Pory

Promptfree, for everyone, and the only version there is

Copy the prompt for Claude CodeCopy the prompt for CodexCopy the prompt for Cursor

Copy

Build me a site generated from my table, replacing Pory: a directory or a listing that updates when the data does, with filters and detail pages.

STACK
- Astro with a Node adapter — server-rendered so the site is current, with aggressive caching so it is fast
- SQLite through better-sqlite3 as the local mirror
- The source can be a spreadsheet, a table product's API, a CSV, or a database. One adapter interface, several implementations
- Caddy in front

THE SHAPE
- The data lives wherever it already lives. This pulls it, mirrors it locally, and serves pages from the mirror
- Serving directly from a spreadsheet API is the obvious approach and the wrong one: it is slow, it is rate-limited, and the site goes down when the source does
- So: sync on a schedule and on demand, serve from the mirror, and show the last successful sync somewhere the owner can see

THE DATA MODEL
- sources: id, kind, config_encrypted, sync_interval_seconds, last_synced_at, last_error, is_active
- collections: id, source_id, name, slug, table_ref, sync_cursor
- fields: id, collection_id, key, label, kind, is_filter, is_search, is_visible, position, format_json
- records: stored in a real table per collection, created and altered by generated migrations, keyed on a stable identifier from the source
- record_revisions: id, collection_id, record_id, field_key, old_value, new_value, at — so a change in the source is visible rather than silent
- sync_runs: id, source_id, started_at, finished_at, created, updated, deleted, status, error
- Identity comes from a column you nominate, never from row position. A sorted spreadsheet must not renumber every page on the site

SYNCING
- Incremental where the source supports it, full otherwise, with a cursor
- A record that disappears from the source is marked missing rather than deleted, and its page returns a clear gone state rather than a broken link. Somebody deleting a row by accident must not silently remove a page search engines have indexed
- Field types inferred on first sync and then fixed; a type change is a deliberate migration with a dry run
- Rate limits respected with backoff, and credentials encrypted at rest
- A manual sync button, and a webhook so the source can trigger one

THE PAGES
- An index with the records, a detail page per record at a permanent slug, and category or tag pages if the data has them
- The slug comes from a nominated field plus the stable identifier, and once published it never changes — a renamed row writes a redirect rather than breaking the link
- Filters generated from the fields marked filterable: checkboxes, ranges, a select. Server-rendered with the state in the query string, so a filtered view is a URL somebody can share and a search engine can index
- Search over the fields marked searchable, using FTS5, server-side, instant
- Sorting, and pagination that is real pages rather than an infinite scroll, because a directory's value is largely in being indexed

WHAT MAKES A DIRECTORY WORTH ANYTHING
- Every detail page has its own title, meta description, canonical URL and JSON-LD of the right type — LocalBusiness, Product, Event, Person, whatever the data actually is
- A sitemap generated from the records, split if it grows past the limit, and submitted
- An RSS feed of new entries
- A generated share image per record from its own title
- Fast: server-rendered HTML, no framework, under 100KB, with the page cached and invalidated by the sync
- Without these a directory is invisible, and being found is the entire business model of one

TEMPLATES
- A handful of layouts — directory, listing, gallery, map — each a real template in the repository rather than a configuration blob
- Fields mapped into template slots, so changing which column is the title is a setting rather than an edit
- Theme tokens: two backgrounds, two text colours, one accent, one border, all measured against WCAG AA
- Dark and light

SUBMISSIONS
- A form that writes a new row back to the source, or into a pending table for review
- Nothing published without review, ever. A directory anybody can add to is a directory full of spam within a week
- Honeypot, minimum time on form, rate limits per address hash

OPERATIONS
- .env: DATABASE_PATH, BASE_URL, SOURCE credentials, SESSION_SECRET, HASH_SALT
- Migrations on boot, each once
- Health endpoint reporting the age of the last successful sync and the last error — a stale site that looks fine is the failure mode here
- Nightly backup off the machine, restore script

WHAT MATTERS MOST
Stable identity and the search-engine surface. Nominate a real identifier column before anything else and never let a slug change, then get the sitemap, the structured data and the per-record titles right. A directory nobody can find is a spreadsheet with extra steps.

Give me the repository, one source adapter, migrations, .env.example, two templates, and a README with deploy steps behind Caddy.