diff --git a/js/src/utils.ts b/js/src/utils.ts
index 060458e4..a7cd0df0 100644
--- a/js/src/utils.ts
+++ b/js/src/utils.ts
@@ -51,32 +51,54 @@ export function isConnectionClosedError(error: unknown): boolean {
 
 export async function* readLines(stream: ReadableStream<Uint8Array>) {
   const reader = stream.getReader()
-  let buffer = ''
+  const decoder = new TextDecoder()
+  // Text accumulated since the last newline, kept as an array of fragments:
+  // appending is O(1), so total work stays linear even when a single line
+  // spans many chunks (growing a string with `buffer += chunk` re-copies the
+  // whole buffer on every chunk, which is O(n²) and can OOM on large stdout).
+  const pending: string[] = []
 
   try {
     while (true) {
       const { done, value } = await reader.read()
 
       if (value !== undefined) {
-        buffer += new TextDecoder().decode(value)
+        // { stream: true } keeps multi-byte UTF-8 sequences split across
+        // chunk boundaries intact instead of decoding them as U+FFFD.
+        const chunk = decoder.decode(value, { stream: true })
+        let newlineIdx = chunk.indexOf('\n')
+
+        if (newlineIdx === -1) {
+          if (chunk.length > 0) {
+            pending.push(chunk)
+          }
+        } else {
+          pending.push(chunk.slice(0, newlineIdx))
+          yield pending.join('')
+          pending.length = 0
+
+          let start = newlineIdx + 1
+          while ((newlineIdx = chunk.indexOf('\n', start)) !== -1) {
+            yield chunk.slice(start, newlineIdx)
+            start = newlineIdx + 1
+          }
+
+          if (start < chunk.length) {
+            pending.push(chunk.slice(start))
+          }
+        }
       }
 
       if (done) {
-        if (buffer.length > 0) {
-          yield buffer
+        const trailing = decoder.decode()
+        if (trailing) {
+          pending.push(trailing)
+        }
+        if (pending.length > 0) {
+          yield pending.join('')
         }
         break
       }
-
-      let newlineIdx = -1
-
-      do {
-        newlineIdx = buffer.indexOf('\n')
-        if (newlineIdx !== -1) {
-          yield buffer.slice(0, newlineIdx)
-          buffer = buffer.slice(newlineIdx + 1)
-        }
-      } while (newlineIdx !== -1)
     }
   } finally {
     reader.releaseLock()
diff --git a/js/tests/utils.test.ts b/js/tests/utils.test.ts
new file mode 100644
index 00000000..7c2e3c25
--- /dev/null
+++ b/js/tests/utils.test.ts
@@ -0,0 +1,96 @@
+import { expect, test } from 'vitest'
+
+import { readLines } from '../src/utils'
+
+function streamFromChunks(chunks: Uint8Array[]): ReadableStream<Uint8Array> {
+  return new ReadableStream<Uint8Array>({
+    start(controller) {
+      for (const chunk of chunks) {
+        controller.enqueue(chunk)
+      }
+      controller.close()
+    },
+  })
+}
+
+function encodeChunks(chunks: string[]): Uint8Array[] {
+  const encoder = new TextEncoder()
+  return chunks.map((chunk) => encoder.encode(chunk))
+}
+
+async function collect(stream: ReadableStream<Uint8Array>): Promise<string[]> {
+  const lines: string[] = []
+  for await (const line of readLines(stream)) {
+    lines.push(line)
+  }
+  return lines
+}
+
+test('yields lines split across a single chunk', async () => {
+  const lines = await collect(streamFromChunks(encodeChunks(['a\nb\nc\n'])))
+  expect(lines).toEqual(['a', 'b', 'c'])
+})
+
+test('yields trailing text without a final newline', async () => {
+  const lines = await collect(streamFromChunks(encodeChunks(['a\nb'])))
+  expect(lines).toEqual(['a', 'b'])
+})
+
+test('handles a line spanning multiple chunks', async () => {
+  const lines = await collect(
+    streamFromChunks(encodeChunks(['first pa', 'rt', ' of line\nsecond\n']))
+  )
+  expect(lines).toEqual(['first part of line', 'second'])
+})
+
+test('handles empty lines', async () => {
+  const lines = await collect(streamFromChunks(encodeChunks(['a\n\nb\n'])))
+  expect(lines).toEqual(['a', '', 'b'])
+})
+
+test('handles chunk boundaries around newlines', async () => {
+  const lines = await collect(
+    streamFromChunks(encodeChunks(['a', '\n', 'b\n', 'c']))
+  )
+  expect(lines).toEqual(['a', 'b', 'c'])
+})
+
+test('yields nothing for an empty stream', async () => {
+  const lines = await collect(streamFromChunks([]))
+  expect(lines).toEqual([])
+})
+
+test('decodes multi-byte UTF-8 characters split across chunks', async () => {
+  // '🚀' is 4 bytes in UTF-8, so split it in the middle of the sequence
+  const bytes = new TextEncoder().encode('before 🚀 after\n')
+  const splitAt = bytes.indexOf(0xf0) + 2
+  const lines = await collect(
+    streamFromChunks([bytes.slice(0, splitAt), bytes.slice(splitAt)])
+  )
+  expect(lines).toEqual(['before 🚀 after'])
+})
+
+test('decodes an incomplete multi-byte sequence at stream end', async () => {
+  // Stream ends mid-🚀: the decoder should flush a replacement character
+  // instead of silently dropping the bytes
+  const bytes = new TextEncoder().encode('abc🚀')
+  const lines = await collect(
+    streamFromChunks([bytes.slice(0, bytes.length - 2)])
+  )
+  expect(lines).toHaveLength(1)
+  expect(lines[0].startsWith('abc')).toBe(true)
+})
+
+test('handles large single-line output split into many chunks', async () => {
+  // A ~8 MB line delivered in 64 KiB chunks must complete without quadratic
+  // buffer rebuilds (this test hangs for minutes with `buffer += chunk`)
+  const chunkSize = 64 * 1024
+  const chunkCount = 128
+  const chunk = 'x'.repeat(chunkSize)
+  const chunks = encodeChunks([
+    ...Array.from({ length: chunkCount }, () => chunk),
+    '\ntail\n',
+  ])
+  const lines = await collect(streamFromChunks(chunks))
+  expect(lines).toEqual(['x'.repeat(chunkSize * chunkCount), 'tail'])
+})