test/onboard-selection.test.ts (1)

680-683: ⚠️ Potential issue | 🟡 Minor

Update stale localhost mock literals in this test case.

Line 680 and Line 682 still match localhost, but the local inference commands now target 127.0.0.1. This can miss runCapture matches and make this test path less reliable.

✅ Suggested patch

-  if (command.includes("localhost:11434/api/tags")) return JSON.stringify({ models: [{ name: "nemotron-3-nano:30b" }] });
+  if (command.includes("127.0.0.1:11434/api/tags")) return JSON.stringify({ models: [{ name: "nemotron-3-nano:30b" }] });
   if (command.includes("ollama list")) return "nemotron-3-nano:30b  abc  24 GB  now";
-  if (command.includes("localhost:8000/v1/models")) return "";
+  if (command.includes("127.0.0.1:8000/v1/models")) return "";

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@test/onboard-selection.test.ts` around lines 680 - 683, Update the two stale
localhost mocks so the test matches the current local-inference addresses:
change the checks that call command.includes("localhost:11434/api/tags") and
command.includes("localhost:8000/v1/models") to recognize 127.0.0.1 (either by
replacing the literal or adding an OR check to also match
"127.0.0.1:11434/api/tags" and "127.0.0.1:8000/v1/models"); this ensures
runCapture-based matching (and the if branches that return model/tag mocks)
still fire for local inference.

src/lib/local-inference.ts (1)

162-208: ⚠️ Potential issue | 🟠 Major

Probe target and failure messages are inconsistent.

validateLocalProvider() (Line 162) still executes a health check built from getLocalProviderHealthEndpoint(), which currently returns localhost (Line 68, Line 70), but the new messages on Line 173, Line 179, Line 201, and Line 207 claim checks against 127.0.0.1. This can produce false negatives and misleading remediation text.

🔧 Proposed fix

 export function getLocalProviderHealthEndpoint(provider: string): string | null {
   switch (provider) {
     case "vllm-local":
-      return "http://localhost:8000/v1/models";
+      return "http://127.0.0.1:8000/v1/models";
     case "ollama-local":
-      return "http://localhost:11434/api/tags";
+      return "http://127.0.0.1:11434/api/tags";
     default:
       return null;
   }
 }

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/lib/local-inference.ts` around lines 162 - 208, The health-check messages
in validateLocalProvider() are inconsistent with the actual probe target
returned by getLocalProviderHealthEndpoint(); update validateLocalProvider() to
derive the host/URL from getLocalProviderHealthEndpoint(provider) (or from
getLocalProviderHealthCheck(provider)'s command) and interpolate that value into
the failure messages for vllm-local and ollama-local instead of hardcoding
"127.0.0.1" (or alternatively change getLocalProviderHealthEndpoint() to return
127.0.0.1 if you intend to probe that address); adjust the
container-reachability messages similarly so they reference the same computed
host/URL used by getLocalProviderContainerReachabilityCheck(provider) and
runCapture.