hte-cli 0.2.0__tar.gz → 0.2.1__tar.gz

{hte_cli-0.2.0 → hte_cli-0.2.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hte-cli
-Version: 0.2.0
+Version: 0.2.1
 Summary: Human Time-to-Completion Evaluation CLI
 Project-URL: Homepage, https://github.com/sean-peters-au/lyptus-mono
 Author: Lyptus Research

{hte_cli-0.2.0 → hte_cli-0.2.1}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "hte-cli"
-version = "0.2.0"
+version = "0.2.1"
 description = "Human Time-to-Completion Evaluation CLI"
 readme = "README.md"
 requires-python = ">=3.11"

{hte_cli-0.2.0 → hte_cli-0.2.1}/src/hte_cli/cli.py

@@ -217,8 +217,10 @@ def session_join(ctx, session_id: str, force_setup: bool):
     console.print()
 
     # Import runner components
+    import time
     from hte_cli.events import EventStreamer
-    from hte_cli.runner import TaskRunner, DockerComposeManager
+    from hte_cli.runner import TaskRunner
+    from hte_cli.image_utils import extract_images_from_compose, pull_image_with_progress
 
     # Create event streamer
     events = EventStreamer(api, session_id)
@@ -264,48 +266,45 @@ def session_join(ctx, session_id: str, force_setup: bool):
     # Build assignment dict for runner compatibility
     assignment = {
         "assignment_id": session_info.get("assignment_id"),
+        "session_id": session_id,
         "task_id": session_info["task_id"],
         "benchmark": session_info["benchmark"],
         "mode": session_info["mode"],
         "time_cap_seconds": session_info.get("time_cap_seconds"),
-    }
-
-    # Build task dict for runner
-    task_data = {
-        "instructions": session_info.get("instructions", ""),
-        "metadata": session_info.get("metadata", {}),
-        "scorer_type": session_info.get("scorer_type"),
-        "intermediate_scoring": session_info.get("intermediate_scoring", False),
+        "task": {
+            "instructions": session_info.get("instructions", ""),
+            "metadata": session_info.get("metadata", {}),
+            "scorer_type": session_info.get("scorer_type"),
+            "intermediate_scoring": session_info.get("intermediate_scoring", False),
+        },
     }
 
     # Step 3: Run setup (skip if reconnecting without force)
+    setup_start_time = time.monotonic()
     if not is_reconnect or force_setup:
         # Send setup_started event
         events.setup_started({"cli_version": __version__, "task_id": assignment["task_id"]})
 
-        # Extract files and run compose
-        runner = TaskRunner(session_id, api, events, console)
-        compose_manager = None
-
+        # Pull images if we have compose
         if compose_yaml:
-            compose_manager = DockerComposeManager(
-                compose_yaml=compose_yaml,
-                files_zip=files_zip,
-                session_id=session_id,
-                task_id=assignment["task_id"],
-                console=console,
-            )
-
-            # Pull images
-            events.image_pull_started({})
-            compose_manager.pull_images()
-            events.image_pull_completed({})
-
-            # Start containers
-            compose_manager.up()
+            images = extract_images_from_compose(compose_yaml)
+            if images:
+                console.print(f"[bold]Pulling {len(images)} Docker image(s)...[/bold]")
+                events.image_pull_started({})
+                for img in images:
+                    short_name = img.split("/")[-1][:40]
+                    with console.status(f"[yellow]↓[/yellow] {short_name}") as status:
+                        success = pull_image_with_progress(img)
+                    if success:
+                        console.print(f"  [green]✓[/green] {short_name}")
+                    else:
+                        console.print(f"  [red]✗[/red] {short_name} (failed)")
+                events.image_pull_completed({})
+                console.print()
 
         # Send setup_completed - THIS STARTS THE TIMER ON SERVER
-        events.setup_completed({"cli_version": __version__})
+        total_setup = time.monotonic() - setup_start_time
+        events.setup_completed(total_seconds=total_setup)
         console.print("[green]Environment ready! Timer started.[/green]")
         console.print()
     else:
@@ -318,19 +317,37 @@ def session_join(ctx, session_id: str, force_setup: bool):
         console.print(Panel(session_info["instructions"], title="Task Instructions"))
         console.print()
 
-    # Step 5: Run the task interaction loop
-    runner = TaskRunner(session_id, api, events, console)
+    # Step 5: Run the task using TaskRunner
+    console.print("[bold]Starting task environment...[/bold]")
+    console.print("[dim]Launching Docker containers...[/dim]")
+    console.print()
+
+    events.docker_started()
+
+    runner = TaskRunner()
+    eval_log_bytes = None
     try:
-        result = runner.run(
+        result = runner.run_from_assignment(
             assignment=assignment,
-            task=task_data,
             compose_yaml=compose_yaml,
             files_zip=files_zip,
         )
+        # Read eval log before cleanup
+        if result.eval_log_path and result.eval_log_path.exists():
+            eval_log_bytes = result.eval_log_path.read_bytes()
     except KeyboardInterrupt:
+        events.docker_stopped(exit_code=130)
         console.print()
         console.print("[yellow]Interrupted. Session remains active - you can reconnect later.[/yellow]")
         sys.exit(0)
+    except Exception as e:
+        events.docker_stopped(exit_code=1)
+        console.print(f"[red]Task execution failed: {e}[/red]")
+        sys.exit(1)
+    finally:
+        runner.cleanup()
+
+    events.docker_stopped(exit_code=0)
 
     # Step 6: Upload result
     if result and result.answer:
@@ -356,7 +373,7 @@ def session_join(ctx, session_id: str, force_setup: bool):
                     session_id=session_id,
                     answer=result.answer or "",
                     client_active_seconds=result.time_seconds,
-                    eval_log_bytes=result.eval_log_bytes,
+                    eval_log_bytes=eval_log_bytes,
                     score=result.score,
                     score_binarized=result.score_binarized,
                     agent_id=result.agent_id,