diff --git a/src/runtime/cli/pack_command.rs b/src/runtime/cli/pack_command.rs
index a1ce7dd2c0e1..aa7f174c72dc 100644
--- a/src/runtime/cli/pack_command.rs
+++ b/src/runtime/cli/pack_command.rs
@@ -2491,9 +2491,9 @@ pub(crate) fn pack<const FOR_PUBLISH: bool>(
     // default is 9
     // https://github.com/npm/cli/blob/ec105f400281a5bfd17885de1ea3d54d0c231b27/node_modules/pacote/lib/util/tar-create-options.js#L12
     let compression_level: &[u8] = opt_pack_gzip_level(ctx.manager).unwrap_or(b"9");
-    write!(&mut print_buf, "{}\x00", bstr::BStr::new(compression_level)).expect("OOM");
-    // SAFETY: print_buf[compression_level.len()] == 0 written above
-    let level_z = ZStr::from_buf(&print_buf[..], compression_level.len());
+    print_buf.extend_from_slice(compression_level);
+    print_buf.push(0);
+    let level_z = ZStr::from_slice_with_nul(&print_buf[..]);
     match archive.write_set_filter_option(None, zstr_lit(b"compression-level\0"), level_z) {
         ArchiveResult::Failed | ArchiveResult::Fatal | ArchiveResult::Warn => {
             Output::err_generic(
@@ -3197,16 +3197,11 @@ fn add_archive_entry(
 ) -> Result<*mut ArchiveEntry, AllocError> {
     // `entry` is the same pointer after `.clear()`.
     let entry = ArchiveEntry::opaque_ref(entry);
-    write!(
-        print_buf,
-        "{}{}\x00",
-        bstr::BStr::new(PACKAGE_PREFIX),
-        bstr::BStr::new(filename.as_bytes())
-    )
-    .expect("OOM");
-    let pathname_len = PACKAGE_PREFIX.len() + filename.as_bytes().len();
-    // SAFETY: print_buf[pathname_len] == 0 written above
-    let pathname = ZStr::from_buf(&print_buf[..], pathname_len);
+    // Not formatted through `BStr`: its `Display` rewrites bytes that are not valid UTF-8.
+    print_buf.extend_from_slice(PACKAGE_PREFIX);
+    print_buf.extend_from_slice(filename.as_bytes());
+    print_buf.push(0);
+    let pathname = ZStr::from_slice_with_nul(&print_buf[..]);
     #[cfg(windows)]
     entry.set_pathname_utf8(pathname);
     #[cfg(not(windows))]
diff --git a/test/cli/install/bun-pack.test.ts b/test/cli/install/bun-pack.test.ts
index 77fe7a55685a..215551038a07 100644
--- a/test/cli/install/bun-pack.test.ts
+++ b/test/cli/install/bun-pack.test.ts
@@ -2,7 +2,7 @@ import { file, spawn, write } from "bun";
 import { readTarball } from "bun:internal-for-testing";
 import { beforeEach, describe, expect, test } from "bun:test";
 import { exists, mkdir, rm } from "fs/promises";
-import { bunEnv, bunExe, pack, runBunInstall, tempDir, tmpdirSync } from "harness";
+import { bunEnv, bunExe, isLinux, pack, runBunInstall, tempDir, tmpdirSync } from "harness";
 import fs from "node:fs/promises";
 import { join } from "path";
 
@@ -1616,6 +1616,74 @@ test("unicode", async () => {
   expect(tarball.entries).toMatchObject([{ "pathname": "package/package.json" }, { "pathname": "package/äöüščří.js" }]);
 });
 
+// Reads the members of a .tgz without decoding their names: `readTarball` turns
+// names into JS strings, which maps the bytes this test is about onto U+FFFD.
+// Names are returned latin1-decoded (one char per stored byte) and collected
+// from every place the archive spells a name: the ustar header `name` field
+// and, when libarchive emits one, the `path` record of the pax header.
+function rawTarballMembers(tgz: Uint8Array) {
+  const tar = Buffer.from(Bun.gunzipSync(tgz));
+  const members: { names: Set<string>; contents: string }[] = [];
+  let paxPath: string | undefined;
+  for (let offset = 0; offset + 512 <= tar.length; ) {
+    const header = tar.subarray(offset, offset + 512);
+    if (header.every(byte => byte === 0)) break;
+    const nameField = header.subarray(0, 100);
+    const nameLength = nameField.indexOf(0) === -1 ? 100 : nameField.indexOf(0);
+    const name = nameField.toString("latin1", 0, nameLength);
+    const size = parseInt(header.toString("latin1", 124, 136), 8);
+    const typeflag = header.toString("latin1", 156, 157);
+    const data = tar.subarray(offset + 512, offset + 512 + size);
+    offset += 512 + Math.ceil(size / 512) * 512;
+
+    if (typeflag === "x") {
+      // pax extended header data is a sequence of "<record length> <key>=<value>\n"
+      for (let pos = 0; pos < data.length; ) {
+        const space = data.indexOf(" ", pos);
+        const recordLength = parseInt(data.toString("latin1", pos, space), 10);
+        const record = data.toString("latin1", space + 1, pos + recordLength - 1);
+        if (record.startsWith("path=")) paxPath = record.slice("path=".length);
+        pos += recordLength;
+      }
+      continue;
+    }
+
+    const names = new Set([name]);
+    if (paxPath !== undefined) names.add(paxPath);
+    paxPath = undefined;
+    members.push({ names, contents: data.toString("latin1") });
+  }
+  return members.sort((a, b) => (a.contents < b.contents ? -1 : 1));
+}
+
+// Only Linux lets a filename carry bytes that are not valid UTF-8 (APFS rejects
+// them and Windows filenames are UTF-16).
+test.skipIf(!isLinux)("filenames that are not valid UTF-8 are stored byte for byte", async () => {
+  const rawPath = (byte: number) => Buffer.concat([Buffer.from(`${packageDir}/x_`), Buffer.from([byte])]);
+  await Promise.all([
+    write(
+      join(packageDir, "package.json"),
+      JSON.stringify({
+        name: "pack-invalid-utf8",
+        version: "1.0.0",
+      }),
+    ),
+    // latin-1 "é", and a byte that is invalid in UTF-8 at any position
+    fs.writeFile(rawPath(0xe9), "ONE"),
+    fs.writeFile(rawPath(0xff), "TWO"),
+  ]);
+
+  const { out } = await pack(packageDir, bunEnv);
+  expect(out).toContain("Total files: 3");
+
+  const members = rawTarballMembers(await file(join(packageDir, "pack-invalid-utf8-1.0.0.tgz")).bytes());
+  expect(members).toEqual([
+    { names: new Set(["package/x_\xe9"]), contents: "ONE" },
+    { names: new Set(["package/x_\xff"]), contents: "TWO" },
+    { names: new Set(["package/package.json"]), contents: expect.stringContaining(`"pack-invalid-utf8"`) },
+  ]);
+});
+
 test("$npm_command is accurate", async () => {
   await write(
     join(packageDir, "package.json"),