Who it is for

Keep what you read in a format that still opens

Bookmarks point at pages that change or disappear, and read-later services close, get acquired or start charging for the export. A clip is the text itself, in a format with no vendor and no expiry, stored where you decide.

Why saved reading disappears

Anyone with a long reading habit has lost a library at least once. A read-later service shut down and the export turned out to be a JSON file of URLs, most of them already dead. A note app changed its format, and the old files needed the old version to open. A bookmark folder from 2014 now resolves, page after page, to a parked domain or a redirect to somebody’s homepage.

The common thread is dependency. If reading you saved needs a company to still exist, an account to still work, or a format only one program understands, then it is borrowed rather than kept. Plain text has none of those dependencies, which is exactly why files written twenty years ago still open without ceremony today.

The failure people notice later is retrieval rather than storage. An archive nobody can search is a landfill with good intentions – you know you read something about a subject, in some year, on a site you cannot name, and there is no way in. Plain text fixes that in a way no interface can, because every tool on every operating system can search it, one grep across a decade of reading returns the file, and the file’s own header says which page it was and when it was published.

What an archive of clips looks like

• A durable, open format: plain text with YAML frontmatter, readable by any editor and by grep.
• No account and nothing uploaded: what you clip stays on your own disk.
• Publication date and source URL captured with the text, so a note is still identifiable years later.
• The extraction field records the route: dom from the rendered page, jsonld-articlebody from the page’s own structured data.
• No limit on how much you clip: no quota, no daily cap, no paid tier.
• A name collision gets a numeric suffix, so an earlier capture of the same page is never overwritten.
• Line-based text compresses, deduplicates, syncs and diffs like source code, so backing up a decade of reading costs very little and restoring it needs no special tool.
• A per-site rule files a whole domain on arrival, so the archive sorts itself while you read instead of waiting for a tidying session that never comes.

---
title: "Things You Should Never Do, Part I"
source: "https://www.joelonsoftware.com/2000/04/06/things-you-should-never-do-part-i/"
author: "Joel Spolsky"
date: "2000-04-06"
extraction: "dom"
---

They did it by making the single worst strategic mistake that any software
company can make: they decided to rewrite the code from scratch.

Setting up an archive folder

The whole configuration exists to make one thing true – clipping must be cheaper than deciding whether to clip. Everything else follows from that.

1. Choose one folder and treat it as the archive – Archive, at the top level of wherever your documents live. One folder, not one per topic – topics are what search is for, and a folder tree is a filing decision you have to make at the moment of reading.
2. Open Options from the extension icon, point the destination there, and set what the icon click does to “save to folder”. No window, no dialogue, no decision: the whole habit depends on it costing one keystroke.
3. Set the filename template to {date}-{domain}-{title}. That gives you chronological order, the site at a glance, and enough uniqueness that two clips of different pages never argue over a name.
4. Turn on every frontmatter field. date is the page’s own publication date, extraction records how the text was obtained, and both are the sort of thing you cannot add back later.
5. Add per-site rules for the handful of sites you read constantly, each with its own subfolder. The archive then sorts itself for the material you clip most, which is where a flat folder gets unwieldy first.
6. Include the folder in whatever backs up your documents, and once, genuinely once, restore a file from that backup and open it. An archive nobody has ever restored from is a hypothesis.
7. Clip as you read rather than in tidying sessions. The pages you would have gone back for are precisely the ones that will have gone.

Settings for keeping things

These values assume the archive will outlive the machine it started on, and probably the browser too. Nothing here creates a dependency that has to survive with it.

Archive/
  2019-08-14-the-website-obesity-crisis.md
  2026-01-07-the-website-obesity-crisis.md
  2026-01-07-the-website-obesity-crisis-2.md

The third file is a second clip made the same day: a name collision takes a
numeric suffix instead of overwriting. The 2019 file still opens in any
editor, and its frontmatter still names the URL it came from.

Three years of the same habit

Reading in the morning, archiving by accident

Six long reads over a week, each clipped with one keystroke as you finish it. No window opens, nothing is filed, no decision is made about whether this one is worth keeping – which is the point, because that decision is what stops archives from forming.

What accumulates is a folder where the filename gives you date, site and title, and the first four lines of each file give you the URL, the author and the publication date. Nothing was organised, and it is still searchable, because the naming did the organising.

Finding something you read years ago

You remember a phrase and nothing else: not the site, not the year, not the author. A single search across the archive folder returns the file, and its header names the page. No index was built, no application had to still exist, and the search worked with the network off.

This is the specific thing a bookmark folder cannot do. A bookmark stores where something was, not what it said, so a search across bookmarks can only match titles that somebody else chose.

The site goes down and the text does not

A blog you read for years stops resolving. The clips are unaffected: the text, the author and the publication dates are in files on your disk, and the source line still records the address even though nothing answers there now.

The images are the honest exception. They were links, so they break when the site does, and a page whose value was its photographs is not preserved by this. That is the trade a text archive makes, and it is worth knowing which of your pages fall on the wrong side of it.

Against the other ways of keeping reading

The strongest option in this table is not this one. A full-fidelity archive keeps more; it also costs more, in space, in setup and in the number of things that have to keep working.

What goes wrong over years

The images in old clips are broken

They are links, and the sites they point at have moved, redesigned or gone. The extension downloads no binary assets, so an archive of clips is an archive of text. Where the pictures are the point (photo essays, diagrams, comics) either save them separately at the time or accept that this is not the tool that preserves them.

The folder is filling up with near-duplicates

That is the collision rule doing its job – a second clip of the same page takes a numeric suffix instead of replacing the first, because the second capture of a page that has changed is often the more interesting one. Nothing deduplicates automatically. On plain text, finding duplicates is a one-line job for any of the usual tools, and comparing two of them is a diff.

Half the archive is called “Home” or “Untitled”

The title comes from the page’s own metadata, and some sites give every page the same one. That is what {domain} and {date} in the filename template are for – with both in the name, an unhelpful title is still a findable file. If the archive already has a hundred of them, they are text files – renaming them by hand or by script is the fix, and nothing inside them is lost either way.

Some pages simply cannot be clipped

Three categories. Pages the browser protects – chrome:// addresses and the extension store – where no extension may run at all. Pages with no article body, such as feeds, storefronts and search results, which are reported as “no article” rather than returned as link lists. And anything that never rendered for you, including content inside an embedded viewer, which is a separate document the clip does not reach into.

What it does not preserve

It saves text, not pages: no layout, no screenshots, no fonts, no scripts and no downloaded images – image links keep pointing at the original site and break when that site does. It is not a web archive and carries no third-party timestamp. It does not read your history or bulk-import your bookmarks; you clip the page you are on. And it captures what the browser rendered, so a page that never renders for you cannot be kept.

Questions