sync-demo-seeder - SKILL.md Agent Skill

name: sync-demo-seeder description: Synchronizes scripts/seed-demo.sql with the current Drizzle schema and migrations. Use this skill whenever a new migration is added or src/lib/db/schema.ts changes, to keep the demo dataset consistent with the database structure.

Synchronize scripts/seed-demo.sql with the current database schema after a Drizzle migration or schema change.

Your Task

When invoked, perform the following steps in order:

Step 1 — Understand the current state

Read all of the following files completely before making any changes:

src/lib/db/schema.ts — source of truth for all tables, columns, and enum values
Every SQL file in drizzle/ (sorted by filename, i.e. migration order)
scripts/seed-demo.sql — the current demo seed that needs updating

Step 2 — Diff the schema against the seed

Compare the schema to what the seed currently does. Identify every discrepancy:

Tables:

Is every table that holds user data present in the TRUNCATE statement?
- Excluded tables (never seed): conversations, messages, documents
- The TRUNCATE must list all remaining user-data tables
Is there a new table that needs demo rows?
Is a table removed? Remove its TRUNCATE entry and INSERT block.

Columns:

Are the column names in each INSERT statement still correct?
- Column names come from the second argument of each Drizzle column definition, e.g. uuid('account_id') → SQL column is account_id
- Never guess — read the schema directly
Are there new NOT NULL columns without a DEFAULT that are missing from the INSERTs? They must be added with a sensible demo value.
Are there removed or renamed columns still referenced in the seed? Remove or rename them.

Enum values:

Are any enum values used in the seed that no longer exist in the schema?
Are there new enum variants that existing demo rows should use?

Step 3 — Plan the changes

Before editing, write out a concise list of every change you are about to make, grouped by table. For example:

accounts       — no changes
expenses       — add new column `budget_id` (nullable, use NULL in all rows)
categories     — enum 'shopping' renamed to 'retail', update 3 rows
budget_groups  — NEW TABLE: add TRUNCATE entry + INSERT block with 2 demo rows

If no changes are required, say so explicitly and stop.

Step 4 — Edit scripts/seed-demo.sql

Apply every planned change to scripts/seed-demo.sql. Follow these rules:

UUID format: All hardcoded UUIDs must use valid hex characters only (0-9, a-f). Use the existing naming convention:

a0000000-0000-0000-0000-00000000000N for accounts
b0000000-0000-0000-0000-00000000000N for incomes
c0000000-0000-0000-0000-00000000000N for categories
d0000000-0000-0000-0000-00000000000N for transfers
e0000000-0000-0000-0000-00000000000N for expenses
For new tables, pick the next unused letter (f, then 0a, etc.)
For tables seeded with gen_random_uuid() (daily_expenses), keep using it

Demo data quality: New demo rows must be realistic and consistent with the existing dataset (German locale, EUR currency, plausible financial amounts and descriptions). Never insert placeholder values like "Test" or "TODO".

Transaction: The entire file must remain wrapped in BEGIN; ... COMMIT;.

TRUNCATE: Keep all truncated tables in a single TRUNCATE TABLE ... CASCADE; statement at the top.

Column order: Match the column order in each INSERT to what's defined in the schema for readability.

Comments: Keep existing section comments (-- === ACCOUNTS === etc.). Add a matching comment for any new table section.

Step 5 — Verify

After editing, re-read the updated scripts/seed-demo.sql and confirm:

Every table referenced in INSERTs exists in the current schema
Every column in every INSERT matches the schema's SQL column name exactly
All hardcoded UUIDs contain only valid hex characters
No enum value is used that isn't declared in the schema
The file starts with BEGIN; and ends with COMMIT;
The TRUNCATE covers all user-data tables

Report the final checklist result to the user.

Context

ORM: Drizzle ORM — the TypeScript property name and the SQL column name often differ. Always use the SQL column name (second argument of the column function). Example: accountId: uuid('account_id') → INSERT uses account_id, not accountId.
Migrations live in drizzle/ — each .sql file represents one schema version.
The seed resets all user data daily via scripts/demo-reset.sh in the demo Docker stack.
Do not touch conversations, messages, documents — these are excluded from seeding by design (AI chat history is ephemeral, documents are large binary blobs).
Do not change the reset logic in scripts/demo-reset.sh unless explicitly asked.