fix(agents): bound OpenRouter model-scan catalog success body

Alix-007 · Alix-007 · commit 285839b1e002 · 2026-06-21T06:29:34.000+08:00
The OpenRouter /models catalog read in fetchOpenRouterModels hardened only the error/early-return path (dbd5689 cancels the body when res.bodyUsed is false), but the success branch still buffered the whole body with an unbounded `await res.json()`. The response is a provider-controlled, runtime-fetched body, so a faulty or hostile provider can stream an effectively unbounded JSON document and exhaust process memory before the parse completes; the finally-cancel is a no-op once .json() has drained. Read the success body through the canonical byte-cap reader (readResponseWithLimit) under a 4 MiB ceiling before JSON.parse, cancelling the stream on overflow and bounding idle stalls with the call's existing timeout. This is the symmetric success-path counterpart to the bounded-stream hardening landed in #95103 (pricing catalog) and #95108 (Anthropic error streams), reusing the same helper rather than a new abstraction.
diff --git a/src/agents/model-scan.test.ts b/src/agents/model-scan.test.ts
@@ -113,6 +113,57 @@ describe("scanOpenRouterModels", () => {
     expect(cancel).toHaveBeenCalledOnce();
   });
 
+  it("bounds an oversized catalog success body and cancels the stream", async () => {
+    // The success body is provider-controlled and runtime-fetched. A faulty or
+    // hostile provider can stream an effectively unbounded JSON document; the
+    // read must stop at the byte cap, cancel the upstream stream, and surface a
+    // clear overflow error instead of buffering the whole payload into memory.
+    const cancel = vi.fn(async () => undefined);
+    let pullCount = 0;
+    const chunk = new Uint8Array(64 * 1024).fill(0x20); // 64 KiB of spaces
+    const fetchImpl = withFetchPreconnect(
+      async () =>
+        new Response(
+          new ReadableStream({
+            start(controller) {
+              // Valid JSON array prefix so the body would parse if ever read in full.
+              controller.enqueue(new TextEncoder().encode('{"data":['));
+            },
+            pull(controller) {
+              pullCount += 1;
+              controller.enqueue(chunk);
+            },
+            cancel,
+          }),
+          { status: 200, headers: { "content-type": "application/json" } },
+        ),
+    );
+
+    await expect(scanOpenRouterModels({ fetchImpl, probe: false })).rejects.toThrow(
+      /OpenRouter \/models response too large/,
+    );
+
+    // The reader stopped early instead of draining an unbounded stream, and
+    // cancelled the upstream body once the byte cap was crossed.
+    expect(cancel).toHaveBeenCalledOnce();
+    expect(pullCount).toBeGreaterThan(0);
+    expect(pullCount).toBeLessThan(4 * 1024 * 1024); // nowhere near draining forever
+  });
+
+  it("rejects a malformed catalog success body", async () => {
+    const fetchImpl = withFetchPreconnect(
+      async () =>
+        new Response("not json at all", {
+          status: 200,
+          headers: { "content-type": "application/json" },
+        }),
+    );
+
+    await expect(scanOpenRouterModels({ fetchImpl, probe: false })).rejects.toThrow(
+      /OpenRouter \/models response is malformed JSON/,
+    );
+  });
+
   it("requires an API key when probing", async () => {
     const fetchImpl = createFetchFixture({ data: [] });
     await withEnvAsync({ OPENROUTER_API_KEY: undefined }, async () => {
diff --git a/src/agents/model-scan.ts b/src/agents/model-scan.ts
@@ -1,6 +1,7 @@
 /**
  * Scans remote provider model catalogs for configured providers.
  */
+import { readResponseWithLimit } from "@openclaw/media-core/read-response-with-limit";
 import { normalizeProviderId } from "@openclaw/model-catalog-core/provider-id";
 import {
   asDateTimestampMs,
@@ -25,6 +26,11 @@ import { inferParamBFromIdOrName } from "../shared/model-param-b.js";
 const OPENROUTER_MODELS_URL = "https://openrouter.ai/api/v1/models";
 const DEFAULT_TIMEOUT_MS = 12_000;
 const DEFAULT_CONCURRENCY = 3;
+// The OpenRouter /models catalog is a provider-controlled, runtime-fetched body
+// (already >100 KB and growing). Read it under a byte cap before JSON.parse so a
+// faulty or hostile provider cannot stream an unbounded document and exhaust
+// process memory. 4 MiB matches the cap used for the same endpoint elsewhere.
+const OPENROUTER_MODELS_BODY_MAX_BYTES = 4 * 1024 * 1024;
 
 const BASE_IMAGE_PNG =
   "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mP8/x8AAwMCAO+X3mIAAAAASUVORK5CYII=";
@@ -184,6 +190,26 @@ async function withTimeout<T>(
   }
 }
 
+// Reads the OpenRouter /models success body under a byte cap before JSON.parse.
+// The success path was previously buffered with an unbounded res.json(); a faulty
+// or hostile provider could stream an effectively endless document and exhaust
+// memory. readResponseWithLimit caps the read, cancels the stream on overflow,
+// and bounds idle stalls with the call's existing timeout.
+async function readOpenRouterModelsJson(response: Response, timeoutMs: number): Promise<unknown> {
+  const buffer = await readResponseWithLimit(response, OPENROUTER_MODELS_BODY_MAX_BYTES, {
+    chunkTimeoutMs: timeoutMs,
+    onOverflow: ({ size, maxBytes }) =>
+      new Error(`OpenRouter /models response too large: ${size} bytes (limit ${maxBytes} bytes)`),
+    onIdleTimeout: ({ chunkTimeoutMs }) =>
+      new Error(`OpenRouter /models response stalled after ${chunkTimeoutMs}ms`),
+  });
+  try {
+    return JSON.parse(buffer.toString("utf8")) as unknown;
+  } catch (cause) {
+    throw new Error("OpenRouter /models response is malformed JSON", { cause });
+  }
+}
+
 async function fetchOpenRouterModels(
   fetchImpl: typeof fetch,
   timeoutMs: number,
@@ -199,7 +225,7 @@ async function fetchOpenRouterModels(
     if (!res.ok) {
       throw new Error(`OpenRouter /models failed: HTTP ${res.status}`);
     }
-    const payload = (await res.json()) as { data?: unknown };
+    const payload = (await readOpenRouterModelsJson(res, timeoutMs)) as { data?: unknown };
     const entries = Array.isArray(payload.data) ? payload.data : [];
 
     return entries
