Rahmath1/self_improving_agent

1from typing import List, Dict, Any, Optional, Union2from pydantic import BaseModel3 4from openenv.core import Action as BaseAction, Observation as BaseObservation5 6 7# ─────────────────────────────────────────8# Action9# ─────────────────────────────────────────10class AdvanceAgentAction(BaseAction):11    action_type: str = "clean_data"12 13 14# ─────────────────────────────────────────15# Observation16# ─────────────────────────────────────────17class AdvanceAgentObservation(BaseObservation):18    dataset_head: List[Dict]     = []19    columns:      List[str]      = []20    stats:        Dict[str, Any] = {}21    history:      List[str]      = []22    task:         str            = ""23    objective:    str            = ""24    difficulty:   str            = ""25 26    # ── Round 2: Self-improvement metadata ──────────────────27    # Included in every observation so agents can read curriculum state28    curriculum_level:      int   = 129    curriculum_level_name: str   = "🌱 Novice"30    episode_number:        int   = 031    rolling_avg_score:     float = 0.032    injection_detected:    bool  = False   # True if dataset contains injections33 34 35# Aliases for backward compatibility36Observation = AdvanceAgentObservation37Action      = AdvanceAgentAction38 39 40# ─────────────────────────────────────────41# Reward — internal use only42# ─────────────────────────────────────────43class Reward(BaseModel):44    score:      float45    feedback:   str46    is_penalty: bool = False