benchflow 0.3.4__tar.gz → 0.4.0__tar.gz

{benchflow-0.3.4 → benchflow-0.4.0}/.gitignore

@@ -184,3 +184,4 @@ tmp/
 tests/.smoke-jobs/
 context/
 tutorials/
+.playwright-mcp/

{benchflow-0.3.4 → benchflow-0.4.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: benchflow
-Version: 0.3.4
+Version: 0.4.0
 Summary: Multi-turn agent benchmarking with ACP — run any agent, any model, any provider.
 Project-URL: Homepage, https://github.com/benchflow-ai/benchflow
 Project-URL: Repository, https://github.com/benchflow-ai/benchflow
@@ -11,7 +11,7 @@ Author-email: Xiangyi Li <xiangyi@benchflow.ai>, Kyoung Whan Choe <choe.kyoung@g
 Maintainer-email: Xiangyi Li <xiangyi@benchflow.ai>, Kyoung Whan Choe <choe.kyoung@gmail.com>
 License: Apache-2.0
 License-File: LICENSE
-Keywords: acp,agent-evaluation,benchmark,llm-agents,multi-turn,skillsbench,terminal-bench
+Keywords: acp,agent-evaluation,benchmark,llm-agents,multi-turn,skillsbench
 Classifier: License :: OSI Approved :: Apache Software License
 Classifier: Operating System :: OS Independent
 Classifier: Programming Language :: Python :: 3
@@ -19,7 +19,6 @@ Classifier: Programming Language :: Python :: 3.12
 Classifier: Programming Language :: Python :: 3.13
 Requires-Python: >=3.12
 Requires-Dist: anyio>=4.0
-Requires-Dist: harbor==0.3.0
 Requires-Dist: httpx>=0.27.0
 Requires-Dist: pydantic>=2.0
 Requires-Dist: pyyaml>=6.0
@@ -33,6 +32,12 @@ Requires-Dist: pytest-asyncio>=0.24.0; extra == 'dev'
 Requires-Dist: pytest>=9.0.3; extra == 'dev'
 Requires-Dist: ruff>=0.7.0; extra == 'dev'
 Requires-Dist: ty>=0.0.1a1; extra == 'dev'
+Provides-Extra: sandbox-daytona
+Requires-Dist: daytona>=0.153.0; extra == 'sandbox-daytona'
+Requires-Dist: tenacity>=8.0; extra == 'sandbox-daytona'
+Provides-Extra: sandbox-modal
+Requires-Dist: modal>=0.73; extra == 'sandbox-modal'
+Requires-Dist: tenacity>=8.0; extra == 'sandbox-modal'
 Description-Content-Type: text/markdown
 
 <div align="center">
@@ -116,7 +121,7 @@ SkillsBench when you need its lockfile to point at the newest BenchFlow commit.
 
 ## Featured
 
-- **Progressive disclosure on SWE-bench Pro** — the `BaseUser` abstraction drives a multi-round rollout: terse round-0 prompt → failing-test hints → full spec. 5/5 oracle on Daytona, runnable demo at [`docs/examples/swebench_pro_progressive_disclosure.ipynb`](./docs/examples/swebench_pro_progressive_disclosure.ipynb). Also benchflow's [Harbor #1316](https://github.com/harbor-ai/harbor/issues/1316) parity answer for the no-second-LLM case. See [Progressive disclosure](./docs/progressive-disclosure.md).
+- **Progressive disclosure on SWE-bench Pro** — the `BaseUser` abstraction drives a multi-round rollout: terse round-0 prompt → failing-test hints → full spec. 5/5 oracle on Daytona, runnable demo at [`docs/examples/swebench_pro_progressive_disclosure.ipynb`](./docs/examples/swebench_pro_progressive_disclosure.ipynb). See [Progressive disclosure](./docs/progressive-disclosure.md).
 
 ## Research artifacts
 
@@ -130,7 +135,7 @@ Two runnable labs validate the security story:
 - **Eval researchers / paper writers** → [Getting started](./docs/getting-started.md) → [Concepts](./docs/concepts.md) → [Use cases](./docs/use-cases.md)
 - **Task authors** → [Task authoring](./docs/task-authoring.md) → [Sandbox hardening](./docs/sandbox-hardening.md)
 - **Agent builders integrating with benchflow** → [Concepts](./docs/concepts.md) → [Python API reference](./docs/reference/python-api.md) → [`benchflow.agents.registry`](./src/benchflow/agents/registry.py)
-- **Existing Harbor users migrating** → [Use cases — migration section](./docs/use-cases.md#migration-from-harbor) → [Progressive disclosure](./docs/progressive-disclosure.md#comparison-with-multi-agent-simulated-user)
+- **External benchmark adapters** → [Task authoring](./docs/task-authoring.md) → [Progressive disclosure](./docs/progressive-disclosure.md#comparison-with-multi-agent-simulated-user)
 
 ## Contributing
 

{benchflow-0.3.4 → benchflow-0.4.0}/README.md

@@ -79,7 +79,7 @@ SkillsBench when you need its lockfile to point at the newest BenchFlow commit.
 
 ## Featured
 
-- **Progressive disclosure on SWE-bench Pro** — the `BaseUser` abstraction drives a multi-round rollout: terse round-0 prompt → failing-test hints → full spec. 5/5 oracle on Daytona, runnable demo at [`docs/examples/swebench_pro_progressive_disclosure.ipynb`](./docs/examples/swebench_pro_progressive_disclosure.ipynb). Also benchflow's [Harbor #1316](https://github.com/harbor-ai/harbor/issues/1316) parity answer for the no-second-LLM case. See [Progressive disclosure](./docs/progressive-disclosure.md).
+- **Progressive disclosure on SWE-bench Pro** — the `BaseUser` abstraction drives a multi-round rollout: terse round-0 prompt → failing-test hints → full spec. 5/5 oracle on Daytona, runnable demo at [`docs/examples/swebench_pro_progressive_disclosure.ipynb`](./docs/examples/swebench_pro_progressive_disclosure.ipynb). See [Progressive disclosure](./docs/progressive-disclosure.md).
 
 ## Research artifacts
 
@@ -93,7 +93,7 @@ Two runnable labs validate the security story:
 - **Eval researchers / paper writers** → [Getting started](./docs/getting-started.md) → [Concepts](./docs/concepts.md) → [Use cases](./docs/use-cases.md)
 - **Task authors** → [Task authoring](./docs/task-authoring.md) → [Sandbox hardening](./docs/sandbox-hardening.md)
 - **Agent builders integrating with benchflow** → [Concepts](./docs/concepts.md) → [Python API reference](./docs/reference/python-api.md) → [`benchflow.agents.registry`](./src/benchflow/agents/registry.py)
-- **Existing Harbor users migrating** → [Use cases — migration section](./docs/use-cases.md#migration-from-harbor) → [Progressive disclosure](./docs/progressive-disclosure.md#comparison-with-multi-agent-simulated-user)
+- **External benchmark adapters** → [Task authoring](./docs/task-authoring.md) → [Progressive disclosure](./docs/progressive-disclosure.md#comparison-with-multi-agent-simulated-user)
 
 ## Contributing
 

{benchflow-0.3.4 → benchflow-0.4.0}/pyproject.toml

@@ -1,12 +1,11 @@
 [project]
 name = "benchflow"
-version = "0.3.4"
+version = "0.4.0"
 description = "Multi-turn agent benchmarking with ACP — run any agent, any model, any provider."
 readme = "README.md"
 requires-python = ">=3.12"
-keywords = ["benchmark", "llm-agents", "acp", "agent-evaluation", "multi-turn", "terminal-bench", "skillsbench"]
+keywords = ["benchmark", "llm-agents", "acp", "agent-evaluation", "multi-turn", "skillsbench"]
 dependencies = [
-    "harbor==0.3.0",
     "httpx>=0.27.0",
     "anyio>=4.0",
     "pydantic>=2.0",
@@ -42,6 +41,14 @@ dev = [
     "ruff>=0.7.0",
     "ty>=0.0.1a1",
 ]
+sandbox-daytona = [
+    "daytona>=0.153.0",
+    "tenacity>=8.0",
+]
+sandbox-modal = [
+    "modal>=0.73",
+    "tenacity>=8.0",
+]
 bedrock = [
     "boto3>=1.40",
 ]
@@ -114,5 +121,24 @@ ignore = [
 [tool.ty.environment]
 python-version = "3.12"
 
+[tool.ty.rules]
+# Many modules lazily import optional deps (daytona, modal, openai, toml, …).
+# These are guarded by try/except at runtime but ty can't resolve them in CI.
+unresolved-import = "ignore"
+
 [tool.ty.src]
 include = ["src"]
+# Modules that heavily use optional-dep types (daytona, modal, openai, boto3, …)
+# produce cascading type errors when those packages aren't installed.
+exclude = [
+    "src/benchflow/sandbox/daytona.py",
+    "src/benchflow/sandbox/modal_impl.py",
+    "src/benchflow/sandbox/docker.py",
+    "src/benchflow/sandbox/_base.py",
+    "src/benchflow/_env_setup.py",
+    "src/benchflow/rewards/llm.py",
+    "src/benchflow/rewards/file_readers.py",
+    "src/benchflow/rewards/rubric_config.py",
+    "src/benchflow/providers/bedrock_runtime.py",
+    "src/benchflow/experimental/mcp/reviewer_server.py",
+]

{benchflow-0.3.4 → benchflow-0.4.0}/src/benchflow/__init__.py

@@ -1,10 +1,12 @@
 """benchflow — ACP-native agent benchmarking framework.
 
-Re-exports environment APIs and adds:
+Public API surface:
+- Sandbox protocol for isolated execution environments
 - ACP client for multi-turn agent communication
 - Trajectory capture (HTTP proxy, OTel collector, ACP native)
 - Rollout lifecycle for single-task execution
 - Evaluation orchestration with retries and concurrency
+- Rewards protocol (composable Rubric + RewardFunc)
 - Metrics collection and aggregation
 """
 
@@ -12,23 +14,9 @@ from importlib.metadata import version as _version
 
 __version__ = _version("benchflow")
 
-# Re-export Harbor's core types for downstream task authors
-from harbor import (
-    BaseAgent,
-    BaseEnvironment,
-    ExecResult,
-    Task,
-    TaskConfig,
-    Verifier,
-    VerifierResult,
-)
-
-# benchflow's additions
-from benchflow._env_setup import stage_dockerfile_deps
-from benchflow._scene import MailboxTransport, Message, MessageTransport, SceneRole
-from benchflow._scene import Scene as SceneRuntime
-from benchflow._snapshot import list_snapshots, restore, snapshot
+# Core types
 from benchflow._types import Role, Scene, Turn
+from benchflow._utils.yaml_loader import rollout_config_from_yaml
 from benchflow.acp.client import ACPClient
 from benchflow.acp.session import ACPSession
 from benchflow.adapters import (
@@ -45,12 +33,6 @@ from benchflow.agents.registry import (
     list_agents,
     register_agent,
 )
-from benchflow.environments import (
-    SERVICES,
-    build_service_hooks,
-    detect_services_from_dockerfile,
-    register_service,
-)
 from benchflow.evaluation import (
     Evaluation,
     EvaluationConfig,
@@ -85,32 +67,39 @@ from benchflow.runtime import (
     RuntimeResult,
     run,
 )  # bf.run() — supports Agent, RolloutConfig, and str calling conventions
+from benchflow.sandbox import (
+    SERVICES,
+    ImageBuilder,
+    ImageConfig,
+    ImageRef,
+    Sandbox,
+    build_service_hooks,
+    detect_services_from_dockerfile,
+    register_service,
+)
 
-# Sandbox protocol (v0.4 — parallel types, Harbor not yet removed)
+# Sandbox protocol (v0.4)
 from benchflow.sandbox import ExecResult as SandboxExecResult
-from benchflow.sandbox import ImageBuilder, ImageConfig, ImageRef, Sandbox
+from benchflow.sandbox.protocol import ExecResult
+from benchflow.sandbox.setup import stage_dockerfile_deps
+from benchflow.sandbox.snapshot import list_snapshots, restore, snapshot
+from benchflow.sandbox.user import BaseUser, FunctionUser, PassthroughUser, RoundResult
+from benchflow.scenes import MailboxTransport, Message, MessageTransport, SceneRole
+from benchflow.scenes import Scene as SceneRuntime
 from benchflow.sdk import SDK
 from benchflow.skills import SkillInfo, discover_skills, install_skill, parse_skill
+from benchflow.task import (
+    Task,
+    TaskConfig,
+    Verifier,
+    VerifierResult,
+)
 from benchflow.trajectories.otel import OTelCollector
 from benchflow.trajectories.proxy import TrajectoryProxy
 from benchflow.trajectories.types import Trajectory
-from benchflow.trial_yaml import trial_config_from_yaml
-from benchflow.user import BaseUser, FunctionUser, PassthroughUser, RoundResult
-
-# Backward-compat aliases
-Trial = Rollout
-TrialConfig = RolloutConfig
-TrialRole = Role
-TrialScene = Scene
-RunResult = RolloutResult
-Job = Evaluation
-JobConfig = EvaluationConfig
-JobResult = EvaluationResult
 
 # Public API surface. Anything not in this list is implementation detail and
-# may change without notice. Names are grouped by source module to match the
-# imports above and to make it obvious to a future agent which module owns
-# what.
+# may change without notice.
 __all__ = [
     "__version__",
     # Rewards protocol (v0.4)
@@ -122,21 +111,17 @@ __all__ = [
     "LLMJudgeRewardFunc",
     "StringMatchRewardFunc",
     "CodeExecRewardFunc",
-    # Rubric config (ENG-55)
     "Criterion",
     "JudgeConfig",
     "RubricConfig",
     "ScoringConfig",
     "load_rubric_toml",
-    # Sandbox protocol (v0.4)
+    # Sandbox protocol
     "Sandbox",
     "SandboxExecResult",
     "ImageBuilder",
     "ImageConfig",
     "ImageRef",
-    # Harbor re-exports
-    "BaseAgent",
-    "BaseEnvironment",
     "ExecResult",
     "Task",
     "TaskConfig",
@@ -152,15 +137,11 @@ __all__ = [
     "is_vertex_model",
     "list_agents",
     "register_agent",
-    # Evaluation orchestration (new names)
+    # Evaluation orchestration
     "Evaluation",
     "EvaluationConfig",
     "EvaluationResult",
     "RetryConfig",
-    # Backward-compat aliases for Job
-    "Job",
-    "JobConfig",
-    "JobResult",
     # Metrics
     "BenchmarkMetrics",
     "collect_metrics",
@@ -168,8 +149,7 @@ __all__ = [
     "AgentInstallError",
     "AgentTimeoutError",
     "RolloutResult",
-    "RunResult",
-    # Runtime (0.3 compat)
+    # Runtime
     "Agent",
     "Environment",
     "Runtime",
@@ -177,7 +157,7 @@ __all__ = [
     "RuntimeResult",
     # Single entry point
     "run",
-    # Canonical declarative types (_types.py — ENG-47)
+    # Declarative types
     "Role",
     "Scene",
     "Turn",
@@ -191,23 +171,18 @@ __all__ = [
     "snapshot",
     "restore",
     "list_snapshots",
-    # Rollout (single execution path — ENG-46)
+    # Rollout
     "Rollout",
     "RolloutConfig",
-    # Backward-compat aliases for Trial
-    "Trial",
-    "TrialConfig",
-    "TrialRole",
-    "TrialScene",
-    "trial_config_from_yaml",
+    "rollout_config_from_yaml",
     # User abstraction (progressive disclosure)
     "BaseUser",
     "FunctionUser",
     "PassthroughUser",
     "RoundResult",
-    # SDK (backwards compat)
+    # SDK
     "SDK",
-    # Environments / dep staging
+    # Sandbox services
     "SERVICES",
     "build_service_hooks",
     "detect_services_from_dockerfile",
@@ -222,7 +197,7 @@ __all__ = [
     "OTelCollector",
     "TrajectoryProxy",
     "Trajectory",
-    # External adapters (ENG-51)
+    # External adapters
     "InspectAdapter",
     "ORSAdapter",
     "to_inspect_task",
@@ -231,8 +206,7 @@ __all__ = [
 
 
 def __getattr__(name: str):
-    """Fall through to harbor for names not explicitly re-exported."""
-    # Let Python's normal submodule resolution handle subpackages first.
+    """Lazy submodule resolution."""
     import importlib
 
     try:
@@ -240,16 +214,4 @@ def __getattr__(name: str):
     except ModuleNotFoundError as e:
         if e.name != f"benchflow.{name}":
             raise
-
-    import harbor
-
-    if hasattr(harbor, name):
-        import warnings
-
-        warnings.warn(
-            f"'{name}' is not directly re-exported by benchflow. Use 'from harbor import {name}' instead.",
-            ImportWarning,
-            stacklevel=2,
-        )
-        return getattr(harbor, name)
     raise AttributeError(f"module 'benchflow' has no attribute {name!r}")

benchflow-0.4.0/src/benchflow/_dotenv.py

@@ -0,0 +1,47 @@
+"""Small `.env` reader shared by CLI/runtime env resolution."""
+
+from __future__ import annotations
+
+import os
+from pathlib import Path
+
+_DEFAULT_DOTENV_PATH = Path(".env")
+_DOTENV_PATH_ENV = "BENCHFLOW_DOTENV_PATH"
+
+
+def load_dotenv_env(path: str | Path | None = None) -> dict[str, str]:
+    """Read a local `.env` file into a plain dict.
+
+    Missing files are treated as empty input. `BENCHFLOW_DOTENV_PATH` lets tests
+    or callers override the implicit `.env` lookup without changing cwd.
+    """
+    if path is not None:
+        dotenv_path = Path(path)
+    else:
+        dotenv_path = Path(os.environ.get(_DOTENV_PATH_ENV, _DEFAULT_DOTENV_PATH))
+    if not dotenv_path.exists() or not dotenv_path.is_file():
+        return {}
+
+    parsed: dict[str, str] = {}
+    for raw_line in dotenv_path.read_text().splitlines():
+        line = raw_line.strip()
+        if not line or line.startswith("#"):
+            continue
+        if line.startswith("export "):
+            line = line[len("export ") :].lstrip()
+        if "=" not in line:
+            continue
+
+        key, value = line.split("=", 1)
+        key = key.strip()
+        value = value.strip()
+        if not key:
+            continue
+
+        if value[:1] in {"'", '"'} and value[-1:] == value[:1]:
+            value = value[1:-1]
+        elif " #" in value:
+            value = value.split(" #", 1)[0].rstrip()
+
+        parsed[key] = value
+    return parsed

benchflow-0.4.0/src/benchflow/_utils/__init__.py

@@ -0,0 +1,11 @@
+"""benchflow._utils — small periphery I/O glue, private.
+
+Holds small (<200 LOC) periphery modules that translate between external
+artifacts (YAML files, git repos, scaffolded task dirs) and benchflow
+shapes.
+
+Members:
+    yaml_loader      — YAML → RolloutConfig/EvaluationConfig
+    benchmark_repos  — clone benchmark repos
+    task_authoring   — init_task / check_task scaffolding
+"""

benchflow-0.3.4/src/benchflow/task_download.py → benchflow-0.4.0/src/benchflow/_utils/benchmark_repos.py

@@ -138,7 +138,11 @@ def resolve_source(repo: str, path: str | None = None, ref: str | None = None) -
 # Format: (org/repo, ref, subpath)
 TASK_ALIASES: dict[str, tuple[str, str | None, str | None]] = {
     "skillsbench": ("benchflow-ai/skillsbench", "main", "tasks"),
-    "programbench": ("benchflow-ai/benchmarks", "main", "datasets/programbench/tasks"),
+    "programbench": (
+        "facebookresearch/programbench",
+        "main",
+        "src/programbench/data/tasks",
+    ),
     "harvey-lab": ("benchflow-ai/benchmarks", "main", "datasets/harvey-lab/tasks"),
 }
 

benchflow-0.4.0/src/benchflow/_utils/config.py

@@ -0,0 +1,22 @@
+"""Shared configuration normalization helpers."""
+
+from __future__ import annotations
+
+from benchflow.agents.registry import parse_agent_spec
+
+
+def normalize_agent_name(agent: str) -> str:
+    """Return the canonical registry name for an ACP agent alias."""
+    protocol, canonical = parse_agent_spec(agent)
+    if protocol == "acp":
+        return canonical
+    return agent
+
+
+def normalize_sandbox_user(sandbox_user: str | None) -> str | None:
+    """Map text root-user sentinels to ``None``."""
+    if sandbox_user is None:
+        return None
+    if sandbox_user.lower() in {"none", "null"}:
+        return None
+    return sandbox_user

benchflow-0.3.4/src/benchflow/trial_yaml.py → benchflow-0.4.0/src/benchflow/_utils/yaml_loader.py

@@ -1,6 +1,6 @@
-"""YAML trial config loader.
+"""YAML rollout config loader.
 
-Parses trial YAML files into TrialConfig with Scene support.
+Parses rollout YAML files into RolloutConfig with Scene support.
 Handles both new scene-based format and legacy flat format.
 
 New format::
@@ -44,13 +44,13 @@ from typing import Any
 import yaml
 
 from benchflow._types import Role, Scene, Turn
-from benchflow.trial import TrialConfig
+from benchflow.rollout import RolloutConfig
 
 logger = logging.getLogger(__name__)
 
 
-def load_trial_yaml(path: str | Path) -> dict:
-    """Load and normalize a trial YAML file."""
+def load_rollout_yaml(path: str | Path) -> dict:
+    """Load and normalize a rollout YAML file."""
     with open(path) as f:
         raw = yaml.safe_load(f)
     if not isinstance(raw, dict):
@@ -58,23 +58,23 @@ def load_trial_yaml(path: str | Path) -> dict:
     return raw
 
 
-def trial_config_from_yaml(
+def rollout_config_from_yaml(
     path: str | Path,
     task_path: Path | None = None,
-) -> TrialConfig:
-    """Parse a YAML file into a TrialConfig.
+) -> RolloutConfig:
+    """Parse a YAML file into a RolloutConfig.
 
     If task_path is provided, it overrides task_dir from the YAML.
     """
-    raw = load_trial_yaml(path)
-    return trial_config_from_dict(raw, task_path=task_path)
+    raw = load_rollout_yaml(path)
+    return rollout_config_from_dict(raw, task_path=task_path)
 
 
-def trial_config_from_dict(
+def rollout_config_from_dict(
     raw: dict[str, Any],
     task_path: Path | None = None,
-) -> TrialConfig:
-    """Convert a raw dict (from YAML or programmatic) into a TrialConfig."""
+) -> RolloutConfig:
+    """Convert a raw dict (from YAML or programmatic) into a RolloutConfig."""
     tp = task_path or Path(raw.get("task_dir", raw.get("task_path", ".")))
 
     # Scene-based format
@@ -105,7 +105,7 @@ def trial_config_from_dict(
     else:
         raise ValueError("YAML must have either 'scenes' or 'agent' at top level")
 
-    return TrialConfig(
+    return RolloutConfig(
         task_path=tp,
         scenes=scenes,
         environment=raw.get("environment", "docker"),
@@ -113,7 +113,7 @@ def trial_config_from_dict(
         sandbox_locked_paths=raw.get("sandbox_locked_paths"),
         sandbox_setup_timeout=raw.get("sandbox_setup_timeout", 120),
         job_name=raw.get("job_name"),
-        trial_name=raw.get("trial_name"),
+        rollout_name=raw.get("rollout_name"),
         jobs_dir=raw.get("jobs_dir", "jobs"),
         context_root=raw.get("context_root"),
         agent=raw.get("agent", "claude-agent-acp"),
@@ -165,9 +165,9 @@ def job_config_from_yaml(path: str | Path) -> dict:
     """Parse a YAML file and return both job-level and trial-level config.
 
     Returns a dict with keys: task_dir, concurrency, max_retries,
-    trial_config (TrialConfig), and any other job-level fields.
+    trial_config (RolloutConfig), and any other job-level fields.
     """
-    raw = load_trial_yaml(path)
+    raw = load_rollout_yaml(path)
     task_dir = Path(raw.get("task_dir", raw.get("tasks_dir", ".")))
     concurrency = raw.get("concurrency", 4)
     max_retries = raw.get("max_retries", 2)
@@ -176,6 +176,6 @@ def job_config_from_yaml(path: str | Path) -> dict:
         "task_dir": task_dir,
         "concurrency": concurrency,
         "max_retries": max_retries,
-        "trial_config": trial_config_from_dict(raw, task_path=task_dir),
+        "trial_config": rollout_config_from_dict(raw, task_path=task_dir),
         "raw": raw,
     }

{benchflow-0.3.4 → benchflow-0.4.0}/src/benchflow/acp/container_transport.py

@@ -5,7 +5,7 @@ import logging
 from pathlib import Path
 from typing import Any
 
-from benchflow.process import LiveProcess
+from benchflow.sandbox.process import LiveProcess
 
 from .transport import Transport, decode_json_rpc_message
 

benchflow-0.3.4/src/benchflow/_acp_run.py → benchflow-0.4.0/src/benchflow/acp/runtime.py

@@ -7,7 +7,7 @@ Owns the live agent-side of a run:
       ACP-native trajectory, and report tool-call counts
 
 The one allowed horizontal phase import in this refactor lives here:
-``from benchflow._sandbox import build_priv_drop_cmd``. connect_acp wraps
+``from benchflow.sandbox.lockdown import build_priv_drop_cmd``. connect_acp wraps
 the agent launch command in the sandbox user's privilege-drop prefix
 before handing it to the transport. It is a single pure-function call
 with no shared state — not a coupling of concerns.
@@ -22,13 +22,13 @@ import contextlib
 import logging
 from pathlib import Path
 
-from benchflow._sandbox import build_priv_drop_cmd
-from benchflow._trajectory import _capture_session_trajectory
 from benchflow.acp.client import ACPClient
 from benchflow.acp.container_transport import ContainerTransport
 from benchflow.agents.providers import find_provider, strip_provider_prefix
 from benchflow.agents.registry import AGENTS
-from benchflow.process import DaytonaProcess, DaytonaPtyProcess, DockerProcess
+from benchflow.sandbox.lockdown import build_priv_drop_cmd
+from benchflow.sandbox.process import DaytonaProcess, DaytonaPtyProcess, DockerProcess
+from benchflow.trajectories._capture import _capture_session_trajectory
 
 logger = logging.getLogger(__name__)
 
@@ -144,7 +144,7 @@ async def connect_acp(
     agent_env: dict,
     sandbox_user: str | None,
     model: str | None,
-    trial_dir: Path,
+    rollout_dir: Path,
     environment: str,
     agent_cwd: str,
 ) -> tuple[ACPClient, object, str]:
@@ -181,18 +181,18 @@ async def connect_acp(
 
         try:
             if environment == "docker":
-                live_proc = DockerProcess.from_harbor_env(env)
+                live_proc = DockerProcess.from_sandbox_env(env)
             else:
                 is_dind = hasattr(env, "_strategy") and hasattr(
                     env._strategy, "_compose_cmd"
                 )
                 if is_dind:
-                    live_proc = await DaytonaPtyProcess.from_harbor_env(env)
+                    live_proc = await DaytonaPtyProcess.from_sandbox_env(env)
                     logger.info("Using PTY transport for DinD compose task")
                 else:
-                    live_proc = await DaytonaProcess.from_harbor_env(env)
+                    live_proc = await DaytonaProcess.from_sandbox_env(env)
 
-            agent_log = trial_dir / "agent" / f"{agent.replace('-', '_')}.txt"
+            agent_log = rollout_dir / "agent" / f"{agent.replace('-', '_')}.txt"
             transport = ContainerTransport(
                 container_process=live_proc,
                 command=agent_launch,

{benchflow-0.3.4 → benchflow-0.4.0}/src/benchflow/acp/transport.py

@@ -6,7 +6,7 @@ import logging
 from abc import ABC, abstractmethod
 from typing import Any
 
-from benchflow.process import drain_oversized_line
+from benchflow.sandbox.process import drain_oversized_line
 
 logger = logging.getLogger(__name__)