feat(labs): unlock native Codex vision for lab OCR

Image input rides the ChatGPT-OAuth codex/responses endpoint via the
input_image content block (protocol spec 2b) — no vision API key. The
codex client folds params.images into input_image blocks alongside the
input_text; text-only calls keep the prior byte-identical wire. The
vision gate now resolves codex capability against the working slug
(GPT-5.x / gpt-4o-class multimodal) instead of a blanket false, with the
OCR capability resolver feeding the codex slug rather than the user's
aiModel. A non-multimodal slug or an image-block rejection degrades to a
clean extract failure / local-OCR text mode.

src/lib/ai/__tests__/codex-client.test.ts

@@ -196,4 +196,66 @@ describe("CodexClient", () => {
     ).rejects.toThrow("Codex request failed (500)");
     expect(onRefresh).not.toHaveBeenCalled();
   });
+
+  it("emits input_image content blocks when images are present", async () => {
+    const mockFetch = vi
+      .fn()
+      .mockResolvedValue(
+        sseResponse([messageDoneEvent('{"summary":"ok"}'), completedEvent(50)]),
+      );
+    vi.stubGlobal("fetch", mockFetch);
+
+    const client = new CodexClient({
+      accessToken: "test-token",
+      accountId: "acct-test",
+      onTokenRefresh: vi
+        .fn()
+        .mockResolvedValue({ accessToken: "x", accountId: "acct-test" }),
+    });
+
+    await client.generateCompletion({
+      systemPrompt: "Transcribe.",
+      userPrompt: "Read this report.",
+      images: [{ mediaType: "image/png", dataBase64: "QUJD" }],
+    });
+
+    const body = JSON.parse(mockFetch.mock.calls[0][1].body);
+    const content = body.input[0].content;
+    expect(content[0]).toEqual({
+      type: "input_text",
+      text: "Read this report.",
+    });
+    expect(content[1]).toEqual({
+      type: "input_image",
+      image_url: "data:image/png;base64,QUJD",
+      detail: "high",
+    });
+  });
+
+  it("keeps the text-only content shape when no images are present", async () => {
+    const mockFetch = vi
+      .fn()
+      .mockResolvedValue(
+        sseResponse([messageDoneEvent('{"summary":"ok"}'), completedEvent(50)]),
+      );
+    vi.stubGlobal("fetch", mockFetch);
+
+    const client = new CodexClient({
+      accessToken: "test-token",
+      accountId: "acct-test",
+      onTokenRefresh: vi
+        .fn()
+        .mockResolvedValue({ accessToken: "x", accountId: "acct-test" }),
+    });
+
+    await client.generateCompletion({
+      systemPrompt: "Answer.",
+      userPrompt: "Plain text only.",
+    });
+
+    const body = JSON.parse(mockFetch.mock.calls[0][1].body);
+    expect(body.input[0].content).toEqual([
+      { type: "input_text", text: "Plain text only." },
+    ]);
+  });
 });

src/lib/ai/__tests__/vision-capability.test.ts

@@ -54,8 +54,24 @@ describe("supportsVisionForConfig", () => {
     expect(supportsVisionForConfig("local", null)).toBe(true);
   });
 
-  it("never reports vision for codex or none", () => {
-    expect(supportsVisionForConfig("codex", "gpt-4o")).toBe(false);
+  it("reports vision for codex only on multimodal slugs", () => {
+    for (const model of [
+      "gpt-5.5",
+      "gpt-5.4",
+      "gpt-5.4-mini",
+      "gpt-4o",
+      "gpt-4.1",
+      "o1",
+    ]) {
+      expect(supportsVisionForConfig("codex", model)).toBe(true);
+    }
+    // Non-multimodal / unknown slugs and a missing slug → false.
+    expect(supportsVisionForConfig("codex", "gpt-4")).toBe(false);
+    expect(supportsVisionForConfig("codex", "gpt-3.5-turbo")).toBe(false);
+    expect(supportsVisionForConfig("codex", null)).toBe(false);
+  });
+
+  it("never reports vision for none", () => {
     expect(supportsVisionForConfig("none", "gpt-4o")).toBe(false);
   });
 });