fix(core): convert multi-line tags in cleanBody

.changeset/cleanbody-multiline-tags.md

@@ -0,0 +1,7 @@
+---
+"@dualmark/core": patch
+---
+
+Fix `cleanBody` leaking raw tags when a tag spans multiple lines.
+
+The HTML-tag replacement (e.g. `<Highlighted>…</Highlighted>` to `**…**`, and any custom `htmlTagReplacements`) used a regex without the dotAll flag, so `.` never matched newlines. A tag whose content spanned more than one line was left untouched, leaking raw markup into the cleaned markdown that AI clients read. The regex now uses the `s` flag while keeping the lazy match, so multi-line tags are converted and adjacent tags are not merged.

packages/core/src/text.ts

@@ -70,7 +70,11 @@ export function cleanBody(body: string, opts: CleanBodyOptions = {}): string {
   let out = stripImg ? stripImages(body) : body;
 
   for (const [tag, marker] of Object.entries(replacements)) {
-    const re = new RegExp(`<${tag}>(.*?)<\\/${tag}>`, "g");
+    // `s` (dotAll) so a tag whose content spans multiple lines is still
+    // replaced; without it a multi-line `<Tag>…</Tag>` leaks its raw markup
+    // into the cleaned markdown. `.*?` stays lazy, so it still stops at the
+    // first closing tag.
+    const re = new RegExp(`<${tag}>(.*?)<\\/${tag}>`, "gs");
     out = out.replace(re, `${marker}$1${marker}`);
   }
 

packages/core/test/text.test.ts

@@ -86,6 +86,18 @@ describe("cleanBody", () => {
     expect(cleanBody("see <Highlighted>this</Highlighted>")).toBe("see **this**");
   });
 
+  it("converts a tag whose content spans multiple lines", () => {
+    expect(cleanBody("see <Highlighted>multi\nline</Highlighted> end")).toBe(
+      "see **multi\nline** end",
+    );
+  });
+
+  it("keeps tag replacement lazy across multiple tags", () => {
+    expect(cleanBody("<Em>a</Em> mid <Em>b</Em>", { htmlTagReplacements: { Em: "*" } })).toBe(
+      "*a* mid *b*",
+    );
+  });
+
   it("converts <br> to newline", () => {
     expect(cleanBody("line1<br>line2<br/>line3")).toBe("line1\nline2\nline3");
   });