diff --git a/wolf/cli.py b/wolf/cli.py
new file mode 100644
index 0000000..0b286c4
--- /dev/null
+++ b/wolf/cli.py
@@ -0,0 +1,94 @@
+import argparse
+import logging
+import sys
+from pathlib import Path
+from .config import Config, setup_logging
+from .gitea import GiteaClient
+from .task import TaskGenerator
+from .runner import AgentRunner
+from .evaluator import Evaluator
+from .leaderboard import Leaderboard
+
+def main():
+    parser = argparse.ArgumentParser(description="Wolf: Model Evaluation Backbone for Sovereign AI Fleet.")
+    parser.add_argument("--config", help="Path to wolf-config.yaml")
+    parser.add_argument("--task-spec", help="Path to task specification JSON")
+    parser.add_argument("--run", action="store_true", help="Run pending tasks")
+    parser.add_argument("--evaluate", action="store_true", help="Evaluate open PRs")
+    parser.add_argument("--leaderboard", action="store_true", help="Show model rankings")
+    
+    args = parser.parse_args()
+    
+    # 1. Setup config and logging
+    config = Config(args.config)
+    setup_logging(config.get('log_dir'))
+    
+    logging.info("Wolf starting...")
+    
+    # 2. Initialize clients
+    gitea_config = config.get('gitea', {})
+    gitea = GiteaClient(gitea_config.get('base_url'), gitea_config.get('token'))
+    
+    # 3. Run tasks
+    if args.run:
+        task_gen = TaskGenerator(gitea)
+        if args.task_spec:
+            tasks = task_gen.from_spec(args.task_spec)
+        else:
+            # Default to fetching issues from a specific repo
+            tasks = task_gen.from_gitea_issues(gitea_config.get('owner'), gitea_config.get('repo'))
+        
+        # Assign tasks to models
+        models = config.get('models', [])
+        if not models:
+            logging.error("No models configured for assignment.")
+            return
+        
+        assigned_tasks = task_gen.assign_tasks(tasks, models)
+        
+        runner = AgentRunner(gitea, config.data)
+        for task in assigned_tasks:
+            runner.execute_task(task)
+            
+    # 4. Evaluate PRs
+    if args.evaluate:
+        evaluator = Evaluator(gitea)
+        leaderboard = Leaderboard(config.get('leaderboard_path'))
+        
+        # Fetch open PRs from the configured repo
+        # This is a simplified version; in reality, we'd track which PRs were created by Wolf
+        prs = gitea.get_issues(gitea_config.get('owner'), gitea_config.get('repo'), state="open")
+        for pr in prs:
+            if "Wolf Task:" in pr['title']:
+                score_data = evaluator.score_pr(gitea_config.get('owner'), gitea_config.get('repo'), pr['number'])
+                
+                # Extract model info from PR body or title
+                # For now, we'll assume the model name is in the PR body
+                model_name = "unknown"
+                provider = "unknown"
+                if "model" in pr['body']:
+                    # Simple regex to find model and provider
+                    import re
+                    match = re.search(r"model (.*) via (.*)\.", pr['body'])
+                    if match:
+                        model_name = match.group(1)
+                        provider = match.group(2)
+                
+                leaderboard.record_score(model_name, provider, pr['number'], score_data)
+                
+    # 5. Show leaderboard
+    if args.leaderboard:
+        leaderboard = Leaderboard(config.get('leaderboard_path'))
+        rankings = leaderboard.get_rankings()
+        
+        print("\n🏆 WOLF LEADERBOARD")
+        print("-" * 60)
+        print(f"{'Model':<25} | {'Provider':<15} | {'Score':<8} | {'Tasks':<6} | {'Ready'}")
+        print("-" * 60)
+        for r in rankings:
+            ready = "✅" if r['serverless_ready'] else "❌"
+            print(f"{r['model_name']:<25} | {r['provider']:<15} | {r['average_score']:<8.2f} | {r['total_tasks']:<6} | {ready}")
+        print("-" * 60)
+
+if __name__ == "__main__":
+    main()