test: address PR feedback for evals and nightly workflow

- Add missing expectation for foundKeywords in pr-review.eval.ts
- Improve formatting of timeout error logs in test-rig.ts
- Add comments explaining retry logic for npm installations
- Remove || true from evaluation script in GitHub Action

Author: cocosheng-g

.github/workflows/evals-nightly.yml

@@ -47,10 +47,12 @@ jobs:
           cache: 'npm'
 
       - name: 'Install dependencies'
+        # Retry logic for transient network or package retrieval failures
         run: |
           npm ci || (sleep 10 && npm ci) || (sleep 30 && npm ci)
 
       - name: 'Install Gemini CLI'
+        # Retry logic for transient network or package retrieval failures
         run: |
           npm install -g @google/gemini-cli@latest || (sleep 10 && npm install -g @google/gemini-cli@latest) || (sleep 30 && npm install -g @google/gemini-cli@latest)
 
@@ -62,7 +64,7 @@ jobs:
           GEMINI_MODEL: '${{ matrix.model }}'
         run: |
           BASE_NAME=$(basename "${{ matrix.eval-file }}" .eval.ts)
-          npm run test:evals -- "${{ matrix.eval-file }}" --reporter=json --outputFile="eval-results-${{ matrix.model }}-${BASE_NAME}.json" || true
+          npm run test:evals -- "${{ matrix.eval-file }}" --reporter=json --outputFile="eval-results-${{ matrix.model }}-${BASE_NAME}.json"
 
       - name: 'Upload Results'
         if: 'always()'
@@ -86,6 +88,7 @@ jobs:
           cache: 'npm'
 
       - name: 'Install dependencies'
+        # Retry logic for transient network or package retrieval failures
         run: |
           npm ci || (sleep 10 && npm ci) || (sleep 30 && npm ci)
 

evals/pr-review.eval.ts

@@ -81,6 +81,7 @@ describe('PR Review Workflow', () => {
         }
 
         expect(stdout.length).toBeGreaterThan(0);
+        expect(foundKeywords.length).toBeGreaterThan(0);
       } finally {
         rig.cleanup();
       }

evals/test-rig.ts

@@ -168,7 +168,7 @@ export class TestRig {
         child.kill('SIGKILL');
         reject(
           new Error(
-            `Timeout: Command exceeded 600 seconds. stdout: ${stdout.substring(0, 500)} stderr: ${stderr.substring(0, 500)}`,
+            `Timeout: Command exceeded 600 seconds. stdout:\n${stdout.slice(-500)}\nstderr: \n${stderr.slice(-500)}\n`,
           ),
         );
       }, 600_000);