create-project

name	create-project
description	Generate a ready-to-import Linkr project ZIP — a project with synthetic dataset(s), a wired dashboard (KPIs + charts + filters), IDE analysis scripts (.py/.r/.sql/.md), and README/tasks, assembled into the importable project ZIP format. Use when the user wants a demo project, sample data, a dashboard, or analysis scripts built from scratch to drag into the app.

Create an importable Linkr project

You build a project ZIP the user imports via "Import a project" in the app. You never touch app source code — the output is a data artifact.

All mechanical ZIP assembly (col-N ids, tree/parent ids, 48-col grid layout, appVersion, IDE scripts/ layout, .gitignore) is owned by assets/build_zip.py. Your job is to author CSV(s), optional IDE script files, and one spec.json, then run the script.

What a project ZIP can contain

Brick	In ZIP?	This skill	Reference
Project metadata (name, description, badges)	✅	✅	references/project-meta.md
README + tasks (todos/notes)	✅	✅	references/project-meta.md
Datasets (CSV → table)	✅	✅	references/datasets.md
Dashboards (8 built-in widgets: KPI, plot, table1, correlation, stat-tests, regression, Kaplan-Meier, map, sankey + inline code + filters)	✅	✅	references/dashboards.md
IDE files (.py / .r / .sql / .md in Lab › IDE)	✅	✅	references/ide-files.md
Cohorts, DB connections, pipelines	✅	❌ (out of scope)	—
Patient Data page	❌	❌ impossible	see below

Patient Data is NOT seedable. It is computed live by SQL against a connected OMOP database (features/…/patient-data/use-patient-data.ts); there is no patient-data payload in the ZIP. The only way to populate it is a real OMOP database + connection + schema mapping — outside the scope of a synthetic-data project. If the user asks for it, say so and offer an OMOP-shaped dataset instead.

Workflow

1. Gather requirements — theme/clinical domain; what indicators/charts; whether they want IDE analysis scripts; row count (default ~200); languages (name/description are {en, fr} — fill both).
2. Author the CSV(s) — one per dataset, clinically coherent. → references/datasets.md
3. Author IDE files (if requested) — real .py/.r/.sql/.md. → references/ide-files.md
4. Author spec.json — read the reference for each brick you use. Start from examples/spec.json (a complete, working example).
5. Build: python3 .claude/skills/create-project/assets/build_zip.py spec.json <slug>.zip
6. Verify & hand off — see below.

spec.json at a glance

{
  "appVersion": "2.0.20",
  "project":   { … },        // references/project-meta.md
  "ide":       [ … ],        // references/ide-files.md  (optional)
  "datasets":  [ … ],        // references/datasets.md
  "dashboards":[ … ]         // references/dashboards.md
}

The full field reference is also in the docstring at the top of assets/build_zip.py. examples/spec.json + examples/*.csv + examples/scripts/ are a runnable set — copy and adapt rather than starting blank.

Step 6 — Verify and hand off

• unzip -l <out>.zip should show project.json, datasets/<name>/_data.json, dashboards/<name>.json, and (if you added IDE files) scripts/_tree.json + each scripts/<path>. Confirm every folder/file name is ASCII (no accents) — a non-ASCII dataset folder imports directly but is silently dropped after a macOS/git round-trip (see Format invariants).
• Validate JSON: python3 -c "import json,zipfile;z=zipfile.ZipFile('out.zip');[json.loads(z.read(n)) for n in z.namelist() if n.endswith('.json')];print('ok')"
• Tell the user the ZIP path and that they import it via the project import flow. Every id is regenerated on import, so re-importing makes a fresh copy.

Format invariants (do not break)

These mirror apps/web/src/lib/entity-io.ts (buildProjectZip / parseProjectZip) and seed-loader.ts. If that file's layout changes, update build_zip.py to match. The per-brick reference files document each invariant next to the brick it governs.

• Dataset names and IDE paths must be ASCII (no accents/spaces) — they become ZIP folder paths and non-ASCII silently breaks import after a macOS/git round-trip (NFC vs NFD + exact-string lookup in parseNewLayout). build_zip.py enforces this.