fix: stabilize live probes and docs

src/agents/models.profiles.live.test.ts

@@ -48,6 +48,16 @@ function isGoogleModelNotFoundError(err: unknown): boolean {
   return false;
 }
 
+function isModelNotFoundErrorMessage(raw: string): boolean {
+  const msg = raw.trim();
+  if (!msg) return false;
+  if (/\b404\b/.test(msg) && /not[_-]?found/i.test(msg)) return true;
+  if (/not_found_error/i.test(msg)) return true;
+  if (/model:\s*[a-z0-9._-]+/i.test(msg) && /not[_-]?found/i.test(msg))
+    return true;
+  return false;
+}
+
 describeLive("live models (profile keys)", () => {
   it(
     "completes across configured models",
@@ -187,6 +197,15 @@ describeLive("live models (profile keys)", () => {
             },
           );
 
+          if (res.stopReason === "error") {
+            const msg = res.errorMessage ?? "";
+            if (ALL_MODELS && isModelNotFoundErrorMessage(msg)) {
+              skipped.push({ model: id, reason: msg });
+              continue;
+            }
+            throw new Error(msg || "model returned error with no message");
+          }
+
           const text = res.content
             .filter((block) => block.type === "text")
             .map((block) => block.text.trim())

src/agents/openai-responses.reasoning-replay.test.ts

@@ -52,7 +52,7 @@ function installFailingFetchCapture() {
 }
 
 describe("openai-responses reasoning replay", () => {
-  it("handles tool-call-only turns without requiring reasoning replay", async () => {
+  it("replays reasoning for tool-call-only turns (required by OpenAI)", async () => {
     const cap = installFailingFetchCapture();
     try {
       const model = buildModel();
@@ -141,11 +141,11 @@ describe("openai-responses reasoning replay", () => {
         )
         .filter((t): t is string => typeof t === "string");
 
+      expect(types).toContain("reasoning");
       expect(types).toContain("function_call");
-      const reasoningIndex = types.indexOf("reasoning");
-      if (reasoningIndex !== -1) {
-        expect(reasoningIndex).toBeLessThan(types.indexOf("function_call"));
-      }
+      expect(types.indexOf("reasoning")).toBeLessThan(
+        types.indexOf("function_call"),
+      );
     } finally {
       cap.restore();
     }