ci: implement multi-model evaluation support by replacing static run configs with dynamic discovery and adding Claude/Gemini-specific configurations.

Author: omkargaikwad23

cloudbuild.yaml

@@ -45,8 +45,8 @@ steps:
         PR_TITLE=$(echo "$$PR_DATA" | jq -r '.title')
 
         # Check if execution labels are present using exact matching via jq
-        if ! jq -e '.labels | any(.name == "autorelease: pending" or .name == "ci:run-evals")' pr_data.json > /dev/null; then
-          echo "PR does not have 'autorelease: pending' or 'ci:run-evals' label. Skipping execution."
+        if ! jq -e '.labels | any(.name == "autorelease: pending" or .name == "ci:run-evals" or .name == "ci:run-evals-gemini" or .name == "ci:run-evals-claude")' pr_data.json > /dev/null; then
+          echo "PR does not have required labels. Skipping execution."
           exit 0
         fi
         echo "Execution label detected. Processing release version context..."
@@ -72,7 +72,6 @@ steps:
         export GOOGLE_CLOUD_PROJECT=$PROJECT_ID
         export EVAL_REPORTING_PROJECT=$_EVAL_REPORTING_PROJECT
 
-
         # Set environment variables for extension
         export CLOUD_SQL_POSTGRES_PROJECT=$PROJECT_ID
         export CLOUD_SQL_POSTGRES_INSTANCE=$_CLOUD_SQL_INSTANCE
@@ -84,18 +83,28 @@ steps:
         # Maps the decrypted DB_PASSWORD to the exact variable expected by gemini_cli and extension skills
         export CLOUD_SQL_POSTGRES_PASSWORD=$$DB_PASSWORD
 
-        # Combine CI metadata with run config
-        cat /workspace/evals/ci_metadata.yaml >> /workspace/evals/run_config.yaml
+        # Combine CI metadata with all available run configs
+        for config in /workspace/evals/*run_config.yaml; do
+          if [ -f "$config" ]; then
+            echo "Appending CI metadata to $config"
+            cat /workspace/evals/ci_metadata.yaml >> "$config"
+          fi
+        done
 
-        # Substitute environment variables in model_config.yaml
+        # Substitute environment variables in all configs
         python3 /workspace/evals/substitute_env.py
 
         cd /evalbench
         export PYTHONPATH=./evalbench:./evalbench/evalproto
         export PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python
 
-        echo "Launching Standalone Evaluation..."
-        python3 evalbench/evalbench.py --experiment_config=/workspace/evals/run_config.yaml
+        # Run evaluations for all available run configs
+        for config in /workspace/evals/*run_config.yaml; do
+          if [ -f "$config" ]; then
+            echo "Launching Evaluation for config: $config"
+            python3 evalbench/evalbench.py --experiment_config="$config"
+          fi
+        done
 
 
 availableSecrets:

evals/claude_code_model.yaml

@@ -0,0 +1,18 @@
+claude_code_version: "@anthropic-ai/claude-code@2.1.85"
+generator: claude_code
+model: "claude-opus-4-6" # Use "claude-opus-4-20250514" for direct API
+
+use_vertex: true
+vertex_project_id: "${GOOGLE_CLOUD_PROJECT}"
+vertex_region: "us-central1"
+
+env:
+  GOOGLE_CLOUD_PROJECT: "${GOOGLE_CLOUD_PROJECT}"
+
+setup:
+  mcp_servers:
+    "cloud-sql":
+      httpUrl: "https://sqladmin.googleapis.com/mcp"
+      authProviderType: google_credentials
+      headers:
+        X-Goog-User-Project: "${GOOGLE_CLOUD_PROJECT}"

evals/claude_run_config.yaml

@@ -0,0 +1,41 @@
+# Copyright 2025 Google LLC
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+dataset_config: /workspace/evals/dataset.json
+dataset_format: gemini-cli-format
+
+orchestrator: agent
+model_config: /workspace/evals/claude_code_model.yaml
+simulated_user_model_config: /workspace/evals/gemini_2.5_pro_model.yaml
+
+scorers:
+  # Qualitative (Judge-based)
+  goal_completion:
+    model_config: /workspace/evals/gemini_2.5_pro_model.yaml
+  behavioral_metrics:
+    model_config: /workspace/evals/gemini_2.5_pro_model.yaml
+  skills_best_practices:
+    model_config: /workspace/evals/gemini_2.5_pro_model.yaml
+    skills_dir: /workspace/cloud-sql-postgresql/skills
+
+  # Performance
+  turn_count: {}
+  end_to_end_latency: {}
+  tool_call_latency: {}
+  token_consumption: {}
+  skills_trajectory: {}
+
+reporting:
+  bigquery:
+    gcp_project_id: "${EVAL_REPORTING_PROJECT}"

evals/gemini_cli_model.yaml

@@ -0,0 +1,33 @@
+# Copyright 2025 Google LLC
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+gemini_cli_version: "@google/gemini-cli@latest"
+generator: gemini_cli
+env:
+  GOOGLE_CLOUD_PROJECT: "${GOOGLE_CLOUD_PROJECT}"
+  GOOGLE_CLOUD_LOCATION: "global"
+  GOOGLE_GENAI_USE_VERTEXAI: "true"
+  GEMINI_CLI_TRUST_WORKSPACE: "true"
+setup:
+  extensions:
+    # Points to the symlink created in cloudbuild.yaml to match the extension ID
+    "/workspace/cloud-sql-postgresql":
+      settings:
+        CLOUD_SQL_POSTGRES_PROJECT: "${CLOUD_SQL_POSTGRES_PROJECT}"
+        CLOUD_SQL_POSTGRES_INSTANCE: "${CLOUD_SQL_POSTGRES_INSTANCE}"
+        CLOUD_SQL_POSTGRES_REGION: "${CLOUD_SQL_POSTGRES_REGION}"
+        CLOUD_SQL_POSTGRES_DATABASE: "${CLOUD_SQL_POSTGRES_DATABASE}"
+        CLOUD_SQL_POSTGRES_USER: "${CLOUD_SQL_POSTGRES_USER}"
+        CLOUD_SQL_POSTGRES_PASSWORD: '${CLOUD_SQL_POSTGRES_PASSWORD}'
+        CLOUD_SQL_POSTGRES_IP_TYPE: "${CLOUD_SQL_POSTGRES_IP_TYPE}"

evals/run_config.yaml evals/gemini_run_config.yamlevals/run_config.yaml renamed to evals/gemini_run_config.yaml

@@ -16,7 +16,7 @@ dataset_config: /workspace/evals/dataset.json
 dataset_format: gemini-cli-format
 
 orchestrator: geminicli
-model_config: /workspace/evals/model_config.yaml
+model_config: /workspace/evals/gemini_cli_model.yaml
 simulated_user_model_config: /workspace/evals/gemini_2.5_pro_model.yaml
 
 scorers:
@@ -39,4 +39,3 @@ scorers:
 reporting:
   bigquery:
     gcp_project_id: "${EVAL_REPORTING_PROJECT}"
-

evals/substitute_env.py

@@ -1,18 +1,23 @@
 import os
 import re
+import glob
 
 def main():
-    yaml_paths = ['/workspace/evals/model_config.yaml', '/workspace/evals/run_config.yaml', '/workspace/evals/dataset.json']
-    for yaml_path in yaml_paths:
-        if os.path.exists(yaml_path):
-            with open(yaml_path, 'r') as f:
-                content = f.read()
-            content = re.sub(r'\${(\w+)}', lambda m: os.environ.get(m.group(1), m.group(0)), content)
-            with open(yaml_path, 'w') as f:
-                f.write(content)
-            print(f"Successfully substituted environment variables in {yaml_path}")
-        else:
-            print(f"File not found: {yaml_path}")
+    # Find all .yaml and .json files in /workspace/evals
+    paths = glob.glob('/workspace/evals/**/*.yaml', recursive=True) + glob.glob('/workspace/evals/**/*.json', recursive=True)
+    
+    for path in paths:
+        if os.path.isfile(path):
+            try:
+                with open(path, 'r') as f:
+                    content = f.read()
+                # Substitute ${VAR} with environment variables
+                content = re.sub(r'\${(\w+)}', lambda m: os.environ.get(m.group(1), m.group(0)), content)
+                with open(path, 'w') as f:
+                    f.write(content)
+                print(f"Successfully substituted environment variables in {path}")
+            except Exception as e:
+                print(f"Error processing {path}: {e}")
 
 if __name__ == '__main__':
     main()