fix(image-gen): use 64 MiB JSON cap + mode-dependent label

wangmiao0668000666 · claude · wangmiao0668000666 · commit 7f7c2bc1fabe · 2026-06-25T21:34:19.000+08:00
P1: Image responses carry base64-encoded image data that can exceed the
default 16 MiB provider JSON cap (4 images × ~7 MiB base64 each = ~28 MiB).
Pass a 64 MiB cap to readProviderJsonResponse.

P3: Use mode-dependent label for bounded-read errors, matching the pattern
already used by assertOkOrThrowHttpError in the same function.

Co-Authored-By: Claude &lt;noreply@anthropic.com&gt;
diff --git a/src/image-generation/openai-compatible-image-provider.ts b/src/image-generation/openai-compatible-image-provider.ts
@@ -272,7 +272,8 @@ export function createOpenAiCompatibleImageGenerationProvider(
         );
         const parsedBody = await readProviderJsonResponse(
           response,
-          `${options.label} image generation`,
+          mode === "edit" ? `${options.label} image edit` : `${options.label} image generation`,
+          { maxBytes: 64 * 1024 * 1024 },
         );
         const images = parseOpenAiCompatibleImageResponse(parsedBody, {
           ...options.response,
