Inspiration
Gopher was the plain-text internet before the web took over. Today a single news article can carry megabytes of ads, trackers and scripts just to show a few paragraphs, which is slow on weak connections and hard on screen readers. For the Resurrection category, I wanted to bring a dead technology back to solve exactly that problem.
What it does
GopherGhost turns any web page into a clean, fast Gopher menu. You paste a web address, and it strips out scripts, ads, menus and footers, then returns an RFC 1436-style Gopher menu with clickable links. A bloat meter shows how heavy the original page was compared with its Gopher form, along with a running total of what has been banished during the session.
It also speaks real Gopher, so you can browse live gopher:// servers such as Floodgap, with clickable folders and text files. Back and forward history, a Trail of visited pages, keyboard shortcuts (slash for the address bar, Backspace to go back) and a raw view that shows the actual Gopher menu text make it easy to move around. The interface is a haunted CRT terminal with a boot sequence, scanlines and an ASCII ghost, and the animations switch off when a visitor prefers reduced motion.
How I built it with Kiro
I built it spec-first in the Kiro IDE. The requirements, design and tasks live in .kiro/specs/gopherghost, and Kiro implemented the work task by task, including the history stack, the raw view, the gopher:// TCP client and the bloat meter. Three steering documents (product, tech and structure) set the rules, such as keeping the core parsing code pure and rendering untrusted text safely.
Two hooks automated the workflow. One runs the tests when code changes, and the other keeps the task list in sync with the code, and it ticked off completed tasks on its own. I used property-based testing with fast-check to check that a menu always round-trips through serialize and parse, that text wrapping never exceeds the width or loses characters, and that the bloat percentage always stays between 0 and 100. Compared with vibe coding, having requirements written first meant Kiro's changes could be checked against something concrete, and the task list became a record of the work. I also used Claude to design the final interface, then connected it to the Kiro-built server.
Challenges
Turning messy real-world HTML into readable fixed-width text was harder than it looks, including word wrapping and link extraction. Safety was another challenge, because the server fetches arbitrary addresses, so it blocks private and internal hosts, re-checks every redirect, and limits fetch time and size. Speaking real Gopher over raw TCP and wiring it into the web interface meant matching two separate server endpoints with the page. Kiro also found and fixed a back-button and history-position bug along the way.
What I learned
Gopher's menu format is tiny and strict, which makes it an ideal target for property-based tests. I also learned that a written spec gives an AI coding tool far better direction than a loose prompt.
What's next
An AI ghostwriter summary at the top of each page, and support for Gopher search items.
Built With
- cheerio
- css
- fast-check
- gopher
- html
- kiro
- node.js
- render
- typescript
- vitest