| """ |
| State types for the Data Cleaning environment. |
| |
| Data Cleaning Agent evaluates LLMs on their ability to identify and fix |
| data quality issues in CSV files using tool calls (read_file, run_python, |
| write_file, submit_cleaned_file). |
| |
| This environment uses the MCP protocol for tool interactions. Use |
| ``CallToolAction`` and ``ListToolsAction`` from ``openenv.core.env_server.mcp_types`` |
| to interact with the environment. |
| """ |
|
|
| from openenv.core.env_server import State |
|
|
|
|
| |
| AVAILABLE_TOOLS = ["read_file", "run_python", "write_file", "submit_cleaned_file"] |
|
|
|
|
| class DataCleanState(State): |
| """ |
| Internal environment state for tracking the current episode. |
| |
| All fields are set during reset() and are essential for episode tracking. |
| |
| Attributes: |
| task_level: Difficulty level (easy, medium, hard) |
| messy_file_path: Path to the input file with data issues |
| clean_file_path: Path to the expected cleaned output |
| task_description: Description of the cleaning task |
| workspace_dir: Working directory for agent operations |
| submitted: Whether the agent has submitted a cleaned file |
| # Inherited from State: episode_id, step_count |
| """ |
|
|
| task_level: str = "" |
| messy_file_path: str = "" |
| clean_file_path: str = "" |
| task_description: str = "" |
| workspace_dir: str = "" |
| submitted: bool = False |
|
|
