Review (via AI-assisted code review): REQUEST CHANGES

The direction (Trusted Types compatibility, dropping custom URL munging in favor of safevalues) is good, but there are blocking issues:

1. New runtime dependency without a lockfile update. safevalues ^1.2.0 is added to packages/quill/package.json dependencies, but no package-lock.json change is included (the PR touches 12 files, none of them the root lockfile). npm ci in CI and for contributors will fail with a lock-out-of-sync error. Please regenerate and commit the lockfile.

2. The central Trusted Types policy is an unsanitized passthrough. core/utils/html.ts creates a policy named quill with createHTML: (s) => s. This provides zero protection: any application that enables Trusted Types and allowlists the quill policy name (a natural thing to do once this ships) re-opens every sink Quill feeds through it with no review. It also silently no-ops when policy creation throws (e.g., CSP trusted-types directive with a restricted policy-name allowlist), which is fine, but the feature as implemented is compliance theater rather than a security improvement. Either sanitize inside the policy, restrict its use to inputs that are actually trusted, or document clearly that it exists only to satisfy require-trusted-types-for and must not be allowlisted blindly.

3. Sink coverage is partial. parseFromString is wrapped, but DOMParser.parseFromString is not in Chromium's default Trusted Types sink set, so the main wrapper addresses a non-violation, while real innerHTML sinks outside the picker/theme path (e.g. tooltips, normalizeExternalHTML) are untouched. The PR claims to fix "innerHTML and parseFromString violations" but only covers the icon/picker subset.

4. Tests don't test the feature. The added image.spec.ts/video.spec.ts only re-assert existing sanitize behavior (javascript: → //:0/about:blank) — nothing exercises Trusted Types enforcement, the policy fallback, or that the safevalues setters behave identically to setAttribute under a require-trusted-types-for 'script' document. At minimum add a test with a TT-enforcing stub.

Also minor: setElementAttribute(node, 'src', ...) from safevalues/dom rejects security-sensitive attribute/element combinations for plain strings — please verify img src with a string value doesn't throw under safevalues' checks. And the diff adds stray blank lines in several files.

Branch is based on current main; the video iframe caveat is acknowledged and reasonable. No CI has run on this fork head yet.