v2.5.1 - width clamp measures the container, not the window

Bug fix. No visual rule changed.

Symptom: at the Ultra preset the left sidebar was intermittently squeezed; a reload cleared it.

Cause: the responsive term in the width clamp was calc(100vw - 3rem), and 100vw is the whole window including the sidebar - so the clamp permitted a column wider than the space <main> actually has. Measured in a mock of the same flex layout: a 1440px window with a 260px sidebar leaves main 1143px while the Ultra clamp still allowed 1248px.

Fix: calc(100cqi - 3rem) - the inline size of the nearest query container, which is the real column. Container inline size is definite by containment, so unlike a percentage it can never fall into cyclic sizing; with no container ancestor it degrades to the old behaviour rather than a broken one.

Also: a layout-pressure probe in docs/selectors.md for capturing the failure live, and a troubleshooting row in both READMEs.

Update in Stylus -> Manage -> Check for updates. Do not re-import: that creates a second copy and resets your saved width.

DenseGPT v2.5.0

Make ChatGPT denser, calmer and easier to scan - visually and linguistically.

Install

Open densegpt.user.css with Stylus installed, then reload chatgpt.com.

To update later, use Stylus -> Manage -> Check for updates. Do not re-import - that creates a second copy of the style, and two copies make your settings look like they do nothing.

What is in it

Visual density. Nine rules on assistant markdown: paragraph rhythm, heading scale, list spacing, inline code, blockquote, italics and the horizontal rule. Body colours, links, tables, bold and fenced code blocks are untouched, and there is no theme detection at all - light and dark are inherited from ChatGPT exactly as shipped.

Content width. Replaces the Widescreen extension. Four presets - Reading 44rem, Balanced 54rem, Wide 66rem, Ultra 78rem - plus a free 36-90rem slider, configurable from the Stylus gear. Every value retreats on narrow windows, and assistant messages, your messages and the composer stay on one centre axis.

Response density. STYLE.md and presets/ are paste-ready instructions that make the model itself answer densely - conclusion first, no closing summaries, no one-sentence paragraphs, Markdown as structure rather than decoration. The ChatGPT preset fits the 1500-character custom-instruction field.

Notable in this release

• Width settings work. Every earlier release used USO-style placeholders under @preprocessor default, which compiles @var into CSS custom properties instead - so the declaration was empty and content width silently did nothing at any preset.
• Width also covers the turn wrapper, which redeclares ChatGPT's width variable on itself and shadowed the value set higher up.
• Blockquote colours restored to the values that were verified by eye, as literals rather than derived tokens.

Full history, including what failed and why, is in CHANGELOG.md. Documentation index is in the README.