Address latest Codex findings: stale detection, model swap, cache

- Clear detectedAgents (and skip the network call entirely) when the panel
  leaves a loopback base, instead of leaving a previous loopback detection
  result marked 'installed' for a command that now targets a LAN/tunnel/
  remote host.
- Add a separate, network-free correction effect keyed on the live
  activeGgufVariant: if codex was auto-picked while a GGUF model was loaded
  and the user then switches to a transformers-backed model while this panel
  stays mounted, steer away from codex instead of leaving a command that
  unsloth_cli's _require_gguf_for_codex will now reject. Never touches a
  manual pick.
- Drop coding-agents.ts's module-lifetime cache. Installed-CLI detection is
  environment state, not a persisted setting, so a stale positive/negative
  from before the user installed something (or reopened the tab) is worse
  than one extra cheap local API call per mount; keep only the in-flight
  de-dupe for concurrent callers.

Verified the correction-effect logic (gguf->non-gguf swap with/without a
fallback, still-gguf no-op, manual pick never overridden) with a standalone
port of the effect.

studio/frontend/src/features/settings/api/coding-agents.ts

@@ -16,7 +16,11 @@ type ApiCodingAgentsInfo = {
   detected: string[];
 };
 
-let cachedInfo: CodingAgentsInfo | null = null;
+// Which CLIs are on PATH is environment state, not a persisted setting -- it
+// can change any time the user installs something new, so this only
+// de-duplicates concurrent in-flight calls (e.g. React strict-mode's double
+// mount) rather than caching the result across the module's lifetime. Every
+// fresh call (each time a settings panel mounts) re-checks PATH for real.
 let inFlightInfo: Promise<CodingAgentsInfo> | null = null;
 
 function fromApi(info: ApiCodingAgentsInfo): CodingAgentsInfo {
@@ -34,16 +38,8 @@ async function fetchCodingAgents(): Promise<CodingAgentsInfo> {
 }
 
 export async function loadCodingAgents(): Promise<CodingAgentsInfo> {
-  if (cachedInfo) {
-    return cachedInfo;
-  }
-  inFlightInfo ??= fetchCodingAgents()
-    .then((info) => {
-      cachedInfo = info;
-      return info;
-    })
-    .finally(() => {
-      inFlightInfo = null;
-    });
+  inFlightInfo ??= fetchCodingAgents().finally(() => {
+    inFlightInfo = null;
+  });
   return inFlightInfo;
 }

studio/frontend/src/features/settings/components/usage-examples.tsx

@@ -501,19 +501,24 @@ export function UsageExamples({ apiKey }: { apiKey?: string | null }) {
   }, []);
 
   useEffect(() => {
+    // shutil.which runs on the Studio backend, so "detected" only means
+    // something for the browser's own machine when the base this panel
+    // targets is loopback; for a LAN/tunnel/remote base the server's
+    // installed CLIs describe a different machine. Skip the request
+    // entirely and clear out any stale result from a previous loopback
+    // base (e.g. the Cloudflare URL arriving after mount, or the user
+    // flipping Secure HTTPS/tunnel) instead of leaving old agents marked
+    // "detected" for a command that now targets somewhere else.
+    if (!isLoopbackBase) {
+      setDetectedAgents([]);
+      return;
+    }
+
     let cancelled = false;
     void loadCodingAgents()
       .then((info) => {
         if (cancelled) return;
         setAvailableAgents(info.agents);
-
-        // shutil.which runs on the Studio backend, so "detected" only means
-        // something for the browser's own machine when the base this panel
-        // targets is loopback; for a LAN/tunnel/remote base the server's
-        // installed CLIs describe a different machine, so don't mark
-        // anything as detected or let it drive the default.
-        if (!isLoopbackBase) return;
-
         setDetectedAgents(info.detected);
         // Prefer an agent that's actually installed over the hardcoded
         // "claude" starting point, but never override a choice the user
@@ -538,6 +543,20 @@ export function UsageExamples({ apiKey }: { apiKey?: string | null }) {
     };
   }, [isLoopbackBase]);
 
+  // The effect above only re-evaluates codex's GGUF requirement at the
+  // moment detection resolves; if the user swaps to a non-GGUF model while
+  // this panel stays mounted, steer the auto-pick away from codex instead of
+  // leaving a command that unsloth_cli's _require_gguf_for_codex will now
+  // reject. No network call here -- it only re-derives from state already in
+  // hand, and it never overrides a choice the user made by hand.
+  const activeGgufVariant = useChatRuntimeStore((s) => s.activeGgufVariant);
+  useEffect(() => {
+    if (agentPickedByUserRef.current) return;
+    if (agent !== "codex" || activeGgufVariant) return;
+    const fallback = detectedAgents.find((id) => id !== "codex");
+    if (fallback) setAgent(fallback);
+  }, [agent, activeGgufVariant, detectedAgents]);
+
   useEffect(() => {
     let cancelled = false;
     void loadOpenAIAutoSwitchSettings()