fix(gateway): resolve dynamic models during warmup

src/gateway/server-startup-post-attach.ts

@@ -86,7 +86,7 @@ async function prewarmConfiguredPrimaryModel(params: {
     { selectAgentHarness },
     { isCliProvider, resolveConfiguredModelRef },
     { ensureOpenClawModelsJson },
-    { resolveModel },
+    { resolveModel, resolveModelAsync },
     { resolveEmbeddedAgentRuntime },
   ] = await Promise.all([
     import("../agents/agent-paths.js"),
@@ -119,10 +119,12 @@ async function prewarmConfiguredPrimaryModel(params: {
       skipProviderRuntimeHooks: true,
     });
     if (!resolved.model) {
-      throw new Error(
-        resolved.error ??
-          `Unknown model: ${provider}/${model} (startup warmup only checks static model resolution)`,
-      );
+      const asyncResolved = await resolveModelAsync(provider, model, agentDir, params.cfg);
+      if (!asyncResolved.model) {
+        throw new Error(
+          resolved.error ?? asyncResolved.error ?? `Unknown model: ${provider}/${model}`,
+        );
+      }
     }
   } catch (err) {
     params.log.warn(`startup model warmup failed for ${provider}/${model}: ${String(err)}`);

src/gateway/server-startup.test.ts

@@ -19,6 +19,20 @@ const resolveModelMock = vi.fn<
     api: "openai-codex-responses",
   },
 }));
+const resolveModelAsyncMock = vi.fn<
+  (
+    provider: unknown,
+    modelId: unknown,
+    agentDir: unknown,
+    cfg: unknown,
+  ) => Promise<{ model?: { id: string; provider: string; api: string }; error?: string }>
+>(async () => ({
+  model: {
+    id: "gpt-5.4",
+    provider: "openai-codex",
+    api: "openai-codex-responses",
+  },
+}));
 const selectAgentHarnessMock = vi.fn((_params: unknown) => ({ id: "pi" }));
 const resolveEmbeddedAgentRuntimeMock = vi.fn(() => "auto");
 
@@ -43,6 +57,8 @@ vi.mock("../agents/pi-embedded-runner/model.js", () => ({
     cfg: unknown,
     options?: unknown,
   ) => resolveModelMock(provider, modelId, agentDir, cfg, options),
+  resolveModelAsync: (provider: unknown, modelId: unknown, agentDir: unknown, cfg: unknown) =>
+    resolveModelAsyncMock(provider, modelId, agentDir, cfg),
 }));
 
 vi.mock("../agents/pi-embedded-runner/runtime.js", () => ({
@@ -61,6 +77,7 @@ describe("gateway startup primary model warmup", () => {
   beforeEach(() => {
     ensureOpenClawModelsJsonMock.mockClear();
     resolveModelMock.mockClear();
+    resolveModelAsyncMock.mockClear();
     selectAgentHarnessMock.mockClear();
     selectAgentHarnessMock.mockReturnValue({ id: "pi" });
     resolveEmbeddedAgentRuntimeMock.mockClear();
@@ -194,4 +211,53 @@ describe("gateway startup primary model warmup", () => {
     expect(ensureOpenClawModelsJsonMock).toHaveBeenCalledWith(cfg, "/tmp/agent");
     expect(resolveModelMock).toHaveBeenCalled();
   });
+
+  it("falls back to async model resolution before warning", async () => {
+    resolveModelMock.mockReturnValueOnce({ model: undefined } as never);
+    resolveModelAsyncMock.mockResolvedValueOnce({
+      model: {
+        id: "gpt-5.4",
+        provider: "codex",
+        api: "openai-codex-responses",
+      },
+    });
+    const warn = vi.fn();
+    const cfg = {
+      agents: {
+        defaults: {
+          model: {
+            primary: "codex/gpt-5.4",
+          },
+        },
+      },
+    } as OpenClawConfig;
+
+    await prewarmConfiguredPrimaryModel({ cfg, log: { warn } });
+
+    expect(resolveModelAsyncMock).toHaveBeenCalledWith("codex", "gpt-5.4", "/tmp/agent", cfg);
+    expect(warn).not.toHaveBeenCalled();
+  });
+
+  it("warns only when both static and async model resolution miss", async () => {
+    resolveModelMock.mockReturnValueOnce({ model: undefined, error: "static miss" } as never);
+    resolveModelAsyncMock.mockResolvedValueOnce({ error: "async miss" });
+    const warn = vi.fn();
+
+    await prewarmConfiguredPrimaryModel({
+      cfg: {
+        agents: {
+          defaults: {
+            model: {
+              primary: "codex/gpt-5.4",
+            },
+          },
+        },
+      } as OpenClawConfig,
+      log: { warn },
+    });
+
+    expect(warn).toHaveBeenCalledWith(
+      expect.stringContaining("startup model warmup failed for codex/gpt-5.4"),
+    );
+  });
 });