test(core): unpin evals while keeping integration tests pinned

- Removes hardcoded model from TestRig for unpinned evaluations.\n- Pins integration tests to gemini-2.5-pro via new GEMINI_TEST_TYPE=integration env var.\n- Moves shell efficiency evals to USUALLY_PASSES to prevent PR blocking while tracking Gemini 3.

Author: NTaylorMullen

evals/shell-efficiency.eval.ts

@@ -20,7 +20,7 @@ describe('Shell Efficiency', () => {
     return typeof args === 'string' ? args : (args as any)['command'];
   };
 
-  evalTest('ALWAYS_PASSES', {
+  evalTest('USUALLY_PASSES', {
     name: 'should use --silent/--quiet flags when installing packages',
     prompt: 'Install the "lodash" package using npm.',
     assert: async (rig) => {
@@ -49,7 +49,7 @@ describe('Shell Efficiency', () => {
     },
   });
 
-  evalTest('ALWAYS_PASSES', {
+  evalTest('USUALLY_PASSES', {
     name: 'should use --no-pager with git commands',
     prompt: 'Show the git log.',
     assert: async (rig) => {
@@ -72,7 +72,7 @@ describe('Shell Efficiency', () => {
     },
   });
 
-  evalTest('ALWAYS_PASSES', {
+  evalTest('USUALLY_PASSES', {
     name: 'should NOT use efficiency flags when enableShellOutputEfficiency is disabled',
     params: {
       settings: {

integration-tests/vitest.config.ts

@@ -20,5 +20,8 @@ export default defineConfig({
         maxThreads: 16,
       },
     },
+    env: {
+      GEMINI_TEST_TYPE: 'integration',
+    },
   },
 });

packages/test-utils/src/test-rig.ts

@@ -408,9 +408,13 @@ export class TestRig {
         ui: {
           useAlternateBuffer: true,
         },
-        model: {
-          name: DEFAULT_GEMINI_MODEL,
-        },
+        ...(env['GEMINI_TEST_TYPE'] === 'integration'
+          ? {
+              model: {
+                name: DEFAULT_GEMINI_MODEL,
+              },
+            }
+          : {}),
         sandbox:
           env['GEMINI_SANDBOX'] !== 'false' ? env['GEMINI_SANDBOX'] : false,
         // Don't show the IDE connection dialog when running from VsCode