Add multi-window baseline vs tuned compare flow

src/aituner/cli.py

@@ -5,6 +5,7 @@ import json
 import sys
 from pathlib import Path
 
+from .compare import run_compare
 from .job import append_job, build_trial_job
 from .llm import build_prompt, call_llm_for_proposal, load_capability_profile, parse_proposal_text
 from .spec import Proposal, SpecError, load_study_spec
@@ -183,6 +184,25 @@ def cmd_worker_run_trial(args: argparse.Namespace) -> int:
     return 0
 
 
+def cmd_compare_run(args: argparse.Namespace) -> int:
+    summary = run_compare(
+        Path(args.spec).resolve(),
+        output_root=Path(args.output_root).resolve() if args.output_root else None,
+    )
+    print(
+        json.dumps(
+            {
+                "compare_id": summary["compare_id"],
+                "compare_root": summary["compare_root"],
+                "window_count": summary["aggregate"]["window_count"],
+                "wins": summary["aggregate"]["wins"],
+            },
+            ensure_ascii=False,
+        )
+    )
+    return 0
+
+
 def build_parser() -> argparse.ArgumentParser:
     parser = argparse.ArgumentParser(description="AITuner CLI")
     subparsers = parser.add_subparsers(dest="command", required=True)
@@ -239,6 +259,13 @@ def build_parser() -> argparse.ArgumentParser:
     run.add_argument("--trial-spec", required=True)
     run.set_defaults(func=cmd_worker_run_trial)
 
+    compare = subparsers.add_parser("compare")
+    compare_sub = compare.add_subparsers(dest="compare_command", required=True)
+    compare_run = compare_sub.add_parser("run")
+    compare_run.add_argument("--spec", required=True)
+    compare_run.add_argument("--output-root")
+    compare_run.set_defaults(func=cmd_compare_run)
+
     return parser