#21: fix the DKIM, onboarding and console defects found reviewing the new work

DKIM rotation broke the domain it was run on. It retired the live key and
inserted a placeholder as ACTIVE - private_key_enc was the literal string
"dummy-encrypted-key" - so FindActiveKeys then failed for the whole
domain and outbound mail stopped being signed, with the real key already
gone. Reproduced against a database before changing anything.

The root cause runs deeper: the auth service derives its encryption key
with a plain SHA-256 of the master key while the protocols service uses
HKDF with its own info string, so nothing the auth service seals can ever
be opened by the signer. Rotation therefore moves to the protocols
service, which owns that vault and the real key generator. It retires the
previous key with a seven-day overlap in the same transaction rather than
deleting it, because mail already sent under it is still being verified.

Domain onboarding had three defects in one function. It built the alias
destination by interpolating the domain name into the SQL text, which was
injectable; it reused one placeholder for a CITEXT and a VARCHAR column,
which made the statement fail and happened to hide the injection; and it
ignored its scope argument entirely, so any DOMAIN_ADMIN could create
domains. It also provisioned postmaster, abuse and security, missing the
dmarc and tlsrpt aliases that the published records point at - PLAN.md
section 7.4b is explicit that reports bounce without them.

The TLS panel returned fixed values: acme_status "VALID" and 72 days to
expiry, whatever the process actually held. It now reads the certificate
watcher, and a compile-time assertion keeps the wiring from silently
regressing into "disabled". The audit log and the readiness scorecard were
fixtures too, and their field names did not match the API they now read.

Two tests failed and a third asserted a fabricated constant: one depended
on test order through a shared fixture that had a second factor enrolled,
one used a fixed domain name and so passed once per database, and one
expected "records: 13" from a function that never touches DNS. All three
are fixed, and the suite now passes twice in a row against one database.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

services/webmail/src/lib/sanitizer.ts

@@ -1,45 +1,50 @@
 import DOMPurify from "dompurify";
 
 /**
- * Sanitizes HTML content using DOMPurify with email-safe configuration (Section 14.2).
- * Strips scripts, dangerous attributes, and executable elements while keeping layout.
+ * Sanitizes message HTML with DOMPurify (PLAN.md section 14.2).
+ *
+ * There is deliberately no hand-rolled regex pass in front of DOMPurify. The
+ * previous one turned "<scr<script>ipt>" into a working "<script>" by deleting
+ * the inner match and joining the halves - it manufactured the tag it was
+ * meant to remove. It also left javascript: URIs and <form> untouched, because
+ * a regex cannot parse HTML. DOMPurify parses, which is why it is the only
+ * thing here.
  */
 export function sanitizeEmailHtml(rawHtml: string): string {
   if (!rawHtml) return "";
 
-  // Base sanitization stripping dangerous tags and inline event handlers
-  let cleaned = rawHtml
-    .replace(/<script\b[^<]*(?:(?!<\/script>)<[^<]*)*<\/script>/gi, "")
-    .replace(/<iframe\b[^<]*(?:(?!<\/iframe>)<[^<]*)*<\/iframe>/gi, "")
-    .replace(/<object\b[^<]*(?:(?!<\/object>)<[^<]*)*<\/object>/gi, "")
-    .replace(/<embed\b[^<]*(?:(?!<\/embed>)<[^<]*)*<\/embed>/gi, "")
-    .replace(/\s*on\w+\s*=\s*(?:"[^"]*"|'[^']*'|[^\s>]+)/gi, "");
-
-  if (typeof window !== "undefined") {
-    cleaned = DOMPurify.sanitize(cleaned, {
-      ALLOWED_TAGS: [
-        "a", "b", "blockquote", "br", "caption", "code", "div", "em", "h1", "h2", "h3",
-        "h4", "h5", "h6", "hr", "i", "img", "li", "ol", "p", "pre", "span", "strong",
-        "table", "tbody", "td", "tfoot", "th", "thead", "tr", "u", "ul", "style", "font",
-        "sub", "sup"
-      ],
-      ALLOWED_ATTR: [
-        "align", "alt", "bgcolor", "border", "cellpadding", "cellspacing", "cite",
-        "class", "color", "colspan", "dir", "height", "href", "id", "lang", "rowspan",
-        "src", "style", "title", "width", "target", "rel", "data-blocked-src"
-      ],
-      ALLOW_DATA_ATTR: true,
-      FORBID_TAGS: ["script", "iframe", "object", "embed", "form", "input", "button"],
-      FORBID_ATTR: ["onerror", "onload", "onclick", "onmouseover", "onfocus"],
-    });
+  // Without a DOM there is nothing to parse with, so nothing is returned.
+  // Failing closed matters: returning "best effort" markup would put
+  // unsanitized HTML on the page whenever this ran outside the browser.
+  if (typeof window === "undefined" || !DOMPurify.isSupported) {
+    return "";
   }
 
-  return cleaned;
+  return DOMPurify.sanitize(rawHtml, {
+    ALLOWED_TAGS: [
+      "a", "b", "blockquote", "br", "caption", "code", "div", "em", "h1", "h2", "h3",
+      "h4", "h5", "h6", "hr", "i", "img", "li", "ol", "p", "pre", "span", "strong",
+      "table", "tbody", "td", "tfoot", "th", "thead", "tr", "u", "font", "sub", "sup",
+    ],
+    ALLOWED_ATTR: [
+      "align", "alt", "bgcolor", "border", "cellpadding", "cellspacing", "cite",
+      "class", "color", "colspan", "dir", "height", "href", "lang", "rowspan",
+      "src", "style", "title", "width", "target", "rel", "data-blocked-src",
+    ],
+    // style, link, meta and base are gone: a stylesheet in a message can cover
+    // the rest of the frame, and <base> rewrites every relative URL in it.
+    // id goes too, so a message cannot collide with the host document's ids.
+    FORBID_TAGS: ["script", "iframe", "object", "embed", "form", "input", "button", "style", "link", "meta", "base"],
+    FORBID_ATTR: ["srcset", "formaction", "ping", "id"],
+    ALLOW_DATA_ATTR: false,
+    // data: is how an anchor smuggles a document onto this origin.
+    ALLOWED_URI_REGEXP: /^(?:https?|mailto|cid|tel):/i,
+  });
 }
 
 /**
- * Replaces remote image sources (http/https) with data-blocked-src to protect user privacy
- * against tracking pixels until remote images are explicitly enabled.
+ * Replaces remote image sources with data-blocked-src so a tracking pixel
+ * cannot fire before the reader asks for images.
  */
 export function blockRemoteImages(html: string): string {
   if (!html) return "";
@@ -49,10 +54,8 @@ export function blockRemoteImages(html: string): string {
     .replace(/url\(\s*['"]?https?:[^)]*\)/gi, "none");
 }
 
-/**
- * Restores blocked remote image sources when user clicks "Load Remote Images".
- */
+/** Restores the blocked sources once the reader opts in. */
 export function unblockRemoteImages(html: string): string {
   if (!html) return "";
-  return html.replace(/(<img[^>]+)data-blocked-src\s*=\s*("|')([^"']+)\2/gi, '$1src="$3"');
+  return html.replace(/(<img[^>]+)data-blocked-src\s*=\s*("|')([^"']+)\2/gi, "$1src=$2$3$2");
 }