SkimRead 0.2.5

Adds Antigravity CLI support and improves two-column PDF reading order.

SkimRead 0.2.4

Improved narrative selection and page-structure filtering, clearer Core/Context and annotation-label controls, and a simpler setup guide.

SkimRead 0.2.3 — cleaner document zoning

Highlights are now filtered using document-wide zoning before model selection.

• Removes repeated running headers and footers
• Detects end-of-document bibliography zones without relying only on English headings
• Keeps focused papers in one full-document selection whenever their context fits
• Removes the Balanced coverage control
• Retains chapter/section-sized selection for documents that exceed the configured context window

The TL;DR, EPUB support, opt-in Zotero annotation export, and provider integrations remain available.

SkimRead 0.2.2 — smarter selection, fixed settings, new icon

Better selection, clearer settings, and a new icon.

Selection

A whole-document pass now decides what gets highlighted. Until now each
passage was judged on its own and every passage's picks went straight to the
output, so nothing ever saw the document as a whole. A new pass reads all
candidates in reading order and marks the subset that together narrates the
document, dropping sentences that merely restate a point already made.

The per-passage quota is also softer. It used to demand a fixed number from
every passage, which forced boilerplate sections to yield their full share while
capping the section carrying the argument. It now asks for up to that number.

Kept sentences are flagged rather than deleted, so the density slider means
something again: core sentences show first, and raising density adds context
around them instead of saturating at roughly two per page. The sidebar reports
how many were kept, for example 24 highlights across 9/12 pages · 11 core.

Sharper labels. Each label now carries what it is not, plus a couple of
example sentences. Neighbouring categories (result vs novelty, goal vs
conclusion) are where smaller models go wrong most often, and stating the
exclusion separates them better than a longer description does. Custom labels
accept antiDescription and examples too.

Fixes

• Highlights no longer start on the author byline. On a title page the byline
  runs into the abstract with no full stop, so the whole run was highlighted
  from the authors onward.
• Margin flags scale with the reader's zoom. They were a fixed size, so zooming
  in made them shrink relative to the text.
• The connection status line named Ollama whatever the provider was, and kept
  showing the previous one after a change. It now names the endpoint actually
  configured, refreshes when settings change, and can be clicked to re-check.
• The Claude model setting did nothing: the dropdown neither loaded nor saved,
  and could not be opened. Codex reasoning had the same fault silently.

Settings

• Claude model is a proper picker (Sonnet / Haiku / Opus, with what each suits).
• Codex model is a picker too, populated from the Codex app server's own model
  list rather than hardcoded names.

Annotations

Saved annotations now carry their label as a Zotero tag. Colour alone is lossy —
easy to change, ambiguous across a shared palette, and invisible to search —
while a tag keeps the category attached, filterable, and groupable in Add Note
from Annotations.

Appearance

New icon: three highlight bands in the same palette the plugin paints, replacing
the previous mark. The sidebar version follows Zotero's theme.

Install

Download skim-read-0.2.2.xpi below, then in Zotero: Tools → Plugins →
gear menu → Install Plugin From File…, and restart Zotero.

Existing cached runs are regenerated on first use, since the prompts changed.

SkimRead 0.2.1 — EPUB annotations

Completes the EPUB support introduced in 0.2.0.

Added

EPUB highlights can now be saved as native Zotero annotations. 0.2.0 shipped
EPUB highlighting with this listed as a known limitation. It turned out not to be
a real limitation: the reader view can produce a Zotero annotation directly from
a DOM range, including the EPUB CFI position, and the overlay painter already
builds exactly those ranges. Save as Zotero annotations… now works for EPUBs
as it does for PDFs, and the results work with Zotero's Add Note from
Annotations command.

Because CFIs are generated from live ranges, only passages in chapters the
reader has already rendered can be converted. Any that cannot are reported
("6 skipped (chapter not rendered yet)") rather than silently omitted. Scrolling
through the book once resolves it.

Saving remains explicit and opt-in. Nothing is written to your library until you
press the button.

Documentation

Added a Limitations and bias section to the README. Choosing what to
highlight also decides what you do not look at, which is worth stating plainly.
It covers the discipline bias of the default Goal / Method / Result / Novelty
scheme, the text dropped during extraction (footnotes especially, which carry
the argument in historical and legal scholarship), weaker judgement from small
models, and the fact that what is not highlighted simply disappears.

Install

Download skim-read-0.2.1.xpi below, then in Zotero: Tools → Plugins →
gear menu → Install Plugin From File…, and restart Zotero.

Release v0.2.0

❤️ Contributors

• Adel Tayebi

Release v0.1.1

📖 Documentation

• Fix Develop test instructions and add Acknowledgements (03cb1d7)

❤️ Contributors

• Adel

Release v0.1.0

No significant changes.

Release Manifest

This release is used to host update.json, please do not delete or modify it!
Updated in UTC 2026-08-17T12:43:25.312Z for version 0.2.5