""" Medical Coding Auditor Environment Client. Provides a typed EnvClient for connecting to the MedicalCodingEnvironment server over HTTP/WebSocket. Supports both direct URL connections and Docker-based launch. """ from typing import Any, Dict, List, Optional from openenv.core import EnvClient from openenv.core.client_types import StepResult from openenv.core.env_server.types import State from .models import MedicalCodingAction, MedicalCodingObservation class MedicalCodingEnv( EnvClient[MedicalCodingAction, MedicalCodingObservation, State] ): """ Client for the Medical Coding Auditor Environment. Maintains a persistent WebSocket connection to the environment server. Each client instance has its own dedicated session. Example — connect to a running server: >>> env = MedicalCodingEnv(base_url="http://localhost:7860") >>> result = await env.reset(task_id="easy_demographic") >>> obs = result.observation >>> print(obs.clinical_note) >>> >>> result = await env.step( ... MedicalCodingAction(action_type="query_guideline", code="O80") ... ) >>> print(result.observation.tool_result) Example — launch from Docker image: >>> env = await MedicalCodingEnv.from_docker_image("medical-coding-env:latest") >>> try: ... result = await env.reset(task_id="medium_ncci_conflict") ... result = await env.step( ... MedicalCodingAction( ... action_type="check_ncci_edits", ... code1="93306", ... code2="93307", ... ) ... ) ... finally: ... await env.close() """ def _step_payload(self, action: MedicalCodingAction) -> Dict[str, Any]: """Serialize MedicalCodingAction to JSON payload for the step endpoint.""" payload: Dict[str, Any] = { "action_type": action.action_type, } if action.code is not None: payload["code"] = action.code if action.code1 is not None: payload["code1"] = action.code1 if action.code2 is not None: payload["code2"] = action.code2 if action.error_type is not None: payload["error_type"] = action.error_type if action.justification is not None: payload["justification"] = action.justification if action.question is not None: payload["question"] = action.question if action.evidence_text is not None: payload["evidence_text"] = action.evidence_text if action.metadata: payload["metadata"] = action.metadata return payload def _parse_result(self, payload: Dict[str, Any]) -> StepResult[MedicalCodingObservation]: """Parse server response into StepResult[MedicalCodingObservation].""" obs_data = payload.get("observation", {}) observation = MedicalCodingObservation( task_id=obs_data.get("task_id", ""), difficulty=obs_data.get("difficulty", ""), patient_demographics=obs_data.get("patient_demographics", {}), clinical_note=obs_data.get("clinical_note", ""), proposed_codes=obs_data.get("proposed_codes", {}), draft_report=obs_data.get("draft_report", []), tool_result=obs_data.get("tool_result", ""), step_count=obs_data.get("step_count", 0), codes_queried=obs_data.get("codes_queried", []), pairs_checked=obs_data.get("pairs_checked", []), clarifications_asked=obs_data.get("clarifications_asked", []), extracted_evidence=obs_data.get("extracted_evidence", []), last_action_error=obs_data.get("last_action_error"), grader_score=obs_data.get("grader_score"), episode_metrics=obs_data.get("episode_metrics"), done=payload.get("done", False), reward=payload.get("reward", 0.0), ) return StepResult( observation=observation, reward=payload.get("reward", 0.0), done=payload.get("done", False), ) def _parse_state(self, payload: Dict[str, Any]) -> State: """Parse server response into State object.""" return State( episode_id=payload.get("episode_id"), step_count=payload.get("step_count", 0), )