fix(plugin): show actual configured model in banner (Closes #24)

[WuKongAI-CMU]

Summary

When the onboard config file is not available (e.g. when running inside the sandbox), the plugin banner in `nemoclaw/src/index.ts` hardcodes the model as `nvidia/nemotron-3-super-120b-a12b` and the endpoint as `build.nvidia.com` regardless of what is actually configured in OpenShell.

This PR queries the live OpenShell inference state via `openshell inference get --json` as a fallback before resorting to hardcoded defaults. The probe has a 3-second timeout and falls back cleanly to the existing defaults if anything goes wrong.

Closes #24.

Changes

• `nemoclaw/src/index.ts` — add `probeOpenShellInference()` helper and wire it into banner resolution after onboard config lookup, before hardcoded defaults

Test plan

• Minimal diff, existing defaults preserved as final fallback
• 3s timeout prevents banner from hanging on a slow/broken openshell binary
• No new runtime dependencies (uses stdlib `node:child_process`)

(Resubmitting — prior PR #1816 was auto-closed by the 10-PR-cap check when our open PR count briefly exceeded the limit; we're now under the cap.)

🤖 Generated with Claude Code

Summary by CodeRabbit

• Bug Fixes
  • Improved configuration resolution for model provider, endpoint, and model: the app now prefers onboard settings, probes live inference state to fill missing values, and falls back to defaults for more reliable initialization and registration.
• Tests
  • Added test coverage verifying fallback to probed inference values and confirming probe details are logged during registration.

Signed-off-by: Intern Dev dev@wukongai.io

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 31399376-708a-42a5-b41a-0462f78f38f1

📥 Commits

Reviewing files that changed from the base of the PR and between a9a973e and ea9aa8f.

📒 Files selected for processing (2)

• nemoclaw/src/index.ts
• nemoclaw/src/register.test.ts

🚧 Files skipped from review as they are similar to previous changes (1)

• nemoclaw/src/register.test.ts

---

📝 Walkthrough

Walkthrough

Adds a synchronous probe of OpenShell inference state and updates provider/model resolution so registration uses onboard config first, then the probe, then original hardcoded defaults; tests were added to validate the probe-driven fallback when onboard config is absent.

Changes

Cohort / File(s)	Summary
Index / resolution logic nemoclaw/src/index.ts	Added probeOpenShellInference() (runs openshell inference get --json with a 3s timeout). activeModelEntries() and registeredProviderForConfig() now accept an optional fallback model. register() was reworked to source banner/provider/model from onboard → probe → hardcoded defaults and passes probed model into provider registration.
Tests nemoclaw/src/register.test.ts	Mocked node:child_process.execFileSync, reset mock in beforeEach(), and added a test for the "onboard unavailable" path that forces a probed JSON response and asserts register(api) registers the probed model and logs probed Endpoint/Provider/Model.

Sequence Diagram(s)

sequenceDiagram
    participant Register as Register()
    participant Onboard as OnboardConfig
    participant Probe as OpenShell CLI (execFileSync)
    participant Defaults as HardcodedDefaults
    participant Registry as ProviderRegistry

    Register->>Onboard: read bannerEndpoint / bannerProvider / bannerModel
    alt onboard has all values
        Register->>Registry: register using onboard values
    else missing any value
        Register->>Probe: execFileSync("openshell inference get --json") (3s)
        Probe-->>Register: { endpoint, provider, model } or error
        alt probe supplies missing values
            Register->>Registry: register using onboard + probed values
        else
            Register->>Defaults: use hardcoded defaults
            Defaults-->>Register: { endpoint, provider, model }
            Register->>Registry: register using onboard + defaults
        end
    end

Loading

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Poem

🐰 I probed the shell where models sleep,
I peeked and parsed what secrets keep.
Onboard first, then live we see,
If silence stays, old names agree.
A hop, a log — the banner speaks true. 🥕

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 40.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly summarizes the main change: fixing the display of the actual configured model in the plugin banner instead of a hardcoded default.
Linked Issues check	✅ Passed	The PR implements all acceptance criteria from issue #24: probes actual OpenShell configuration, displays configured provider/model in banner, includes tests for non-default providers, and preserves fallback behavior.
Out of Scope Changes check	✅ Passed	All changes are directly scoped to fixing the banner model display issue: probeOpenShellInference() helper, banner resolution logic updates, and corresponding tests for the new functionality.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 2

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

nemoclaw/src/index.ts (1)

272-275: ⚠️ Potential issue | 🟠 Major

Resolve the effective model before registerProvider().

The probe only updates banner strings after registerProvider() has already been called. When onboardCfg is missing, activeModelEntries() still publishes the baked-in catalog, so the registered provider/UI can disagree with the banner and with the live OpenShell model.

Also applies to: 283-290

🤖 Prompt for AI Agents

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

In `@nemoclaw/src/index.ts` around lines 272 - 275, The provider is being
registered before the effective model/catalog is resolved, so the UI banner and
registered provider can disagree; before calling
api.registerProvider(registeredProviderForConfig(onboardCfg,
providerCredentialEnv)) ensure the effective model is resolved by invoking the
same resolution path used by the probe—e.g., call activeModelEntries() (or
otherwise trigger the probe/model resolution logic) right after
loadOnboardConfig() and before computing registeredProviderForConfig so the
banner/catalog are up-to-date when api.registerProvider runs (also apply this
ordering fix to the similar block around the code at lines 283–290).

🤖 Prompt for all review comments with AI agents

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

Inline comments:
In `@nemoclaw/src/index.ts`:
- Around line 25-40: probeOpenShellInference currently only returns endpoint and
model so callers (banner logic) always fall back to "NVIDIA Endpoints"; modify
probeOpenShellInference to also return the provider string by reading
parsed.provider (e.g., return { provider: parsed.provider ?? parsed.endpoint ??
"", endpoint: parsed.endpoint ?? parsed.provider ?? "", model: parsed.model ??
"" }) and update any banner/display code that uses probeOpenShellInference() to
prefer the returned provider when present (instead of hardcoding "NVIDIA
Endpoints"), e.g., use the provider value to construct the banner label.
- Around line 25-44: Add a Vitest unit test in the existing register.test.ts
that mocks node:child_process execFileSync to return a JSON string with
provider, model, and/or endpoint (e.g.
{"provider":"custom","model":"gpt-x","endpoint":"https://api"}), call
probeOpenShellInference and assert it returns the probed non-default
provider/model/endpoint; also add an integration-style assertion that the code
path which consumes probeOpenShellInference (the registration logic referenced
around probeOpenShellInference and the similar block at lines ~283-290) uses
those values instead of defaults. Use Vitest's vi.mock or vi.spyOn to stub
execFileSync, ensure you restore/reset the mock after the test, and cover both
provider->endpoint fallback and explicit endpoint cases.

---

Outside diff comments:
In `@nemoclaw/src/index.ts`:
- Around line 272-275: The provider is being registered before the effective
model/catalog is resolved, so the UI banner and registered provider can
disagree; before calling
api.registerProvider(registeredProviderForConfig(onboardCfg,
providerCredentialEnv)) ensure the effective model is resolved by invoking the
same resolution path used by the probe—e.g., call activeModelEntries() (or
otherwise trigger the probe/model resolution logic) right after
loadOnboardConfig() and before computing registeredProviderForConfig so the
banner/catalog are up-to-date when api.registerProvider runs (also apply this
ordering fix to the similar block around the code at lines 283–290).

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: cef3faf3-a982-4de0-b744-e941f6c51da9

📥 Commits

Reviewing files that changed from the base of the PR and between d4aac4c and dbd9b9f.

📒 Files selected for processing (1)

• nemoclaw/src/index.ts

[wscurran]

✨
Possibly related open PRs:

• #1816 fix(plugin): show actual configured model in banner (Closes #24)

---

Possibly related open PRs:

• #1816 fix(plugin): show actual configured model in banner (Closes #24)

---

Possibly related open issues:

• #24 Fix hardcoded model identity shown by the in-sandbox plugin/banner so it reflects the actual configured inference provider

[cjagwani]

Good fix @WuKongAI-CMU — the probe fallback chain is clean and the 3-second timeout is well-chosen. One gap to address before merge:

Provider still hardcoded. When onboardCfg is missing (inside the sandbox), bannerProvider falls back to "NVIDIA Endpoints" even if the probe returns Ollama or OpenAI. The model and endpoint are now dynamic but the provider line stays wrong.

CodeRabbit flagged this too — here's the minimal fix:

-function probeOpenShellInference(): { endpoint: string; model: string } {
+function probeOpenShellInference(): { endpoint: string; provider: string; model: string } {
   // ...
   return {
     endpoint: parsed.endpoint ?? parsed.provider ?? "",
+    provider: parsed.provider ?? "",
     model: parsed.model ?? "",
   };
   // catch:
-  return { endpoint: "", model: "" };
+  return { endpoint: "", provider: "", model: "" };
 }

-  const bannerProvider = onboardCfg ? describeOnboardProvider(onboardCfg) : "NVIDIA Endpoints";
+  let bannerProvider = onboardCfg ? describeOnboardProvider(onboardCfg) : "";

-  if (!bannerEndpoint || !bannerModel) {
+  if (!bannerEndpoint || !bannerProvider || !bannerModel) {
     const probed = probeOpenShellInference();
     if (!bannerEndpoint) bannerEndpoint = probed.endpoint;
+    if (!bannerProvider) bannerProvider = probed.provider;
     if (!bannerModel) bannerModel = probed.model;
   }
+  if (!bannerProvider) bannerProvider = "NVIDIA Endpoints";

This way Ollama users see "Ollama" instead of "NVIDIA Endpoints" in the banner.

Also: DCO — add Signed-off-by: Your Name <your@email.com> to the PR description body.

[WuKongAI-CMU]

Confirmed the requested follow-up is already in the latest head (a9a973eb): probeOpenShellInference() now returns provider, the banner uses the probed provider before falling back to NVIDIA Endpoints, and the registered provider catalog uses the probed model when onboard config is unavailable. The PR body also includes Signed-off-by: Intern Dev <dev@wukongai.io>.

Verification:

• npm test -- nemoclaw/src/register.test.ts -> 1 file passed, 8 tests passed
• cd nemoclaw && npm run build
• git diff --check

[cv]

Looks good to me.

• banner now prefers onboard config, then live openshell inference get --json, then hardcoded defaults
• registered model catalog also reflects the probed active model when onboard config is unavailable
• regression coverage added for a non-default provider/model path

I also re-ran a local targeted check on the PR branch:

• npx vitest run nemoclaw/src/register.test.ts
• cd nemoclaw && npm run check