Prompt library

Build [a returns tracker for a small online seller].

The one job: [a customer raises a return with a reason, I approve or reject it,
and I can see what's pending pickup versus refunded].

Data has to survive a refresh, and there are accounts — this is not a demo.
Seed it with realistic sample rows, not lorem ipsum.
Vibe: [calm, dense, muted — it's an ops tool, not a marketing site].

Build [a habit tracker] as a mobile app.

The one job: [see today's habits, tick one off, and open a habit to see its streak].
Bottom tabs for [Today / Stats / Settings], plus a modal sheet for adding a habit.
Portrait only. Data survives an app restart.
Vibe: [big tap targets, soft shadows, rounded corners].

Add accounts. Email and password sign-in, plus a sign-up flow that handles
email confirmation. Three roles: admin, editor, viewer.

Roles live in their own table, checked in RLS policies — not a column on profiles.
Admins can see everything; editors can only edit their own rows; viewers are read-only.
The logged-out and logged-in chrome should actually differ.

Add accounts to the app. Email and password sign-in and sign-up screens in an
(auth) route group, one useSession hook, and an auth gate on the tabs layout.
Handle both sign-up outcomes — instant session, and "check your email".
Email links have to come back into the app through the deep link scheme.

Add a dashboard as the landing screen for signed-in users.

Four KPI tiles across the top: [total outstanding, collected this month,
overdue count, average days to pay]. Below them, [a monthly trend chart] and
[a table of the ten oldest unpaid invoices] with the overdue ones first.

Every number comes from the real query, not a constant. Empty and loading
states for each tile.

Add a stats screen as a tab. Three summary cards, then a 30-day chart below them.
Pull-to-refresh. Render all four states — loading, error, empty, populated.
Use the shared formatting helpers for every date and number.

Let users attach [a bill photo] to each [expense claim].

Private storage — a user can only read their own files. Show upload progress,
accept images and PDF up to 10 MB, reject anything else with an inline message,
and render a thumbnail once the upload finishes.

Let users attach a photo to each entry — camera or photo library, their choice.
Ask for the permission at the moment they tap Attach, not on launch.
Upload to private storage, show progress, and render the thumbnail when it lands.

Add CSV import and export to [the transactions list].

Import: file picker, parse in the browser, show me a preview table with the
column mapping before anything is written, flag rows that would fail validation,
and let me import only the valid ones.

Export: current filters applied, real column headers, downloads as a file.

Add CSV export to the entries screen — generate the file, then open the system
share sheet so I can mail it to myself or drop it in Files.
Import is out of scope on mobile; keep it on the web app.

Add subscriptions. Three tiers — [Free, Pro at $19/mo, Team at $49/mo] — with
[the feature matrix below]. Stripe Checkout for the upgrade, a webhook that keeps
the subscription row in sync, and a billing page where someone can see their
plan and open the customer portal.

Gate [the export and the API key sections] behind Pro.

Add a paywall for [unlimited habits]. Present it from the blocked action — the
moment they try to add the sixth habit — not on launch. Also add an upgrade row
in Settings and one at the end of onboarding.

Monthly and annual cards, annual pre-selected, savings badge computed from the
two prices. Restore Purchases button, Terms and Privacy links.

Add an in-app notification centre. A bell in the header with an unread count,
a panel listing recent notifications, mark-one-read and mark-all-read, and a
row in the database per notification so it survives a refresh.

Also send an email when [an invoice goes 30 days overdue] — from an Edge
Function, with the provider key kept server-side.

Add push notifications. Ask permission the first time they turn on reminders in
Settings — never on launch. Store the device token against the user row.
Local scheduled reminders at a user-chosen time, plus server pushes for
[when someone comments on your entry].

Make this multi-tenant. Every row belongs to a team, users belong to one or more
teams, and someone only ever sees their own team's data — enforced in RLS, not
in the UI.

Add an invite flow: an owner invites by email, the invitee accepts and lands in
the team. A team switcher in the header if someone is in more than one.

Add teams. A team switcher in the Settings screen, team-scoped data everywhere,
and an accept-invite screen reachable from the invite deep link.

Add an AI assistant to [the notes screen]. It reads the current [note] and can
[summarise it, suggest tags, and rewrite it in a chosen tone].

Streaming response, a stop button, and a visible error state when the call fails.
Use Vibely AI so there's no API key to manage.

Add the assistant as a sheet on the entry screen. Stream the response, keep the
keyboard out of the way, and show a retry on failure.

Add semantic search over [documents]. Enable pgvector, store an embedding per
[document], generate it on insert and update, and add a search box that returns
the ten closest matches with a relevance score.

Embedding generation goes in an Edge Function with the key server-side.
Keep the existing keyword filter — I want both.

Add the same search to the mobile app — one search field above the list,
debounced, with a loading row and a designed "no matches" state.

Go through every public page and add SEO: a unique title and meta description in
each route's head, one h1 per page, Open Graph and Twitter card tags, JSON-LD on
[the product pages], a sitemap, and alt text on every image.

Skip the signed-in app routes — those should not be indexed.

A native app has no meta tags — the store listing is its metadata. Write the App
Store and Play listing copy for this app: name (30 chars), subtitle (30),
promotional text, full description, and a keyword list. Base it on what the app
actually does, not on what it could do.

Add an audit log. Every create, update and delete on [invoices, payments and
team membership] writes a row: who, what, which record, before and after, when.

Only admins can read it. Nobody can update or delete a row — inserts only.
Add an admin screen with filters for actor, action and date range.

Surface the audit log read-only in the mobile app for admins — a list with
filters in a sheet, paginated, no write actions at all.

Add feature flags. A flags table with a key, a description and an enabled boolean,
a hook that reads them once and caches, and per-team overrides so I can turn
[the new dashboard] on for one team before everyone.

Admin screen to flip them. Default to off for anything unset.

Same flags on mobile, but cache the last fetched values so the app doesn't render
the wrong UI on a cold start with no network. Fall back to off, never to on.

Add [a due_date and a priority enum] to [tasks].

Write it as a new additive migration — don't edit the applied one. Existing rows
have to stay valid, so give me a sensible default or make it nullable. Update the
types, the list, the create form and the edit form in the same turn.

Same migration, and update the entry sheet, the detail screen and the list row.
Anything I've already saved on-device has to keep loading — handle the missing
field on old records rather than crashing on undefined.

The [dashboard] feels generic. Make it [denser and more like Linear] —
[mono-spaced tabular numbers, tighter row height, one accent hue instead of four,
borders doing the separating instead of shadows].

Change the tokens, not the individual components. Keep the rest of the app
consistent with whatever you change.

Restyle the app to feel [premium and quiet]: [a deeper neutral base, one accent,
tighter type, less shadow]. Change the theme constants for BOTH colour schemes,
and check the empty, loading and error screens carry the new direction too.

[Clicking Save on the edit form does nothing.]

Steps: [open an invoice → change the amount → click Save].
Expected: [the dialog closes and the row updates].
Actually happens: [the dialog stays open, the row is unchanged, no error appears].
Started after: [the turn where we added validation].

Investigate first and tell me the cause before you change anything.

[Tapping a habit row crashes the app.]

Steps, expected, actual, and: it happens in Expo Go on my iPhone but NOT in the
web preview. Investigate before fixing.

Refactor [the invoice fetching] to [go through one shared hook instead of four
copies]. Behaviour must not change — same data, same states, same URLs.

List the files you'll touch first, then do it. Don't rename anything public and
don't reformat files you aren't otherwise changing.

Refactor [the four screens that each hand-roll useState + useEffect + try/catch]
to use the shared async-data hook. Same four states on every screen. No behaviour
change, no new dependencies.

Add reminders. A time picker in Settings, permission requested at the moment they
flip the toggle on, and a daily local notification at that time.

Also register the device token against the user row so I can send server pushes
later. Tapping a notification should deep-link into [the habit it's about],
not just open the app.

Add [barcode scanning] to [the stock screen]. Full-screen camera with a scan
frame overlay, a haptic on a successful read, and the matched product shown in a
sheet with an Add button. Permission asked when they tap Scan.

If the camera isn't available (web preview), show a designed fallback with manual
entry — don't render a dead black rectangle.

Gate the app behind Face ID. On launch, if a session already exists, require a
biometric check before revealing content. Fall back to the device passcode.

If the device has no biometrics enrolled, let them in on the existing session —
don't lock them out. Add a toggle in Settings to turn it off.

Make [the entries list] work offline. Cache the last fetched rows on the device,
render them immediately on launch while the network request runs, and reconcile
when it returns.

Writes made offline queue and flush when connectivity comes back. Show a quiet
"offline — showing saved data" banner, not an error.

Set up deep links. [habits://habit/123] opens that habit's detail screen,
[habits://invite/<token>] opens the accept-invite screen, and anything unknown
lands on the home tab instead of a blank screen.

Handle both cases: the app already running, and a cold start from the link.

Add [Pro] as an in-app purchase. Real packages from the store, no fake unlock.

Paywall at src/app/paywall.tsx, presented from the blocked action. Monthly and
annual, annual pre-selected, savings computed from the two prices. Restore
Purchases, Terms and Privacy links. Entitlement checked on launch and after a
purchase.

If the purchase SDK isn't configured yet, say so on the button — don't flip a
local boolean.

Apple rejected the build. Here's the reviewer note verbatim:

"[paste the full rejection text, including the guideline number]"

Work out what in our app triggered it, fix it, and tell me what changed. If the
fix needs something only I can do in App Store Connect, say exactly what.