strip URL fields from $set / $set_once

src/analytics/PosthogAnalytics.test.ts

@@ -138,6 +138,65 @@ describe("PosthogAnalytics", () => {
     it("passes null events through unchanged", () => {
       expect(applyPrivacyFilters(null, Anonymity.Pseudonymous)).toBeNull();
     });
+
+    it("strips URL fields nested inside $set_once", () => {
+      const secretUrl =
+        "https://call.example.com/room/#/?password=hunter2&roomId=abc";
+      const out = applyPrivacyFilters(
+        makeEvent({
+          $current_url: "https://call.example.com/x",
+          $set_once: {
+            $current_url: secretUrl,
+            $initial_current_url: secretUrl,
+            $session_entry_url: secretUrl,
+            $initial_person_info: { r: "x", u: secretUrl },
+          },
+        }),
+        Anonymity.Pseudonymous,
+      );
+
+      const setOnce = out?.properties["$set_once"] as Record<string, unknown>;
+      expect(setOnce["$current_url"]).not.toContain("password");
+      expect(setOnce["$initial_current_url"]).not.toContain("password");
+      expect(setOnce).not.toHaveProperty("$session_entry_url");
+      expect(setOnce).not.toHaveProperty("$initial_person_info");
+    });
+
+    it("strips URL fields nested inside $set", () => {
+      const secretUrl =
+        "https://call.example.com/room/#/?password=hunter2&roomId=abc";
+      const out = applyPrivacyFilters(
+        makeEvent({
+          $current_url: "https://call.example.com/x",
+          $set: {
+            $current_url: secretUrl,
+            $session_entry_url: secretUrl,
+          },
+        }),
+        Anonymity.Pseudonymous,
+      );
+
+      const set = out?.properties["$set"] as Record<string, unknown>;
+      expect(set["$current_url"]).not.toContain("password");
+      expect(set).not.toHaveProperty("$session_entry_url");
+    });
+
+    it("nulls referrer fields inside $set_once when anonymous", () => {
+      const out = applyPrivacyFilters(
+        makeEvent({
+          $current_url: "https://x/y",
+          $set_once: {
+            $initial_referrer: "https://leaky",
+            $initial_referring_domain: "leaky",
+          },
+        }),
+        Anonymity.Anonymous,
+      );
+
+      const setOnce = out?.properties["$set_once"] as Record<string, unknown>;
+      expect(setOnce["$initial_referrer"]).toBeNull();
+      expect(setOnce["$initial_referring_domain"]).toBeNull();
+    });
   });
 
   // Verifies that applyPrivacyFilters is actually wired into posthog.init via

src/analytics/PosthogAnalytics.ts

@@ -66,9 +66,47 @@ export enum RegistrationType {
   Registered,
 }
 
+// Sanitize URL / referrer / device fields on a single posthog properties bag.
+// Applied to event.properties and to the person-profile bags ($set / $set_once),
+// since posthog mirrors the same URL fields into those.
+function stripPrivacyFields(
+  obj: Properties | undefined,
+  anonymity: Anonymity,
+): void {
+  if (!obj) return;
+
+  if (anonymity === Anonymity.Anonymous) {
+    // drop referrer information for anonymous users
+    obj["$referrer"] = null;
+    obj["$referring_domain"] = null;
+    obj["$initial_referrer"] = null;
+    obj["$initial_referring_domain"] = null;
+
+    // drop device ID, which is a UUID persisted in local storage
+    obj["$device_id"] = null;
+  }
+
+  // the url leaks a lot of private data like the call name or the user
+  // (room password / room ID can land in the hash/query). Strip down to
+  // scheme + host so we still get host-level insights (develop / main / sfu).
+  for (const key of ["$current_url", "$initial_current_url"]) {
+    if (typeof obj[key] === "string") {
+      obj[key] = (obj[key] as string).split("/").slice(0, 3).join("");
+    }
+  }
+
+  // $session_entry_url carries the full untrimmed URL; $initial_person_info
+  // bundles initial referrer + URL into a nested object that bypasses the
+  // per-key strips above. Drop both.
+  delete obj["$session_entry_url"];
+  delete obj["$initial_person_info"];
+}
+
 /**
  * Strip PII from posthog's built-in properties (URL, referrer fields,
- * device ID, $initial_person_info) before events leave the client.
+ * device ID, $initial_person_info, $session_entry_url) before events leave
+ * the client. Also applied to the person-profile bags ($set / $set_once),
+ * which mirror the same URL fields.
  * See src/utils/event-utils.ts in posthog-js (getEventProperties, getPersonInfo)
  * for the list of properties posthog sets automatically.
  */
@@ -77,30 +115,15 @@ export function applyPrivacyFilters(
   anonymity: Anonymity,
 ): CaptureResult | null {
   if (event === null) return null;
-  const properties = event.properties;
 
-  if (anonymity === Anonymity.Anonymous) {
+  stripPrivacyFields(event.properties, anonymity);
-    // drop referrer information for anonymous users
+  // posthog can stash person-profile updates either at the top level
-    properties["$referrer"] = null;
+  // of CaptureResult or nested inside properties depending on the pipeline
-    properties["$referring_domain"] = null;
+  // stage; clean both spots so nothing slips through.
-    properties["$initial_referrer"] = null;
+  stripPrivacyFields(event.$set, anonymity);
-    properties["$initial_referring_domain"] = null;
+  stripPrivacyFields(event.$set_once, anonymity);
-
+  stripPrivacyFields(event.properties["$set"], anonymity);
-    // drop device ID, which is a UUID persisted in local storage
+  stripPrivacyFields(event.properties["$set_once"], anonymity);
-    properties["$device_id"] = null;
-  }
-
-  // the url leaks a lot of private data like the call name or the user.
-  // Its stripped down to the bare minimum to only give insights about the host (develop, main or sfu)
-  if (typeof properties["$current_url"] === "string") {
-    properties["$current_url"] = properties["$current_url"]
-      .split("/")
-      .slice(0, 3)
-      .join("");
-  }
-
-  // drop $initial_person_info for increased privacy.
-  delete properties["$initial_person_info"];
 
   return event;
 }