hte-cli 0.2.29__tar.gz → 0.2.31__tar.gz

{hte_cli-0.2.29 → hte_cli-0.2.31}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hte-cli
-Version: 0.2.29
+Version: 0.2.31
 Summary: Human Time-to-Completion Evaluation CLI
 Project-URL: Homepage, https://github.com/sean-peters-au/lyptus-mono
 Author: Lyptus Research

{hte_cli-0.2.29 → hte_cli-0.2.31}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "hte-cli"
-version = "0.2.29"
+version = "0.2.31"
 description = "Human Time-to-Completion Evaluation CLI"
 readme = "README.md"
 requires-python = ">=3.11"

{hte_cli-0.2.29 → hte_cli-0.2.31}/src/hte_cli/api_client.py

@@ -265,3 +265,27 @@ class APIClient:
             json=payload,
             timeout=UPLOAD_TIMEOUT,
         )
+
+    def upload_partial_log(
+        self,
+        session_id: str,
+        eval_log_bytes: bytes,
+    ) -> dict:
+        """Upload partial eval log for interrupted sessions.
+
+        Args:
+            session_id: The session ID
+            eval_log_bytes: Partial eval log content
+
+        Returns:
+            Response dict with status and log_path
+        """
+        payload = {
+            "eval_log_base64": base64.b64encode(eval_log_bytes).decode("ascii"),
+        }
+
+        return self.post(
+            f"/sessions/{session_id}/partial-log",
+            json=payload,
+            timeout=UPLOAD_TIMEOUT,
+        )

{hte_cli-0.2.29 → hte_cli-0.2.31}/src/hte_cli/cli.py

@@ -23,6 +23,42 @@ console = Console()
 SUPPORT_EMAIL = "jacktpayne51@gmail.com"
 
 
+def _find_eval_log_bytes(runner) -> bytes | None:
+    """Find and read eval log bytes from runner's work directory.
+
+    Used for interrupted sessions to upload partial logs.
+    """
+    try:
+        # Look for eval logs in the work directory
+        if not runner.work_dir.exists():
+            return None
+
+        # Find any .eval files in the work directory tree
+        eval_files = list(runner.work_dir.rglob("*.eval"))
+        if not eval_files:
+            return None
+
+        # Get the most recent one
+        eval_files.sort(key=lambda p: p.stat().st_mtime, reverse=True)
+        return eval_files[0].read_bytes()
+    except Exception:
+        return None
+
+
+def _upload_partial_log(api: APIClient, session_id: str, eval_log_bytes: bytes, console: Console) -> None:
+    """Upload partial eval log for interrupted session.
+
+    Best-effort: silently handles failures to not block exit.
+    """
+    try:
+        size_kb = len(eval_log_bytes) / 1024
+        console.print(f"[dim]Uploading partial eval log ({size_kb:.0f} KB)...[/dim]")
+        api.upload_partial_log(session_id, eval_log_bytes)
+        console.print("[dim]Partial eval log uploaded.[/dim]")
+    except Exception as e:
+        console.print(f"[dim]Could not upload partial log: {e}[/dim]")
+
+
 @click.group()
 @click.version_option(__version__, prog_name="hte-cli")
 @click.pass_context
@@ -282,6 +318,11 @@ def session_join(ctx, session_id: str, force_setup: bool):
     # Validate compose for sandbox-required benchmarks
     benchmark = session_info.get("benchmark", "").lower()
     SANDBOX_REQUIRED_BENCHMARKS = {"cybench", "intercode-ctf", "cvebench", "cybergym", "nyuctf"}
+
+    # NYU CTF requires a shared Docker network for agent/game-server communication
+    if benchmark == "nyuctf":
+        _ensure_nyuctf_network()
+
     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]"
@@ -333,15 +374,11 @@ def session_join(ctx, session_id: str, force_setup: bool):
             host_platform = get_host_docker_platform()
 
             if is_linux_arm:
-                console.print(
-                    f"[yellow]![/yellow] Detected [bold]Linux ARM64[/bold] environment"
-                )
+                console.print("[yellow]![/yellow] Detected [bold]Linux ARM64[/bold] environment")
                 console.print(
                     f"  [dim]Will verify cached images match host architecture ({host_platform})[/dim]"
                 )
-                console.print(
-                    f"  [dim]Mismatched images will be automatically re-pulled[/dim]"
-                )
+                console.print("  [dim]Mismatched images will be automatically re-pulled[/dim]")
                 console.print()
 
             console.print(f"[bold]Step 2:[/bold] Pulling {len(images)} Docker image(s)...")
@@ -377,7 +414,7 @@ def session_join(ctx, session_id: str, force_setup: bool):
                             f"      [dim]Cached image: {image_arch} | Host: {host_arch}[/dim]"
                         )
                         console.print(
-                            f"      [dim]Removing cached image and re-pulling correct architecture...[/dim]"
+                            "      [dim]Removing cached image and re-pulling correct architecture...[/dim]"
                         )
 
                         needed_fix, fix_msg = fix_image_architecture(img)
@@ -391,7 +428,7 @@ def session_join(ctx, session_id: str, force_setup: bool):
                             # No ARM variant available - this is an x86-only image
                             # Re-pull the amd64 version and warn about QEMU
                             console.print(
-                                f"      [dim]No ARM variant available - re-pulling x86 version...[/dim]"
+                                "      [dim]No ARM variant available - re-pulling x86 version...[/dim]"
                             )
                             success = pull_image_with_progress(img)
                             if success:
@@ -402,7 +439,9 @@ def session_join(ctx, session_id: str, force_setup: bool):
                                 pulled_images.append(img)
                                 continue
                             else:
-                                console.print(f"  [red]✗[/red] {short_name} [dim](failed to pull)[/dim]")
+                                console.print(
+                                    f"  [red]✗[/red] {short_name} [dim](failed to pull)[/dim]"
+                                )
                                 failed_images.append(img)
                                 pull_errors[img] = "failed to pull x86 fallback"
                                 continue
@@ -426,12 +465,12 @@ def session_join(ctx, session_id: str, force_setup: bool):
 
                     def show_progress(image: str, line: str):
                         # Show docker output directly - includes MB progress from PTY
-                        # Lines look like: "abc123: Downloading  360.9MB/4.075GB"
+                        # Lines look like: "abc123: Downloading [======>    ]  360.9MB/4.075GB"
                         if ": " in line:
                             parts = line.split(": ", 1)
                             if len(parts) == 2:
                                 layer_id = parts[0][-8:]
-                                layer_status = parts[1][:45]
+                                layer_status = parts[1][:70]  # Increased to include size info
                                 display = f"{layer_id}: {layer_status}"
                                 if display != last_status[0]:
                                     last_status[0] = display
@@ -450,6 +489,7 @@ def session_join(ctx, session_id: str, force_setup: bool):
                     # On Linux ARM64, verify pulled image architecture
                     if is_linux_arm:
                         from hte_cli.image_utils import get_image_architecture
+
                         pulled_arch = get_image_architecture(img)
 
                         if pulled_arch == "arm64":
@@ -462,7 +502,7 @@ def session_join(ctx, session_id: str, force_setup: bool):
                                 f"  [yellow]![/yellow] {short_name} [dim](downloaded, arch: amd64)[/dim]"
                             )
                             console.print(
-                                f"      [yellow]This is an x86 image - requires QEMU emulation on ARM[/yellow]"
+                                "      [yellow]This is an x86 image - requires QEMU emulation on ARM[/yellow]"
                             )
                             x86_images_on_arm.append(img)
                         else:
@@ -494,9 +534,7 @@ def session_join(ctx, session_id: str, force_setup: bool):
                 console.print(
                     f"[yellow]⚠ Warning:[/yellow] {len(x86_images_on_arm)} x86 image(s) detected on ARM host"
                 )
-                console.print(
-                    "  These require QEMU emulation. If container fails to start, run:"
-                )
+                console.print("  These require QEMU emulation. If container fails to start, run:")
                 console.print(
                     "  [bold]docker run --privileged --rm tonistiigi/binfmt --install all[/bold]"
                 )
@@ -513,14 +551,21 @@ def session_join(ctx, session_id: str, force_setup: bool):
 
                 # Architecture-specific advice
                 if is_linux_arm:
-                    console.print(f"  2. You're on Linux ARM64 - try: docker pull <image> --platform linux/arm64")
-                    console.print("  3. For x86-only images, enable QEMU: docker run --privileged --rm tonistiigi/binfmt --install all")
+                    console.print(
+                        "  2. You're on Linux ARM64 - try: docker pull <image> --platform linux/arm64"
+                    )
+                    console.print(
+                        "  3. For x86-only images, enable QEMU: docker run --privileged --rm tonistiigi/binfmt --install all"
+                    )
                 else:
                     console.print("  2. Try manual pull: docker pull <image>")
 
                 console.print("  4. Check network connectivity")
                 console.print()
-                console.print("Session remains active - you can retry with: hte-cli session join " + session_id)
+                console.print(
+                    "Session remains active - you can retry with: hte-cli session join "
+                    + session_id
+                )
                 sys.exit(1)
 
         # Send setup_completed - THIS STARTS THE TIMER ON SERVER
@@ -574,6 +619,10 @@ def session_join(ctx, session_id: str, force_setup: bool):
             eval_log_bytes = result.eval_log_path.read_bytes()
     except KeyboardInterrupt:
         events.docker_stopped(exit_code=130)
+        # Try to find and upload any partial eval log before exiting
+        eval_log_bytes = _find_eval_log_bytes(runner)
+        if eval_log_bytes:
+            _upload_partial_log(api, session_id, eval_log_bytes, console)
         console.print()
         console.print(
             "[yellow]Interrupted. Session remains active - you can reconnect later.[/yellow]"
@@ -581,6 +630,10 @@ def session_join(ctx, session_id: str, force_setup: bool):
         sys.exit(0)
     except Exception as e:
         events.docker_stopped(exit_code=1)
+        # Try to upload partial log on failure too
+        eval_log_bytes = _find_eval_log_bytes(runner)
+        if eval_log_bytes:
+            _upload_partial_log(api, session_id, eval_log_bytes, console)
         console.print(f"[red]Task execution failed: {e}[/red]")
         sys.exit(1)
     finally:
@@ -930,7 +983,9 @@ def diagnose_cmd():
                 console.print("  [red]✗[/red] QEMU x86 emulation NOT working")
                 console.print()
                 console.print("  [yellow]To enable QEMU emulation, run:[/yellow]")
-                console.print("  [bold]docker run --privileged --rm tonistiigi/binfmt --install all[/bold]")
+                console.print(
+                    "  [bold]docker run --privileged --rm tonistiigi/binfmt --install all[/bold]"
+                )
         except subprocess.TimeoutExpired:
             console.print("  [yellow]![/yellow] QEMU test timed out")
         except Exception as e:
@@ -974,7 +1029,10 @@ def _check_docker() -> tuple[bool, str | None]:
             timeout=10,
         )
         if result.returncode != 0:
-            return False, "Docker is not running. Start Docker (Docker Desktop, colima, or dockerd)."
+            return (
+                False,
+                "Docker is not running. Start Docker (Docker Desktop, colima, or dockerd).",
+            )
     except FileNotFoundError:
         return False, "Docker is not installed. Install from https://docs.docker.com/get-docker/"
     except Exception as e:
@@ -1023,5 +1081,41 @@ def _check_docker() -> tuple[bool, str | None]:
     return True, None
 
 
+def _ensure_nyuctf_network() -> None:
+    """Ensure the ctfnet Docker network exists for NYU CTF challenges.
+
+    NYU CTF tasks use a shared Docker network ('ctfnet') for communication
+    between the agent container and game-server container. This network must
+    exist before docker compose up is called, since it's declared as external.
+    """
+    import subprocess
+
+    NETWORK_NAME = "ctfnet"
+
+    try:
+        # Check if network exists
+        result = subprocess.run(
+            ["docker", "network", "inspect", NETWORK_NAME],
+            capture_output=True,
+            text=True,
+            timeout=10,
+        )
+        if result.returncode == 0:
+            return  # Network exists
+
+        # Create the network
+        subprocess.run(
+            ["docker", "network", "create", NETWORK_NAME],
+            capture_output=True,
+            text=True,
+            check=True,
+            timeout=10,
+        )
+    except subprocess.CalledProcessError:
+        pass  # Network creation failed, will error later with clearer message
+    except (subprocess.TimeoutExpired, FileNotFoundError):
+        pass  # Docker not available, will error later
+
+
 if __name__ == "__main__":
     cli()

{hte_cli-0.2.29 → hte_cli-0.2.31}/src/hte_cli/image_utils.py

@@ -124,6 +124,7 @@ def is_running_in_linux_vm_on_arm() -> bool:
         True if running Linux on ARM64
     """
     import sys
+
     return sys.platform == "linux" and get_host_architecture() in ("aarch64", "arm64")
 
 
@@ -313,10 +314,6 @@ def pull_image_with_progress(
 
         # Read output from master with timeout
         output_buffer = ""
-        # Regex to parse docker progress: "abc123: Downloading [===>  ] 10.5MB/50MB"
-        progress_pattern = re.compile(
-            r"([a-f0-9]+):\s*(Downloading|Extracting|Verifying Checksum|Download complete|Pull complete|Already exists|Waiting)(?:\s+\[.*?\]\s+)?(\d+\.?\d*\s*[kMG]?B)?(?:/(\d+\.?\d*\s*[kMG]?B))?"
-        )
 
         while True:
             # Check if process is done

{hte_cli-0.2.29 → hte_cli-0.2.31}/tests/unit/test_image_utils.py

@@ -610,7 +610,9 @@ class TestFixImageArchitecture:
     @patch("hte_cli.image_utils.remove_image")
     @patch("hte_cli.image_utils.check_image_architecture_matches_host")
     @patch("hte_cli.image_utils.platform.machine")
-    def test_returns_false_when_repull_fails(self, mock_machine, mock_check, mock_remove, mock_pull):
+    def test_returns_false_when_repull_fails(
+        self, mock_machine, mock_check, mock_remove, mock_pull
+    ):
         """Returns (False, message) when re-pull fails."""
         mock_machine.return_value = "aarch64"
         mock_check.return_value = (False, "amd64", "aarch64")

{hte_cli-0.2.29 → hte_cli-0.2.31}/uv.lock

@@ -625,7 +625,7 @@ wheels = [
 
 [[package]]
 name = "hte-cli"
-version = "0.2.29"
+version = "0.2.30"
 source = { editable = "." }
 dependencies = [
     { name = "click" },