hte-cli 0.1.28__py3-none-any.whl → 0.2.8__py3-none-any.whl

hte_cli/api_client.py

@@ -201,6 +201,31 @@ class APIClient:
             },
         )
 
+    # =========================================================================
+    # Session-based API (new flow: hte-cli session join <session_id>)
+    # =========================================================================
+
+    def join_session(self, session_id: str) -> dict:
+        """Join an existing session created by web UI.
+
+        Returns session info including task data, benchmark, mode, etc.
+        Sets cli_connected_at on the server.
+        """
+        return self.post(f"/sessions/{session_id}/join")
+
+    def get_session_files(self, session_id: str) -> bytes:
+        """Download task files for a session as zip."""
+        return self.get_raw(f"/sessions/{session_id}/files")
+
+    def get_session_compose(self, session_id: str) -> str:
+        """Get compose.yaml content for a session."""
+        content = self.get_raw(f"/sessions/{session_id}/compose")
+        return content.decode("utf-8")
+
+    # =========================================================================
+    # Result Upload
+    # =========================================================================
+
     def upload_result(
         self,
         session_id: str,

hte_cli/cli.py

@@ -147,13 +147,319 @@ def auth_status(ctx):
 
 
 # =============================================================================
-# Tasks Commands
+# Session Commands (New flow: session join <session_id>)
+# =============================================================================
+
+
+@cli.group()
+def session():
+    """Session management commands."""
+    pass
+
+
+@session.command("join")
+@click.argument("session_id")
+@click.option("--force-setup", is_flag=True, help="Re-run setup even if reconnecting")
+@click.pass_context
+def session_join(ctx, session_id: str, force_setup: bool):
+    """Join an existing session by ID.
+
+    This is the primary way to start working on a task:
+    1. Start the task from the web UI (creates session)
+    2. Run this command with the session ID shown in the web UI
+    3. The environment will be set up and the timer will start
+    """
+    config: Config = ctx.obj["config"]
+
+    if not config.is_authenticated():
+        console.print("[red]Not logged in. Run: hte-cli auth login[/red]")
+        sys.exit(1)
+
+    api = APIClient(config)
+
+    # Step 1: Join session
+    console.print()
+    with Progress(
+        SpinnerColumn(),
+        TextColumn("[progress.description]{task.description}"),
+        console=console,
+    ) as progress:
+        progress.add_task("Joining session...", total=None)
+
+        try:
+            session_info = api.join_session(session_id)
+        except APIError as e:
+            if "Invalid session ID format" in str(e):
+                console.print(f"[red]{e}[/red]")
+            elif e.status_code == 404:
+                console.print("[red]Session not found. Check the session ID and try again.[/red]")
+            elif e.status_code == 400 and "paused" in str(e).lower():
+                console.print("[yellow]Session is paused. Please resume from the web UI first.[/yellow]")
+            else:
+                console.print(f"[red]Error: {e}[/red]")
+            sys.exit(1)
+
+    # Check if reconnecting (session already in_progress)
+    is_reconnect = session_info.get("status") == "in_progress"
+
+    if is_reconnect and not force_setup:
+        console.print("[yellow]Reconnecting to existing session...[/yellow]")
+        console.print()
+
+    console.print(
+        Panel(
+            f"[bold]Task:[/bold] {session_info['task_id']}\n"
+            f"[bold]Benchmark:[/bold] {session_info['benchmark']}\n"
+            f"[bold]Mode:[/bold] {session_info['mode']}",
+            title="Session Joined",
+        )
+    )
+    console.print()
+
+    # Import runner components
+    import time
+    from hte_cli.events import EventStreamer
+    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)
+
+    # Step 2: Download task files and compose (skip if reconnecting without force)
+    files_zip = None
+    compose_yaml = None
+
+    if not is_reconnect or force_setup:
+        console.print("[bold]Step 1:[/bold] Downloading task resources...")
+
+        with console.status("[dim]Fetching task files...[/dim]"):
+            try:
+                files_zip = api.get_session_files(session_id)
+                console.print("  [green]✓[/green] Task files downloaded")
+            except APIError as e:
+                console.print(f"  [dim]○ No task files (optional)[/dim]")
+                files_zip = None
+
+        with console.status("[dim]Fetching compose configuration...[/dim]"):
+            try:
+                compose_yaml = api.get_session_compose(session_id)
+                console.print("  [green]✓[/green] Docker compose downloaded")
+            except APIError as e:
+                console.print(f"  [dim]○ No compose file (optional)[/dim]")
+                compose_yaml = None
+
+        console.print()
+
+    # Validate compose for sandbox-required benchmarks
+    benchmark = session_info.get("benchmark", "").lower()
+    SANDBOX_REQUIRED_BENCHMARKS = {"cybench", "intercode-ctf", "cvebench", "cybergym", "nyuctf"}
+    if benchmark in SANDBOX_REQUIRED_BENCHMARKS and not compose_yaml and not is_reconnect:
+        console.print(
+            f"[red]Error: {benchmark} requires a Docker sandbox but no compose file was found.[/red]"
+        )
+        console.print()
+        console.print(
+            f"Please contact support: {SUPPORT_EMAIL}"
+        )
+        sys.exit(1)
+
+    # 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"),
+        "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()
+    images = []
+    pulled_images = []
+    cached_images = []
+    failed_images = []
+
+    if not is_reconnect or force_setup:
+        # Extract images from compose
+        if compose_yaml:
+            images = extract_images_from_compose(compose_yaml)
+
+        # Send setup_started event
+        events.setup_started(images=images)
+
+        # Pull images if we have any
+        if images:
+            from hte_cli.image_utils import check_image_exists_locally
+
+            console.print(f"[bold]Step 2:[/bold] Pulling {len(images)} Docker image(s)...")
+            pull_start = time.monotonic()
+
+            for img in images:
+                short_name = img.split("/")[-1][:40]
+
+                # Check if already cached
+                if check_image_exists_locally(img):
+                    console.print(f"  [green]✓[/green] {short_name} [dim](cached)[/dim]")
+                    cached_images.append(img)
+                    continue
+
+                # Need to pull - show progress
+                last_status = ["connecting..."]  # Use list for closure mutability
+                with console.status(f"[yellow]↓[/yellow] {short_name} [dim]connecting...[/dim]") as status:
+                    def show_progress(image: str, line: str):
+                        # Parse docker pull output for layer progress
+                        # Lines look like: "abc123: Downloading [====>   ] 10MB/50MB"
+                        # Or: "abc123: Extracting [====>   ] 10MB/50MB"
+                        # Or: "abc123: Pull complete", "Digest: sha256:...", "Status: ..."
+                        display_text = None
+                        if ": " in line:
+                            parts = line.split(": ", 1)
+                            if len(parts) == 2:
+                                layer_id = parts[0][-8:]  # Last 8 chars of layer ID
+                                layer_status = parts[1]  # Don't truncate - keep full progress
+                                # Show progress bars for Downloading/Extracting with MB info
+                                if "Downloading" in layer_status or "Extracting" in layer_status:
+                                    # Keep progress: "[====>   ] 10.5MB/50MB"
+                                    display_text = f"{layer_id}: {layer_status[:50]}"
+                                elif "Pull complete" in layer_status:
+                                    display_text = f"{layer_id}: done"
+                                elif "Download complete" in layer_status:
+                                    display_text = f"{layer_id}: download done"
+                                elif "Already exists" in layer_status:
+                                    display_text = f"{layer_id}: cached"
+                                elif "Waiting" in layer_status:
+                                    display_text = f"{layer_id}: waiting"
+                                elif "Verifying" in layer_status:
+                                    display_text = f"{layer_id}: verifying"
+                                else:
+                                    display_text = line[:55]
+                        elif line.strip():
+                            display_text = line[:55]
+
+                        if display_text and display_text != last_status[0]:
+                            last_status[0] = display_text
+                            status.update(f"[yellow]↓[/yellow] {short_name} [dim]{display_text}[/dim]")
+
+                    success = pull_image_with_progress(img, on_progress=show_progress)
+
+                if success:
+                    console.print(f"  [green]✓[/green] {short_name} [dim](downloaded)[/dim]")
+                    pulled_images.append(img)
+                else:
+                    console.print(f"  [red]✗[/red] {short_name} [dim](failed)[/dim]")
+                    failed_images.append(img)
+
+            pull_duration = time.monotonic() - pull_start
+            events.image_pull_completed(
+                duration_seconds=pull_duration,
+                pulled=pulled_images,
+                cached=cached_images,
+                failed=failed_images,
+            )
+            console.print()
+
+        # Send setup_completed - THIS STARTS THE TIMER ON SERVER
+        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:
+        # Reconnecting - compose should already be running
+        console.print("[dim]Skipping setup (use --force-setup to re-run)[/dim]")
+        console.print()
+
+    # Step 4: Show instructions
+    if session_info.get("instructions"):
+        console.print(Panel(session_info["instructions"], title="Task Instructions"))
+        console.print()
+
+    # Step 3: Run the task using TaskRunner
+    step_num = "3" if (not is_reconnect or force_setup) and images else "2" if (not is_reconnect or force_setup) else "1"
+    console.print(f"[bold]Step {step_num}:[/bold] Starting task environment...")
+
+    events.docker_started()
+
+    runner = TaskRunner()
+    eval_log_bytes = None
+    try:
+        with console.status("[dim]Launching Docker containers (this may take a minute)...[/dim]"):
+            result = runner.run_from_assignment(
+                assignment=assignment,
+                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:
+        events.session_completed(
+            elapsed_seconds=result.time_seconds,
+            answer=result.answer,
+        )
+
+        console.print()
+        console.print("[green]Task completed![/green]")
+        console.print(f"Answer: {result.answer}")
+        console.print(f"Time: {result.time_seconds:.1f}s")
+
+        # Upload to server
+        with Progress(
+            SpinnerColumn(),
+            TextColumn("[progress.description]{task.description}"),
+            console=console,
+        ) as progress:
+            progress.add_task("Uploading result...", total=None)
+            try:
+                upload_result = api.upload_result(
+                    session_id=session_id,
+                    answer=result.answer or "",
+                    client_active_seconds=result.time_seconds,
+                    eval_log_bytes=eval_log_bytes,
+                    score=result.score,
+                    score_binarized=result.score_binarized,
+                    agent_id=result.agent_id,
+                )
+            except APIError as e:
+                console.print(f"[red]Failed to upload result: {e}[/red]")
+                sys.exit(1)
+
+        if upload_result.get("score") is not None:
+            console.print(f"Score: {upload_result['score']}")
+
+        console.print()
+        console.print("[green]Done! Return to the web UI to see your results.[/green]")
+
+
+# =============================================================================
+# Tasks Commands (DEPRECATED - use 'session join' instead)
 # =============================================================================
 
 
 @cli.group()
 def tasks():
-    """Task commands."""
+    """Task commands (deprecated - use 'session join' instead)."""
     pass
 
 
@@ -223,7 +529,23 @@ def tasks_list(ctx):
 @click.argument("task_id", required=False)
 @click.pass_context
 def tasks_run(ctx, task_id: str | None):
-    """Run a task (default: highest priority pending task)."""
+    """[DEPRECATED] Run a task - use 'session join' instead."""
+    console.print()
+    console.print("[red]This command is deprecated.[/red]")
+    console.print()
+    console.print("The new workflow is:")
+    console.print("  1. Start the task from the web UI: https://cyber-task-horizons.com")
+    console.print("  2. Run the command shown: [bold]hte-cli session join <session_id>[/bold]")
+    console.print()
+    console.print("This ensures accurate timing by starting the timer only when")
+    console.print("the environment is ready, not including Docker setup time.")
+    console.print()
+    sys.exit(1)
+
+
+# Keep the old implementation as _tasks_run_legacy for testing if needed
+def _tasks_run_legacy(ctx, task_id: str | None):
+    """Legacy implementation of tasks run (for testing only)."""
     config: Config = ctx.obj["config"]
 
     if not config.is_authenticated():

{hte_cli-0.1.28.dist-info → hte_cli-0.2.8.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hte-cli
-Version: 0.1.28
+Version: 0.2.8
 Summary: Human Time-to-Completion Evaluation CLI
 Project-URL: Homepage, https://github.com/sean-peters-au/lyptus-mono
 Author: Lyptus Research
@@ -23,6 +23,10 @@ Requires-Dist: platformdirs>=4.0
 Requires-Dist: pydantic>=2.0
 Requires-Dist: pyyaml>=6.0
 Requires-Dist: rich>=13.0
+Provides-Extra: dev
+Requires-Dist: pexpect>=4.8; extra == 'dev'
+Requires-Dist: pytest>=7.0; extra == 'dev'
+Requires-Dist: requests>=2.28; extra == 'dev'
 Description-Content-Type: text/markdown
 
 # hte-cli

{hte_cli-0.1.28.dist-info → hte_cli-0.2.8.dist-info}/RECORD

@@ -1,7 +1,7 @@
 hte_cli/__init__.py,sha256=fDGXp-r8bIoLtlQnn5xJ_CpwMhonvk9bGjZQsjA2mDI,914
 hte_cli/__main__.py,sha256=63n0gNGfskidWDU0aAIF2N8lylVCLYKVIkrN9QiORoo,107
-hte_cli/api_client.py,sha256=mO4buDND5cIWESg4gSKb8WkdA1iPwkmTa0L3xL6lvNQ,8153
-hte_cli/cli.py,sha256=yxqh-NacsrILjidE7CD2IHQ7fzZyaDCsPgEFl4lCG_w,29328
+hte_cli/api_client.py,sha256=m42kfFZS72Nu_VuDwxRsLNy4ziCcvgk7KNWBh9gwqy0,9257
+hte_cli/cli.py,sha256=W9R_jHBLhLho2GyroKzCCg6EhBluCrFJdZ9zCaKFGuo,42745
 hte_cli/config.py,sha256=42Xv__YMSeRLs2zhGukJkIXFKtnBtYCHnONfViGyt2g,3387
 hte_cli/errors.py,sha256=1J5PpxcUKBu6XjigMMCPOq4Zc12tnv8LhAsiaVFWLQM,2762
 hte_cli/events.py,sha256=Zn-mroqaLHNzdT4DFf8st1Qclglshihdc09dBfCN070,5522
@@ -9,7 +9,7 @@ hte_cli/image_utils.py,sha256=454yoZEI1duNYrZC8UjhfZzDRP4Nxdrf2TvnZ_54G1k,4439
 hte_cli/runner.py,sha256=DhC8FMjHwfLR193iP4thLDRZrNssYA9KH1WYKU2JKeg,13535
 hte_cli/scorers.py,sha256=sFoPJePRt-K191-Ga4cVmrldruJclYXTOLkU_C9nCDI,6025
 hte_cli/version_check.py,sha256=WVZyGy2XfAghQYdd2N9-0Qfg-7pgp9gt4761-PnmacI,1708
-hte_cli-0.1.28.dist-info/METADATA,sha256=JRF7leLDC5EpYXnq6yBVb9YNMpvI8Ijgodnvh5gvMqs,3615
-hte_cli-0.1.28.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
-hte_cli-0.1.28.dist-info/entry_points.txt,sha256=XbyEEi1H14DFAt0Kdl22e_IRVEGzimSzYSh5HlhKlFA,41
-hte_cli-0.1.28.dist-info/RECORD,,
+hte_cli-0.2.8.dist-info/METADATA,sha256=XT1WLIfvC2PRQtcFB19qTpBdMApP_UKabzal-Fcn8Cw,3767
+hte_cli-0.2.8.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
+hte_cli-0.2.8.dist-info/entry_points.txt,sha256=XbyEEi1H14DFAt0Kdl22e_IRVEGzimSzYSh5HlhKlFA,41
+hte_cli-0.2.8.dist-info/RECORD,,