diff --git a/packages/domain/src/index.ts b/packages/domain/src/index.ts
index 500a18a5d..20008d2be 100644
--- a/packages/domain/src/index.ts
+++ b/packages/domain/src/index.ts
@@ -68,17 +68,20 @@ export interface BrowserSession {
   previewUrl?: string;
   consoleUrl?: string;
 }
+// Invisible characters that reorder or hide text without showing on the review card. Joiners and
+// directional marks stay because emoji sequences, Persian and Arabic text depend on them.
+const hiddenText = /[\u200B\u202A-\u202E\u2060-\u2064\u2066-\u2069\uFEFF\u{E0000}-\u{E007F}]/gu;
+const visibleText = () => z.string().overwrite((text) => text.replace(hiddenText, ""));
 export const emailDraftSchema = z.object({
   to: z.array(z.email()).min(1).max(50),
   cc: z.array(z.email()).max(50).default([]),
   bcc: z.array(z.email()).max(50).default([]),
-  subject: z
-    .string()
+  subject: visibleText()
     .trim()
     .min(1)
     .max(998)
     .refine((s) => !/[\r\n]/.test(s), "Subject must be a single line"),
-  body: z.string().min(1).max(100000),
+  body: visibleText().min(1).max(100000),
   attachmentIds: z.array(z.string()).max(10).default([]),
   threadId: z.string().optional(),
   replyToMessageId: z.string().optional(),
@@ -86,13 +89,13 @@ export const emailDraftSchema = z.object({
 export const eventDraftSchema = z
   .object({
     calendarId: z.string().default("primary"),
-    title: z.string().trim().min(1).max(500),
+    title: visibleText().trim().min(1).max(500),
     start: z.string().min(1),
     end: z.string().min(1),
     allDay: z.boolean().default(false),
     timeZone: z.string().default("America/Los_Angeles"),
-    location: z.string().max(2000).default(""),
-    description: z.string().max(10000).default(""),
+    location: visibleText().max(2000).default(""),
+    description: visibleText().max(10000).default(""),
     attendees: z.array(z.email()).max(50).default([]),
   })
   .superRefine((value, ctx) => {
diff --git a/packages/integrations/src/pdf.ts b/packages/integrations/src/pdf.ts
index a20430c1e..385318873 100644
--- a/packages/integrations/src/pdf.ts
+++ b/packages/integrations/src/pdf.ts
@@ -1,6 +1,9 @@
+import { AsyncLocalStorage } from "node:async_hooks";
 import {
+  decodePDFRawStream,
   PDFArray,
   PDFCheckBox,
+  PDFContext,
   PDFDict,
   PDFDocument,
   PDFDropdown,
@@ -10,6 +13,7 @@ import {
   type PDFObject,
   PDFOptionList,
   PDFRadioGroup,
+  PDFRawStream,
   PDFString,
   PDFTextField,
   rgb,
@@ -18,6 +22,7 @@ import {
 
 const MAX_PDF_BYTES = 10 * 1024 * 1024;
 const MAX_PDF_PAGES = 500;
+const MAX_DECODED_BYTES = 10 * MAX_PDF_BYTES;
 
 export class PdfError extends Error {
   readonly status = 422;
@@ -41,6 +46,30 @@ export interface PdfInspection {
   fields: { name: string; value: string; type: "text" | "checkbox" | "unsupported" }[];
 }
 
+// pdf-lib decompresses object and cross-reference streams while loading, with no output limit,
+// so a file under MAX_PDF_BYTES could still expand without bound. Every pdf-lib decoder grows its
+// output buffer through DecodeStream.ensureBuffer, so cap what one load may allocate there.
+const decodeBudget = new AsyncLocalStorage<{ remaining: number }>();
+let decodeStream = Object.getPrototypeOf(
+  decodePDFRawStream(
+    PDFRawStream.of(PDFContext.create().obj({ Filter: "ASCIIHexDecode" }), new Uint8Array()),
+  ),
+);
+while (decodeStream && !Object.hasOwn(decodeStream, "ensureBuffer"))
+  decodeStream = Object.getPrototypeOf(decodeStream);
+if (!decodeStream) throw new Error("pdf-lib no longer decodes through DecodeStream");
+const ensureBuffer = decodeStream.ensureBuffer;
+decodeStream.ensureBuffer = function (this: { buffer: Uint8Array }, requested: number) {
+  const before = this.buffer.byteLength;
+  const buffer: Uint8Array = ensureBuffer.call(this, requested);
+  const budget = decodeBudget.getStore();
+  if (budget) {
+    budget.remaining -= buffer.byteLength - before;
+    if (budget.remaining < 0) throw new PdfError("PDF content expands beyond the supported size");
+  }
+  return buffer;
+};
+
 async function loadPdf(bytes: Uint8Array): Promise<PDFDocument> {
   if (
     !bytes.length ||
@@ -49,12 +78,16 @@ async function loadPdf(bytes: Uint8Array): Promise<PDFDocument> {
   ) {
     throw new PdfError("Invalid PDF: expected nonempty PDF bytes up to 10 MiB");
   }
+  const budget = { remaining: MAX_DECODED_BYTES };
   try {
-    const doc = await PDFDocument.load(new Uint8Array(bytes), {
-      ignoreEncryption: false,
-      throwOnInvalidObject: true,
-      updateMetadata: false,
-    });
+    const doc = await decodeBudget.run(budget, () =>
+      PDFDocument.load(new Uint8Array(bytes), {
+        ignoreEncryption: false,
+        throwOnInvalidObject: true,
+        updateMetadata: false,
+      }),
+    );
+    if (budget.remaining < 0) throw new PdfError("PDF content expands beyond the supported size");
     if (doc.isEncrypted) throw new PdfError("Encrypted PDFs are not supported");
     if (doc.catalog.lookupMaybe(PDFName.of("AcroForm"), PDFDict)?.has(PDFName.of("XFA"))) {
       throw new PdfError("Unsupported XFA PDF form");
@@ -63,6 +96,8 @@ async function loadPdf(bytes: Uint8Array): Promise<PDFDocument> {
       throw new PdfError("PDF must contain between 1 and 500 pages");
     return doc;
   } catch (error) {
+    // pdf-lib's parser can catch the limit error, then succeed or fail with its own error.
+    if (budget.remaining < 0) throw new PdfError("PDF content expands beyond the supported size");
     if (error instanceof PdfError) throw error;
     if (error instanceof Error && /encrypt/i.test(error.message))
       throw new PdfError("Encrypted PDFs are not supported");
diff --git a/tests/domain.test.ts b/tests/domain.test.ts
index 7c78849d8..ac2d22e4b 100644
--- a/tests/domain.test.ts
+++ b/tests/domain.test.ts
@@ -163,3 +163,31 @@ test("proposal schema discriminates on kind and validates nested data", () => {
   );
   assert.equal(proposalSchema.safeParse({ kind: "unknown.kind", data: {} }).success, false);
 });
+
+test("drafts drop invisible characters so the review card shows what is sent", () => {
+  const hidden = "\u200B\u202E\u2066\u2069\u2060\uFEFF\u{E0041}";
+  const draft = emailDraftSchema.parse({
+    ...email,
+    subject: `Your${hidden} visit`,
+    body: `See you${hidden} soon.`,
+  });
+  assert.equal(draft.subject, "Your visit");
+  assert.equal(draft.body, "See you soon.");
+  const created = eventDraftSchema.parse({
+    ...event,
+    title: `Museum${hidden} visit`,
+    location: `Hall${hidden} A`,
+    description: `Bring${hidden} tickets`,
+  });
+  assert.deepEqual(
+    [created.title, created.location, created.description],
+    ["Museum visit", "Hall A", "Bring tickets"],
+  );
+  // A subject made only of invisible characters is empty once they are dropped.
+  assert.equal(emailDraftSchema.safeParse({ ...email, subject: hidden }).success, false);
+});
+
+test("drafts keep joiners and directional marks that real text uses", () => {
+  const body = "Family 👨\u200D👩\u200D👧, می\u200Cخواهم, \u200Fשלום\u200E";
+  assert.equal(emailDraftSchema.parse({ ...email, body }).body, body);
+});
diff --git a/tests/pdf.test.ts b/tests/pdf.test.ts
index 50bfb2591..f81a6413c 100644
--- a/tests/pdf.test.ts
+++ b/tests/pdf.test.ts
@@ -1,5 +1,6 @@
 import assert from "node:assert/strict";
 import test from "node:test";
+import { deflateSync } from "node:zlib";
 import { PDFDocument, PDFName, PDFString } from "pdf-lib";
 import {
   createSamplePdf,
@@ -152,3 +153,34 @@ test("PDF failures expose safe typed 422 errors including malformed field and fo
     (error: unknown) => error instanceof PdfError && /malformed|inspect|field/i.test(error.message),
   );
 });
+
+test("PDFs whose compressed content expands past the decode limit are rejected", async () => {
+  // A valid one-page PDF plus an object stream that decompresses to just over 100 MiB.
+  const content = deflateSync(
+    Buffer.concat([
+      Buffer.from("5 0 "),
+      Buffer.alloc(100 * 1024 * 1024, 0x20),
+      Buffer.from("null"),
+    ]),
+  );
+  const oversized = Buffer.concat([
+    Buffer.from(
+      "%PDF-1.5\n1 0 obj\n<< /Type /Catalog /Pages 2 0 R >>\nendobj\n" +
+        "2 0 obj\n<< /Type /Pages /Kids [3 0 R] /Count 1 >>\nendobj\n" +
+        "3 0 obj\n<< /Type /Page /Parent 2 0 R /MediaBox [0 0 10 10] >>\nendobj\n" +
+        `4 0 obj\n<< /Type /ObjStm /N 1 /First 4 /Filter /FlateDecode /Length ${content.length} >>\nstream\n`,
+    ),
+    content,
+    Buffer.from("\nendstream\nendobj\ntrailer\n<< /Root 1 0 R /Size 5 >>\n%%EOF\n"),
+  ]);
+  assert.ok(oversized.length < 10 * 1024 * 1024);
+  const [rejected, sample] = await Promise.allSettled([
+    inspectPdf(oversized),
+    createSamplePdf().then(inspectPdf),
+  ]);
+  assert.equal(rejected.status, "rejected");
+  assert.ok(rejected.reason instanceof PdfError);
+  assert.match(rejected.reason.message, /expands beyond the supported size/);
+  // The limit is per load: a concurrent ordinary PDF is unaffected.
+  assert.equal(sample.status === "fulfilled" && sample.value.pageCount, 2);
+});