[michaelwang11394]

What

Hardening across the docs site, the SVG sanitizer, lint, the GSAP writers and the registry:

• Docs catalog preview: variable values are encoded for the inline script they land in. The install-command copy button quotes URL parameters for the shell.
• SVG sanitizer: CR LF is folded before CSS escapes are decoded, so an escaped function name followed by a line break is still caught. The media-use bundle is regenerated.
• Lint: the template-id rule no longer builds a backtracking regex from composition text, and the CSS-rule scanners run in linear time, including on a closed <style> holding many { with no }. The two gsap-rule scans use a small scanner that returns the same selector and body pairs as the old regex. Findings are identical on every tracked HTML file.
• GSAP writers escape </script and <!-- in values they write into inline scripts, and refuse a tween method other than set/to/from/fromTo or a non-numeric motion-path value.
• scripts/host-registry-assets.ts refuses paths outside the item and symlinks, with a test.
• The five map blocks pin d3@7.9.0, us-atlas@3.0.1 and world-atlas@2.0.2 with subresource integrity, and a test rejects floating CDN versions.

Why

Each change keeps untrusted text (composition HTML, URL parameters, SVG input, registry paths) from becoming code or reaching files it should not, and keeps third-party scripts from changing under a floating version.

Related work

#3953 also edits the map blocks and will need a small rebase after this lands.

How

Local fixes in each module. d3 is pinned to 7.9.0, the long-standing release (7.9.1 and 7.9.2 were published days ago); the producer's d3 fixture already uses 7.9.0.

Test plan

• Unit tests added/updated; each fails without its fix (the old template-id code hung past a 120 s timeout).
• Lint findings are identical before and after the scanner change on all 1089 tracked HTML files (1633 findings). The scanner returns the same pairs as the old regex on 200,000 random inputs that include nested @media blocks. The full lint (1288) and parsers (1426) suites pass. scripts/registry-cdn-pins.test.ts passes.
• The 7.9.0 integrity hash matches the CDN file. All five map blocks load in headless Chrome with d3 7.9.0 under that hash, draw their state or country paths, and log no errors.
• The generated catalog pages and payloads are not committed; the catalog publish job regenerates them from the item sources.
• Comment ratchet and pre-commit hooks pass.
• fails without the fix: scans a closed style block of many unclosed rules in linear time (lint, 2870 ms against a 2000 ms limit with the previous gsap.ts), matches a many-placeholder template id against a long near-miss id in linear time (lint), drops: CSS-escaped image-set() followed by CR LF in a style element and ... in style= (core), the six escaping, tween-method and motion-path cases in gsapWriter.scriptContext.test.ts (parsers), a manifest path outside the item is refused before anything is staged or removed and an asset that is a symlink is refused before its target is read (scripts), shown by running each test file with the merge-base version of the source it covers. scripts/catalog-detail.test.ts does not load without the fix, because the helpers it extracts are absent. The form-feed sanitizer case already passes on the base; it is kept as a guard for the line-ending fold.
• Not done here: the <script>/<style> lint regexes are still slow on very large unclosed input, server-side lint has no time budget yet, and the docs catalog preview is not moved to an opaque-origin sandbox (it would break the preview).

🤖 Generated with Claude Code

[mintlify]

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

Project	Status	Preview	Updated
hyperframes	🟢 Ready	View Preview	Nov 10, 2026, 6:24 PM

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

[github-actions]

Edit accuracy: accurate 267 (base branch 267), smooth 90 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 (8)

• seqplainkeys-none-center-r0-nested-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): pass, passes
• seqplainkeys-none-center-r0-root-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): pass, passes
• seqplainkeys-none-center-r30-root-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): not run
• seqplainkeys-none-pct-r0-root-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): fail / pass / pass, passes
• seqplainkeys-none-pct-r30-root-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): not run
• seqplainkeys-none-px-r0-nested-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): pass, passes
• seqplainkeys-none-px-r30-nested-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): not run
• seqplainkeys-none-px-r30-root-z100 (fixed by studio: undo/redo on a keyframed plain element sometimes leaves the box 230-400px off (edit-accuracy seqplainkeys-none-*) #6989): not run

Unstable (2)

• seqrotatenudge-none-pct-r0-root-z100: tracking 0.03, pressJump 0, drop 0, reload 3.03, render 0.04, renderKey -, undo false, teleport true / tracking 0.03, pressJump 0, drop 0, reload 0, render 0.08, renderKey -, undo true, teleport true / tracking 0.03, pressJump 0, drop 0, reload 0, render 0.08, renderKey -, undo true, teleport true
• seqplainkeys-none-px-r0-root-z100: 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 / tracking 0.06, pressJump 0, drop 0, reload 0, render 0.03, renderKey -, undo true, teleport true

[heygengenesis]

Verdict: REQUEST CHANGES · 1 blocker · 0 important · 2 nits · head ae81778

• blocker packages/lint/src/rules/gsap.ts:426 and :2354: both CSS-rule scans still use /(?<![^{}])([^{}]+)\{([^}]+)\}/g. The body [^}]+ can cross {, so the scan is quadratic. A closed <style> holding "x{".repeat(40000) (80 KB) takes about 4.7 s per regex, and lintHyperframeHtml takes about 10 s end to end (N=10k/20k: 271 → 1150 ms in plain node). This is a closed tag, not the "very large unclosed input" the body leaves out, and it contradicts "the CSS-rule scanners run in linear time". Fix: use a [^{}]+ body or an indexOf scan, rerun the lint-parity check (nested @media … { .a {…} attribution), and add an x{×N timing test.
• nit sanitizeSvg.test.ts: the "CSS-escaped url() followed by a form feed" case passes on base, because the old [ \t\n\r\f]? already took a lone \f. Only the two CR LF cases are real regressions.
• nit: the body has no fails without the fix: line naming tests.

Scope

I ran 29 SVG bypass probes (xlink/aliased javascript:, external <use>, foreignObject, animate/set, mixed case, escaped url()/@import/image-set, feImage, PI, entities, CDATA), and all were stripped or inert. Also checked: the GSAP writers (no bare JSON.stringify left), registry asset hosting (realpath + O_NOFOLLOW), and the d3@7.9.0 sha384, which matches jsDelivr. Docs filters drop no real registry entry. Tests pass: parsers 8/8, core sanitizeSvg 59/59, lint gsap 241/241, and the node --test scripts. CI is green.

— Genesis

[michaelwang11394]

Review follow-up:

• blocker (gsap.ts CSS-rule scans quadratic on "x{".repeat(40000)): fixed in bcb4e00. Both scans now use a linear scanner that returns the same selector and body pairs as the old regex (checked on 200,000 random inputs with nested @media). Lint findings are identical on all 1089 tracked HTML files. New test scans a closed style block of many unclosed rules in linear time fails with the previous gsap.ts (2870 ms against a 2000 ms limit).
• nit (form-feed sanitizer case passes on base): relabelled as a guard for the line-ending fold in 006da3d. Only the two CR LF cases are claimed as regressions.
• nit (no fails without the fix: line): added to the PR body.

[heygengenesis]

Verdict: APPROVE — 2 resolved, 0 open, 0 new · ae81778 → 006da3d. styleRuleBodies is linear: 0 mismatches against the old regex on 300k random inputs of my own, and about 1 ms on x{/a{b{/}{/a{}×40k. The new timing test fails on the previous gsap.ts (9.0 s). lint gsap 242/242, sanitizeSvg 59/59. CI is green.

— Genesis