briefs/lead.md` and follow it as your brief, with the city,
prefix, branch and scratch directory confirmed above. It points to
everything else to read, and to `docs/briefs/builder.md` for the
sub-agents you start.

Keep the user posted at each stage: the candidate list (let them edit it
before building), each model as it passes review, and the draft PR link at
the end.

Before handing off, run types, tests, validation, and the static build. Check
LOD defaults, history visibility, all model metadata, credits/avatars, and
dark/night controls against `docs/website.md`. Existing IDs remain stable for
model updates; reconcile place records manually during catalog merges.