Curated

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 link-roundup newsletter tool that replaces Curated: file a link the moment I see it, and let the issue assemble itself.

STACK
- Node 20+ with Fastify, server-rendered HTML
- SQLite through better-sqlite3, WAL mode
- SMTP for sending, through a provider set in .env
- Caddy in front

THE DATA MODEL
- links: id, url, canonical_url, title, description, author, site_name, image_path, note_md, category_id, status, source, added_at, used_in_issue_id
- categories: id, name, position, blurb — the sections an issue is built from
- issues: id, number, subject, preheader, intro_md, outro_md, status, scheduled_for, sent_at, web_slug
- issue_items: id, issue_id, link_id, category_id, position, note_override
- subscribers: id, email, token, confirmed_at, unsubscribed_at, source, ip_hash, created_at
- deliveries: id, issue_id, subscriber_id, sent_at, bounced_at, opened_at, complained_at
- clicks: id, issue_id, link_id, subscriber_id, created_at
- Every issue keeps the note as it was sent. Editing a link later must never change an issue already delivered

FILING A LINK, WHICH IS THE HABIT BEING SOLD
- A bookmarklet: one click on any page files it, with the title and description already filled from the page
- An inbox address: forward an email or send a URL and it appears in the queue
- A paste box that accepts a list of URLs at once
- Every capture fetches the page once, politely, and stores the title, description, author, site name and preview image locally — never hotlinking somebody else's image into my newsletter
- Duplicate detection on the canonical URL, and a warning if the link went out in an earlier issue, with the issue number

BUILDING AN ISSUE
- The queue on one side, the issue on the other, drag to place, sections in order
- Write a line about each link — that line is the product, so the editor is a plain textarea with Markdown and nothing that gets in the way
- Reorder, move between sections, drop back to the queue
- A preview that matches what will be sent, at phone width and at desktop width
- The subject and preheader written together, with the character counts that actually get truncated shown beside them

SENDING
- HTML built with tables and inline styles, because that is what mail clients render, and a plain-text alternative generated from the same source rather than written twice
- A test send to me first, always, and the interface refuses to send an issue that has not had one
- Batched sending with a rate the provider will accept, resumable, and idempotent per subscriber so a crash halfway never doubles up
- Bounces and complaints consumed from the provider's webhook and acted on: a hard bounce unsubscribes, a complaint unsubscribes and is never mailed again

THE LIST
- Double opt-in, a signup form and an embeddable one, a honeypot, a stored consent sentence with its timestamp
- One-click unsubscribe that needs no login, plus the List-Unsubscribe headers that mail clients use
- Import and export as CSV, complete, with the opt-in dates intact
- Segments by signup date and by engagement, no more

THE ARCHIVE
- Every sent issue on a public page at its own permanent URL, with the same content
- An RSS feed of issues, and one of links if I want it
- Open Graph tags and a generated preview image per issue
- Sitemap and canonical URLs

MEASURING
- Opens, clicks per link, unsubscribes per issue, list growth over time
- Open rate labelled as unreliable, because privacy proxies pre-fetch images
- Click tracking through my own redirect, on by default, switchable off per issue
- No third-party analytics anywhere

OPERATIONS
- .env: DATABASE_PATH, BASE_URL, SMTP_URL, FROM_ADDRESS, INBOUND_SECRET, SESSION_SECRET, HASH_SALT
- Migrations on boot, each once
- SPF, DKIM and DMARC documented in the README; a newsletter that lands in spam is not a newsletter
- Nightly backup off the machine, restore script
- Health endpoint that checks the SMTP connection

WHAT MATTERS MOST
The bookmarklet and the duplicate check. Build capture first and use it for a fortnight before writing the send path — if filing a link is not a one-click habit, the issue never gets written and the rest of the tool is furniture.

Give me the repository, the bookmarklet, migrations, .env.example, a seed with two issues and forty links, and a README with deploy steps behind Caddy and the DNS records to set.