RSS Reader

Project roadmap

1. 01

   Define the data and API

   ~1.5h

   Write DESIGN.md: feeds (URL, title, last fetched, ETag and Last-Modified, and last error), articles (feed, a unique id from the feed's guid or link, title, link, author, published date, a cleaned summary, read and starred flags), and the API: add and remove feeds, list articles (by feed, unread only, starred, paginated), and mark read, unread and starred. Done when: DESIGN.md defines the data and every endpoint.

2. 02

   Fetch and parse feeds

   ~2.5h

   In Node.js, fetch a feed with a timeout and a descriptive user agent, sending If-None-Match and If-Modified-Since from the last fetch (a 304 means nothing changed). Parse RSS and Atom (with a library such as fast-xml-parser or rss-parser) into one article shape, choosing a stable id and parsing dates in their different formats. Done when: several real feeds of both formats are parsed into consistent articles.

3. 03

   Store and deduplicate

   ~2h

   Store feeds and articles in SQLite (for example with better-sqlite3), with a unique index on feed plus article id, and insert new articles while ignoring ones already stored. Clean article HTML with a sanitizer (such as sanitize-html) allowing only safe tags, and create a short plain-text summary. Done when: fetching the same feed twice doesn't create duplicates, and stored HTML contains no scripts.

4. 04

   Refresh on a schedule

   ~2h

   Refresh all feeds every 30 minutes (and one feed on demand), a few at a time. Record errors per feed without stopping others, back off feeds that keep failing, and mark feeds as broken after many failures. Log each refresh's result. Done when: a broken feed URL doesn't stop the others from updating, and its error is shown in the feed list.

5. 05

   Build the API

   ~1.5h

   Expose the endpoints from DESIGN.md with Express (or another framework), validating input: adding a feed checks the URL and that it actually returns a feed before saving. Return unread counts per feed. Done when: every endpoint works with curl.

6. 06

   Build the React reader

   ~2.5h

   Build a React interface with a feed list showing unread counts, an article list (newest first, with unread in bold), and a reading pane showing the cleaned content with a link to the original. Opening an article marks it read; add buttons for unread, star, mark all as read, and filters for unread and starred. Add keyboard shortcuts (j and k to move, s to star). Done when: reading, marking and filtering work smoothly.

7. 07

   Test it

   ~2h

   Write tests with saved feed files (RSS, Atom, broken XML, missing dates and duplicate items): parsing, id choice, deduplication, HTML cleaning, the 304 handling with a fake server, and the API endpoints. Done when: all tests pass without network access.

Resources

• DocsRSS 2.0 specification
• DocsRFC 4287: the Atom syndication format
• DocsMDN: HTTP conditional requests
• Docssanitize-html
• DocsReact: quick start