[miguel-heygen]

What this fixes

A CSS rule that selects on the inline style attribute ([style], [style*=…], [style^=…] and the other attribute forms) makes every inline style write expensive. GSAP writes inline styles on every animated element every frame, so a composition with such a rule plays slowly in the preview:

• Chrome restyles the written element, and its whole subtree when the rule reaches descendants ([style*="visibility: hidden"] *). With a few hundred animated elements, that is most of the page on every frame.
• Safari re-encodes the element's whole inline style on each write once any rule selects on style.

The runtime's own two rules of this kind are removed in #5974. Authored rules of the same shape bring the cost back, and the lint today recommends one: the fix hint for runtime_hidden_style_opacity suggests [data-composition-src][style*="visibility: hidden"] guards.

Changes

• New rule inline_style_attribute_selector (error): flags any selector that reads the style attribute, including inside :not()/:is() and in any letter case. It reuses the CSS-selector pass and postcss-selector-parser the lint already has, and checks inline <style> blocks and the linked stylesheets the lint already reads. Each selector is reported once. Lint errors block a render only under --strict, as for the other rules.
• Fix hint (and docs) for runtime_hidden_style_opacity: no longer recommends a [style…] guard. It now says the runtime already hides a hidden clip and its contents, and to animate on the timeline or toggle a class or data attribute instead. The rule's detection is unchanged: it is about the video replacement frame, and a host-scoped guard does not affect that. The new rule flags such guards for their playback cost.
• docs/reference/lint-rules.mdx: the new rule, and the corrected fix.

Tests

packages/lint/src/rules/core.test.ts:

• Each selector form errors: [style], *=, ^=, $=, ~=, |=, uppercase with the i flag, and inside :not().
• Lookalikes pass: a class, [data-state], [data-style], [styles], [data-hf-style].
• A selector repeated across rules is reported once, and each selector of a list on its own.
• The runtime_hidden_style_opacity fix hint contains no [style.

On main the new tests fail (10 of them). Each was checked by breaking the rule:

• Matching only lowercase [style*=…] turns the other forms red.
• Skipping attributes nested in a pseudo-class turns the :not() case red.
• Matching any attribute that contains "style" turns the lookalikes red.

The whole lint suite passes three runs in a row. The lint package typecheck, bun run lint, oxfmt and the fallow audit on the changed files are clean.

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated
hyperframes	🟢 Ready	View Preview	Oct 25, 2026, 12:56 AM

💡 Tip: Enable Automations to automatically generate PRs for you.

[github-actions]

Edit accuracy: accurate 271 (base branch 271), smooth 142 of those

The gate passes.
Smoothness is reported in the artifact, not gated. A case fails only if it fails 2 of 3 runs.

Quarantined, measured but not gated (0)

Unstable (2)

• seqplainkeys-none-pct-r0-root-z100: tracking 0.06, pressJump 0, drop 0, reload 0, render 0.03, renderKey -, undo false, teleport true / tracking 0.06, pressJump 0, drop 0, reload 0, render 0.03, renderKey -, undo true, teleport true / tracking 0.06, pressJump 0, drop 0, reload 0, render 0.03, renderKey -, undo true, teleport true
• seqheldtwice-tlhold-px-r0-nested-z100: tracking 120.29, pressJump 312.3, drop 0.07, reload 0.07, render 0.03, renderKey -, undo true, teleport false / tracking 0.41, pressJump 0.11, drop 0.39, reload 0.39, render 0.02, renderKey -, undo true, teleport true / tracking 0.41, pressJump 0.11, drop 0.39, reload 0.39, render 0.02, renderKey -, undo true, teleport true

[heygengenesis]

Verdict: APPROVE · 0 blocker · 0 important · 1 nit · head b986bf81

• nit packages/lint/src/rules/core.ts:764 — a selector that trips runtime_hidden_style_opacity (e.g. [style*="visibility: hidden"] { opacity: 0 }) now produces two errors for the same rule. Both give the same fix, so this is only noise. Skip the new finding when the hidden-style one fires if you want a single error.

Scope

Read the full diff. selectsOnInlineStyle walks every attribute node, including those nested in :not()/:is(), and compares names case-insensitively. It runs over resolvedRuleSelectors, which covers inline and linked stylesheets. The runtime's own [style*=…] rules (timedClipHide.ts, pending #5974) are injected by JS and never linted. No skill, template, or doc in the repo still recommends a [style…] guard. The compiler's static guard only warns on errors, so the new rule blocks a render only under --strict. CI passed at this head.