Merge pull request #34 from cooperbench/mini_Swe_agent_v2

Adding support for minisweagent_v2

Author: akhatua2

pyproject.toml

@@ -85,6 +85,7 @@ swe-agent = [
     "unidiff>=0.7",
 ]
 gcp = [
+    "cooperbench",
     "google-cloud-batch>=0.17",
     "google-cloud-compute>=1.0",
     "google-cloud-storage>=2.0",
@@ -186,7 +187,7 @@ exclude_lines = [
 ]
 
 [tool.uv.sources]
-cooperbench = { workspace = true }
 openhands-sdk = { path = "src/cooperbench/agents/openhands_agent_sdk/openhands-sdk", editable = true }
 openhands-tools = { path = "src/cooperbench/agents/openhands_agent_sdk/openhands-tools", editable = true }
 openhands-workspace = { path = "src/cooperbench/agents/openhands_agent_sdk/openhands-workspace", editable = true }
+cooperbench = { workspace = true }

src/cooperbench/agents/__init__.py

@@ -104,6 +104,7 @@ def run(
 # Add your agent's shorthand here when registering a new adapter
 AGENT_SHORTHANDS = {
     "mini_swe_agent": "msa",
+    "mini_swe_agent_v2": "msa_v2",
     "swe_agent": "sw",
     "openhands_sdk": "oh",
 }

src/cooperbench/agents/mini_swe_agent/environments/modal.py

@@ -69,8 +69,7 @@ def _get_or_build_image(image_name: str) -> modal.Image:
         if image_name in _image_cache:
             return _image_cache[image_name]
 
-        # Build and cache (CACHE_BUST env forces new image hash) TODO: @arpan remove this after testing
-        image = modal.Image.from_registry(image_name).entrypoint([]).env({"COOPERBENCH_CACHE": "v6"})
+        image = modal.Image.from_registry(image_name).entrypoint([])
         _image_cache[image_name] = image
         return image
 

src/cooperbench/agents/mini_swe_agent_v2/__init__.py

@@ -0,0 +1,102 @@
+"""
+mini-swe-agent v2 - A Minimal AI Agent for Software Engineering (Tool-calling version)
+
+Source: https://github.com/SWE-agent/mini-swe-agent
+Version: 2.1.0 (commit 56613dd)
+License: MIT
+Copyright (c) 2025 Kilian A. Lieret and Carlos E. Jimenez
+
+This code is copied directly from mini-swe-agent with modifications:
+- Import paths changed from 'minisweagent' to 'cooperbench.agents.mini_swe_agent_v2'
+- Removed text-based (regex) action parsing (v1 compat), keeping only tool calling
+- Removed interactive agent, non-Docker environments, non-litellm models
+- Added inter-agent communication (messaging + git connectors) for CooperBench
+
+Citation:
+    @misc{minisweagent2025,
+        title={mini-swe-agent: A Minimal AI Agent for Software Engineering},
+        author={Lieret, Kilian A. and Jimenez, Carlos E.},
+        year={2025},
+        url={https://github.com/SWE-agent/mini-swe-agent}
+    }
+
+This file provides:
+- Path settings for global config file & relative directories
+- Version numbering
+- Protocols for the core components of mini-swe-agent.
+"""
+
+__version__ = "2.1.0"
+
+import os
+from pathlib import Path
+from typing import Any, Protocol
+
+import dotenv
+from platformdirs import user_config_dir
+
+from cooperbench.agents.mini_swe_agent_v2.utils.log import logger
+
+package_dir = Path(__file__).resolve().parent
+
+global_config_dir = Path(os.getenv("MSWEA_GLOBAL_CONFIG_DIR") or user_config_dir("mini-swe-agent"))
+global_config_dir.mkdir(parents=True, exist_ok=True)
+global_config_file = Path(global_config_dir) / ".env"
+
+dotenv.load_dotenv(dotenv_path=global_config_file)
+
+
+# === Protocols ===
+# You can ignore them unless you want static type checking.
+
+
+class Model(Protocol):
+    """Protocol for language models."""
+
+    config: Any
+
+    def query(self, messages: list[dict[str, str]], **kwargs) -> dict: ...
+
+    def format_message(self, **kwargs) -> dict: ...
+
+    def format_observation_messages(
+        self, message: dict, outputs: list[dict], template_vars: dict | None = None
+    ) -> list[dict]: ...
+
+    def get_template_vars(self, **kwargs) -> dict[str, Any]: ...
+
+    def serialize(self) -> dict: ...
+
+
+class Environment(Protocol):
+    """Protocol for execution environments."""
+
+    config: Any
+
+    def execute(self, action: dict, cwd: str = "") -> dict[str, Any]: ...
+
+    def get_template_vars(self, **kwargs) -> dict[str, Any]: ...
+
+    def serialize(self) -> dict: ...
+
+
+class Agent(Protocol):
+    """Protocol for agents."""
+
+    config: Any
+
+    def run(self, task: str, **kwargs) -> dict: ...
+
+    def save(self, path: Path | None, *extra_dicts) -> dict: ...
+
+
+__all__ = [
+    "Agent",
+    "Model",
+    "Environment",
+    "package_dir",
+    "__version__",
+    "global_config_file",
+    "global_config_dir",
+    "logger",
+]

src/cooperbench/agents/mini_swe_agent_v2/adapter.py

@@ -0,0 +1,145 @@
+"""Mini-SWE-Agent v2 adapter for CooperBench.
+
+This adapter wraps the mini-swe-agent v2 framework (tool-calling version)
+to conform to the AgentRunner interface used by CooperBench.
+"""
+
+import yaml
+
+from cooperbench.agents import AgentResult
+from cooperbench.agents.mini_swe_agent_v2.agents.default import DefaultAgent
+from cooperbench.agents.mini_swe_agent_v2.config import get_config_path
+from cooperbench.agents.mini_swe_agent_v2.connectors import GitConnector
+from cooperbench.agents.mini_swe_agent_v2.connectors.messaging import MessagingConnector
+from cooperbench.agents.mini_swe_agent_v2.models.litellm_model import LitellmModel
+from cooperbench.agents.mini_swe_agent_v2.models.utils.actions_toolcall import SEND_MESSAGE_TOOL
+from cooperbench.agents.registry import register
+
+
+@register("mini_swe_agent_v2")
+class MiniSweAgentV2Runner:
+    """Adapter for mini-swe-agent v2 framework (tool-calling)."""
+
+    def run(
+        self,
+        task: str,
+        image: str,
+        *,
+        agent_id: str = "agent",
+        model_name: str = "gpt-4o",
+        agents: list[str] | None = None,
+        comm_url: str | None = None,
+        git_server_url: str | None = None,
+        git_enabled: bool = False,
+        messaging_enabled: bool = True,
+        config: dict | None = None,
+        agent_config: str | None = None,
+        log_dir: str | None = None,
+    ) -> AgentResult:
+        """Run mini-swe-agent v2 on a task."""
+        # Always load default config, then merge with any overrides
+        config_path = get_config_path("mini")
+        with open(config_path) as f:
+            default_config = yaml.safe_load(f)
+
+        # Merge passed config overrides into default config
+        if config is not None:
+            default_config.update(config)
+
+        agent_cfg = default_config.get("agent", {})
+        model_cfg = default_config.get("model", {})
+        env_cfg = default_config.get("environment", {})
+        backend = default_config.get("backend", "modal")
+
+        # Create environment based on backend
+        env_kwargs = {
+            "image": image,
+            "cwd": "/workspace/repo",
+            "timeout": 3600,
+        }
+        if env_cfg.get("env"):
+            env_kwargs["env"] = env_cfg["env"]
+
+        if backend == "docker":
+            from cooperbench.agents.mini_swe_agent_v2.environments.docker import DockerEnvironment
+
+            if config and config.get("git_network"):
+                env_kwargs["network"] = config["git_network"]
+            env = DockerEnvironment(**env_kwargs)
+        else:
+            from cooperbench.agents.mini_swe_agent_v2.environments.modal import ModalEnvironment
+
+            env = ModalEnvironment(**env_kwargs)
+
+        # Capture base commit for patch generation
+        base_commit_result = env.execute({"command": "git rev-parse HEAD"})
+        base_commit = base_commit_result.get("output", "").strip()
+
+        # Setup messaging connector if enabled
+        comm = None
+        use_messaging = messaging_enabled and comm_url and agents and len(agents) > 1
+        if use_messaging:
+            comm = MessagingConnector(agent_id=agent_id, agents=agents, url=comm_url)
+
+        # Create LLM model with send_message tool if messaging is enabled
+        extra_tools = [SEND_MESSAGE_TOOL] if use_messaging else None
+        model = LitellmModel(model_name=model_name, extra_tools=extra_tools, **model_cfg)
+
+        # Setup git connector if enabled
+        if git_enabled and git_server_url and agents:
+            git_connector = GitConnector(
+                agent_id=agent_id,
+                agents=agents,
+                server_url=git_server_url,
+            )
+            git_connector.setup(env)
+
+        # Create agent with template variables for collaboration
+        extra_vars = {
+            "agent_id": agent_id if (agents and len(agents) > 1) else None,
+            "agents": agents if agents else [],
+            "git_enabled": git_enabled,
+            "messaging_enabled": messaging_enabled,
+        }
+
+        agent = DefaultAgent(
+            model=model,
+            env=env,
+            comm=comm,
+            agent_id=agent_id,
+            **agent_cfg,
+        )
+        agent.extra_template_vars.update(extra_vars)
+
+        # Run agent
+        error_msg = None
+        try:
+            result = agent.run(task=task)
+            status = result.get("exit_status", "Submitted")
+        except Exception as e:
+            status = "Error"
+            error_msg = str(e)
+
+        # Extract patch (committed + uncommitted changes)
+        patch = self._get_patch(env, base_commit)
+
+        # Cleanup
+        env.cleanup()
+
+        return AgentResult(
+            status=status,
+            patch=patch,
+            cost=agent.cost,
+            steps=agent.n_calls,
+            messages=agent.messages,
+            sent_messages=agent.sent_messages,
+            error=error_msg,
+        )
+
+    def _get_patch(self, env, base_commit: str) -> str:
+        """Extract git diff from base commit to current working tree state."""
+        try:
+            result = env.execute({"command": f"git diff {base_commit}"})
+            return result.get("output", "").strip()
+        except Exception:
+            return ""

src/cooperbench/agents/mini_swe_agent_v2/agents/__init__.py

@@ -0,0 +1 @@
+"""Agent implementations for mini-SWE-agent v2."""