Upload folder using huggingface_hub

Dockerfile

@@ -0,0 +1,81 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+# Multi-stage build using openenv-base
+# This Dockerfile is flexible and works for both:
+# - In-repo environments (with local OpenEnv sources)
+# - Standalone environments (with openenv from PyPI/Git)
+# The build script (openenv build) handles context detection and sets appropriate build args.
+
+ARG BASE_IMAGE=ghcr.io/meta-pytorch/openenv-base:latest
+FROM ${BASE_IMAGE} AS builder
+
+WORKDIR /app
+
+# Ensure git is available (required for installing dependencies from VCS)
+RUN apt-get update && \
+    apt-get install -y --no-install-recommends git && \
+    rm -rf /var/lib/apt/lists/*
+
+# Build argument to control whether we're building standalone or in-repo
+ARG BUILD_MODE=in-repo
+ARG ENV_NAME=myenv
+
+# Copy environment code (always at root of build context)
+COPY . /app/env
+
+# For in-repo builds, openenv is already vendored in the build context
+# For standalone builds, openenv will be installed via pyproject.toml
+WORKDIR /app/env
+
+# Ensure uv is available (for local builds where base image lacks it)
+RUN if ! command -v uv >/dev/null 2>&1; then \
+        curl -LsSf https://astral.sh/uv/install.sh | sh && \
+        mv /root/.local/bin/uv /usr/local/bin/uv && \
+        mv /root/.local/bin/uvx /usr/local/bin/uvx; \
+    fi
+    
+# Install dependencies using uv sync
+# If uv.lock exists, use it; otherwise resolve on the fly
+RUN --mount=type=cache,target=/root/.cache/uv \
+    if [ -f uv.lock ]; then \
+        uv sync --frozen --no-install-project --no-editable; \
+    else \
+        uv sync --no-install-project --no-editable; \
+    fi
+
+RUN --mount=type=cache,target=/root/.cache/uv \
+    if [ -f uv.lock ]; then \
+        uv sync --frozen --no-editable; \
+    else \
+        uv sync --no-editable; \
+    fi
+
+# Final runtime stage
+FROM ${BASE_IMAGE}
+
+WORKDIR /app
+
+# Copy the virtual environment from builder
+COPY --from=builder /app/env/.venv /app/.venv
+
+# Copy the environment code
+COPY --from=builder /app/env /app/env
+
+# Set PATH to use the virtual environment
+ENV PATH="/app/.venv/bin:$PATH"
+
+# Set PYTHONPATH so imports work correctly
+ENV PYTHONPATH="/app/env:$PYTHONPATH"
+
+ENV ENABLE_WEB_INTERFACE=true
+# Health check
+HEALTHCHECK --interval=30s --timeout=3s --start-period=5s --retries=3 \
+    CMD curl -f http://localhost:8000/health || exit 1
+
+# Run the FastAPI server
+# The module path is constructed to work with the /app/env structure
+CMD ["sh", "-c", "cd /app/env && uvicorn server.app:app --host 0.0.0.0 --port 8000"]

README.md

@@ -1,10 +1,142 @@
 ---
-title: Doc
-emoji: 📊
-colorFrom: yellow
-colorTo: blue
+title: Taskmanager Environment Server
+emoji: 🎬
+colorFrom: green
+colorTo: red
 sdk: docker
 pinned: false
+app_port: 8000
+base_path: /web
+tags:
+  - openenv
+  - rl
+  - scheduling
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# Taskmanager Environment
+
+A reinforcement learning environment that simulates a real-world engineering workflow. The agent must prioritize tickets (bugs, features, UI enhancements) to maximize business impact and avoid SLA violations. 
+
+## Quick Start
+
+The simplest way to use the Taskmanager environment is through the `TaskmanagerEnv` class:
+
+```python
+from taskmanager import TaskmanagerAction, TaskmanagerEnv
+
+try:
+    # Create environment from Docker image
+    env = TaskmanagerEnv.from_docker_image("taskmanager:latest")
+
+    # Reset to start a new episode
+    result = env.reset()
+    
+    print(f"Current Time: {result.observation.current_time}")
+    print(f"Available Tasks: {len(result.observation.tasks)}")
+
+    # Execute tasks until the episode is done
+    done = False
+    while not done:
+        # Simple policy: pick the first available task
+        if not result.observation.tasks:
+            break
+            
+        task_to_execute = result.observation.tasks[0]
+        task_id = task_to_execute["id"]
+        
+        # Take a step
+        result = env.step(TaskmanagerAction(task_id=task_id))
+        
+        print(f"Executed Task ID: {task_id}")
+        print(f"  → Reward: {result.reward}")
+        print(f"  → Current Time: {result.observation.current_time}")
+        print(f"  → Tasks Remaining: {len(result.observation.tasks)}")
+        
+        done = result.done
+
+    print("Episode completed!")
+
+finally:
+    # Always clean up
+    env.close()
+```
+
+## Environment Details
+
+### Action
+**TaskmanagerAction**: Contains a single field specifying the task to execute.
+- `task_id` (int) - The ID of the task/ticket to execute
+
+### Observation
+**TaskmanagerObservation**: Contains the current state of the environment.
+- `tasks` (List[Dict]) - List of remaining tickets. Each ticket has:
+  - `id` (int): Unique identifier
+  - `type` (str): "bug", "feature", or "enhancement"
+  - `effort` (int): Time required to complete the ticket
+  - `priority` (int): Importance (1-5)
+  - `deadline` (int): Target completion time
+- `current_time` (int) - Current time in the schedule
+- `steps_left` (int) - Steps remaining in the episode
+- `reward` (float) - Reward received from the previous action
+- `done` (bool) - Whether the episode is complete (all tasks done or max steps reached)
+- `metadata` (dict) - Additional info like step count
+
+### Reward Function
+The reward function is designed to simulate business impact:
+1. **Base Reward**: `priority * 3` if completed before the deadline.
+2. **Penalty**: If delayed, the reward is reduced based on the delay (`priority - delay * 0.5`), with a minimum of -2.
+3. **Type Multipliers**:
+   - **Bugs**: 2.0x multiplier (Critical)
+   - **Features**: 1.5x multiplier
+   - **Enhancements**: 1.0x multiplier
+4. **Invalid Action**: -1 reward for attempting a non-existent task ID.
+
+## Building the Docker Image
+
+Before using the environment, you need to build the Docker image.
+
+**To create the Docker image:**
+```bash
+# From project root
+docker build -t taskmanager .
+```
+
+**To run the Docker image locally:**
+```bash
+# Run the container in detached mode and map port 8000
+docker run -d -p 8000:8000 --name taskmanager_server taskmanager
+```
+
+## Deploying to Hugging Face Spaces
+
+You can easily deploy your OpenEnv environment to Hugging Face Spaces using the `openenv push` command:
+
+```bash
+# From the environment directory (where openenv.yaml is located)
+openenv push
+
+# Or specify options
+openenv push --namespace my-org --private
+```
+
+After deployment, your space will be available at:
+`https://huggingface.co/spaces/<repo-id>`
+
+The deployed space includes:
+- **Web Interface** at `/web` - Interactive UI for exploring the environment
+- **API Documentation** at `/docs` - Full OpenAPI/Swagger interface
+- **Health Check** at `/health` - Container health monitoring
+- **WebSocket** at `/ws` - Persistent session endpoint for low-latency interactions
+
+## Project Structure
+
+```
+taskmanager/
+├── client.py              # Environment client implementation
+├── models.py              # Action and Observation Pydantic models
+├── openenv.yaml           # OpenEnv manifest
+├── server/
+│   ├── app.py             # FastAPI application
+│   └── taskmanager_environment.py  # Core environment logic and reward function
+└── Dockerfile             # Container definition
+```

__init__.py

@@ -0,0 +1,16 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+"""Taskmanager Environment."""
+
+from .client import TaskmanagerEnv
+from .models import TaskmanagerAction, TaskmanagerObservation
+
+__all__ = [
+    "TaskmanagerAction",
+    "TaskmanagerObservation",
+    "TaskmanagerEnv",
+]

client.py

@@ -0,0 +1,72 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+
+"""Taskmanager Environment Client."""
+
+from typing import 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 TaskmanagerAction, TaskmanagerObservation
+
+
+class TaskmanagerEnv(EnvClient[TaskmanagerAction, TaskmanagerObservation, State]):
+    """
+    Client for the Task Scheduling Environment.
+
+    Supports:
+    - default reset()
+    - custom reset with user-defined tasks
+    """
+
+    # ================= STEP =================
+
+    def _step_payload(self, action: TaskmanagerAction) -> Dict:
+        return {
+            "task_id": action.task_id,
+        }
+
+    # ================= PARSE RESULT =================
+
+    def _parse_result(self, payload: Dict) -> StepResult[TaskmanagerObservation]:
+        obs_data = payload.get("observation", {})
+
+        observation = TaskmanagerObservation(
+            tasks=obs_data.get("tasks", []),
+            current_time=obs_data.get("current_time", 0),
+            steps_left=obs_data.get("steps_left", 0),
+            done=payload.get("done", False),
+            reward=payload.get("reward"),
+            metadata=obs_data.get("metadata", {}),
+        )
+
+        return StepResult(
+            observation=observation,
+            reward=payload.get("reward"),
+            done=payload.get("done", False),
+        )
+
+    # ================= PARSE STATE =================
+
+    def _parse_state(self, payload: Dict) -> State:
+        return State(
+            episode_id=payload.get("episode_id"),
+            step_count=payload.get("step_count", 0),
+        )
+
+    # ================= CUSTOM RESET =================
+
+    async def reset(
+        self, tasks: Optional[List[Dict]] = None, **kwargs
+    ) -> StepResult[TaskmanagerObservation]:
+        """
+        Reset the environment. If tasks are provided, they will be used instead of the predefined task list.
+        """
+        payload = kwargs.pop("config", {}) or {}
+        if tasks is not None:
+            payload["tasks"] = tasks
+
+        # Call underlying reset via super
+        return await super().reset(config=payload, **kwargs)

grader.py

@@ -0,0 +1,30 @@
+# grader.py
+
+
+def compute_score(total_reward, max_possible_reward):
+    """
+    Normalize score between 0 and 1
+    """
+    if max_possible_reward <= 0:
+        return 0.001
+
+    score = total_reward / max_possible_reward
+    return max(0.001, min(0.999, score))
+
+
+def evaluate_episode(rewards, max_per_step=15):
+    total_reward = sum(rewards)
+    max_possible = len(rewards) * max_per_step
+    return compute_score(total_reward, max_possible)
+
+
+def evaluate_task1(rewards):
+    return evaluate_episode(rewards, 15)
+
+
+def evaluate_task2(rewards):
+    return evaluate_episode(rewards, 15)
+
+
+def evaluate_task3(rewards):
+    return evaluate_episode(rewards, 15)

inference.py

@@ -0,0 +1,144 @@
+import asyncio
+import os
+from typing import List, Optional
+
+from openai import OpenAI
+
+from client import TaskmanagerEnv
+from models import TaskmanagerAction
+from grader import compute_score  # ✅ GRADER USED
+
+
+# ================= CONFIG =================
+
+API_KEY = os.environ.get("API_KEY", "dummy")
+API_BASE_URL = os.environ.get("API_BASE_URL", "http://localhost:4000")
+MODEL_NAME = os.environ.get("MODEL_NAME", "Qwen/Qwen2.5-72B-Instruct")
+
+BENCHMARK = "taskmanager"
+MAX_STEPS = 20
+SUCCESS_SCORE_THRESHOLD = 0.6
+
+# ================= LOGGING =================
+
+
+def log_start(task: str, env: str, model: str):
+    print(f"[START] task={task} env={env} model={model}", flush=True)
+
+
+def log_step(step: int, action: str, reward: float, done: bool, error: Optional[str]):
+    error_val = error if error else "null"
+    print(
+        f"[STEP] step={step} action={action} reward={reward:.2f} done={str(done).lower()} error={error_val}",
+        flush=True,
+    )
+
+
+def log_end(success: bool, steps: int, score: float, rewards: List[float]):
+    rewards_str = ",".join(f"{r:.2f}" for r in rewards)
+    print(
+        f"[END] success={str(success).lower()} steps={steps} score={score:.3f} rewards={rewards_str}",
+        flush=True,
+    )
+
+
+# ================= SMART POLICY =================
+
+
+def choose_best_ticket(tickets):
+    if not tickets:
+        return None
+
+    def score(ticket):
+        type_score = {"bug": 3, "feature": 2, "enhancement": 1}
+        return (
+            type_score.get(ticket["type"], 0),
+            ticket["priority"],
+            -ticket["deadline"],
+        )
+
+    best = sorted(tickets, key=score, reverse=True)[0]
+    return best["id"]
+
+
+# ================= MAIN =================
+
+
+async def main():
+    client = OpenAI(base_url=API_BASE_URL, api_key=API_KEY)
+    env = TaskmanagerEnv(base_url="http://localhost:8000")
+
+    try:
+        tasks_to_run = ["task-1-easy", "task-2-medium", "task-3-hard"]
+
+        for task_idx, TASK_NAME in enumerate(tasks_to_run):
+            rewards: List[float] = []
+            steps_taken = 0
+            success = False
+            score = 0.0
+
+            log_start(TASK_NAME, BENCHMARK, MODEL_NAME)
+
+            # 🔥 RESET ENV (tickets auto-generated)
+            result = await env.reset()
+            obs = result.observation
+
+            try:
+                client.chat.completions.create(
+                    model=MODEL_NAME,
+                    messages=[{"role": "user", "content": "hello"}],
+                    max_tokens=1,
+                )
+            except Exception:
+                pass
+
+            for step in range(1, MAX_STEPS + 1):
+                if result.done:
+                    break
+
+                tickets = obs.tasks
+                ticket_id = choose_best_ticket(tickets)
+
+                if ticket_id is None:
+                    break
+
+                # 🔥 STEP
+                result = await env.step(TaskmanagerAction(task_id=ticket_id))
+                obs = result.observation
+
+                reward = result.reward or 0.0
+                done = result.done
+                error = None
+
+                rewards.append(reward)
+                steps_taken = step
+
+                log_step(
+                    step=step,
+                    action=f"resolve_ticket_{ticket_id}",
+                    reward=reward,
+                    done=done,
+                    error=error,
+                )
+
+                if done:
+                    break
+
+            # ================= GRADER =================
+            total_reward = sum(rewards)
+            max_per_step = 15
+            max_possible = len(rewards) * max_per_step
+            score = compute_score(total_reward, max_possible)  # ✅ GRADER USED
+            success = score >= SUCCESS_SCORE_THRESHOLD
+
+            log_end(success, steps_taken, score, rewards)
+
+    finally:
+        try:
+            await env.close()
+        except Exception:
+            pass
+
+
+if __name__ == "__main__":
+    asyncio.run(main())

models.py

@@ -0,0 +1,31 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+"""
+Data models for the Taskmanager Environment.
+
+The taskmanager environment is a simple test environment that echoes back messages.
+"""
+
+from openenv.core.env_server.types import Action, Observation
+from pydantic import Field
+from typing import List, Dict
+
+
+class TaskmanagerAction(Action):
+    """Action for the Taskmanager environment - just a message to echo."""
+
+    task_id: int = Field(..., description="Task to execute")
+
+
+class TaskmanagerObservation(Observation):
+    """Observation for Task Scheduling Environment"""
+
+    tasks: List[Dict] = Field(
+        default_factory=list, description="List of remaining tickets"
+    )
+    current_time: int = Field(default=0, description="Current time in the schedule")
+    steps_left: int = Field(default=0, description="Steps remaining in episode")

openenv.yaml

@@ -0,0 +1,19 @@
+name: ai-ticket-prioritization-env
+
+description: >
+  A reinforcement learning environment simulating real-world software engineering
+  ticket prioritization. The agent must decide the optimal order to resolve tickets
+  (bugs, features, enhancements) under deadlines and resource constraints.
+
+entry_point: server.taskmanager_environment:TaskmanagerEnvironment
+
+action_space: TaskmanagerAction
+observation_space: TaskmanagerObservation
+
+max_steps: 20
+
+tags:
+  - openenv
+  - reinforcement-learning
+  - scheduling
+  - software-engineering

openenv_taskmanager.egg-info/PKG-INFO

@@ -0,0 +1,9 @@
+Metadata-Version: 2.4
+Name: openenv-taskmanager
+Version: 0.1.0
+Summary: Taskmanager environment for OpenEnv
+Requires-Python: >=3.10
+Requires-Dist: openenv-core[core]>=0.2.1
+Provides-Extra: dev
+Requires-Dist: pytest>=8.0.0; extra == "dev"
+Requires-Dist: pytest-cov>=4.0.0; extra == "dev"

openenv_taskmanager.egg-info/SOURCES.txt

@@ -0,0 +1,14 @@
+README.md
+pyproject.toml
+./__init__.py
+./client.py
+./models.py
+openenv_taskmanager.egg-info/PKG-INFO
+openenv_taskmanager.egg-info/SOURCES.txt
+openenv_taskmanager.egg-info/dependency_links.txt
+openenv_taskmanager.egg-info/entry_points.txt
+openenv_taskmanager.egg-info/requires.txt
+openenv_taskmanager.egg-info/top_level.txt
+server/__init__.py
+server/app.py
+server/taskmanager_environment.py

openenv_taskmanager.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

openenv_taskmanager.egg-info/entry_points.txt

@@ -0,0 +1,2 @@
+[console_scripts]
+server = taskmanager.server.app:main

openenv_taskmanager.egg-info/requires.txt

@@ -0,0 +1,5 @@
+openenv-core[core]>=0.2.1
+
+[dev]
+pytest>=8.0.0
+pytest-cov>=4.0.0

openenv_taskmanager.egg-info/top_level.txt

@@ -0,0 +1 @@
+taskmanager

pyproject.toml

@@ -0,0 +1,45 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+[build-system]
+requires = ["setuptools>=45", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "openenv-taskmanager"
+version = "0.1.0"
+description = "Taskmanager environment for OpenEnv"
+requires-python = ">=3.10"
+dependencies = [
+    # Core OpenEnv runtime (provides FastAPI server + HTTP client types)
+    # install from github
+    # "openenv-core[core] @ git+https://github.com/meta-pytorch/OpenEnv.git",
+    "openenv-core[core]>=0.2.1",
+    # Environment-specific dependencies
+    # Add all dependencies needed for your environment here
+    # Examples:
+    # "numpy>=1.19.0",
+    # "torch>=2.0.0",
+    # "gymnasium>=0.29.0",
+    # "openspiel>=1.0.0",
+    # "smolagents>=1.22.0,<2",
+]
+
+[project.optional-dependencies]
+dev = [
+    "pytest>=8.0.0",
+    "pytest-cov>=4.0.0",
+]
+
+[project.scripts]
+# Server entry point - enables running via: uv run --project . server
+# or: python -m taskmanager.server.app
+server = "taskmanager.server.app:main"
+
+[tool.setuptools]
+include-package-data = true
+packages = ["taskmanager", "taskmanager.server"]
+package-dir = { "taskmanager" = ".", "taskmanager.server" = "server" }

server/__init__.py

@@ -0,0 +1,11 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+"""Taskmanager environment server components."""
+
+from .taskmanager_environment import TaskmanagerEnvironment
+
+__all__ = ["TaskmanagerEnvironment"]

server/app.py

@@ -0,0 +1,108 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+
+"""
+FastAPI application for the Taskmanager Environment.
+"""
+
+from fastapi.responses import HTMLResponse
+from fastapi import APIRouter
+import os
+
+try:
+    from openenv.core.env_server.http_server import create_app
+except Exception as e:
+    raise ImportError(
+        "openenv is required. Install dependencies with 'uv sync'"
+    ) from e
+
+try:
+    from ..models import TaskmanagerAction, TaskmanagerObservation
+    from .taskmanager_environment import TaskmanagerEnvironment
+except ImportError:
+    from models import TaskmanagerAction, TaskmanagerObservation
+    from server.taskmanager_environment import TaskmanagerEnvironment
+
+
+# ================= CREATE APP =================
+
+app = create_app(
+    TaskmanagerEnvironment,
+    TaskmanagerAction,
+    TaskmanagerObservation,
+    env_name="taskmanager",
+    max_concurrent_envs=1,
+)
+
+# ================= ROUTER =================
+
+router = APIRouter()
+
+# 🔥 Serve demo UI
+@router.get("/", response_class=HTMLResponse)
+def home():
+    file_path = os.path.join(os.path.dirname(__file__), "..", "demo.html")
+    with open(file_path, "r", encoding="utf-8") as f:
+        return f.read()
+
+
+# 🔥 Run full agent (for UI)
+@router.get("/run-agent")
+def run_agent():
+    env = TaskmanagerEnvironment()
+
+    obs = env.reset()
+    total_reward = 0
+    steps = []
+
+    for _ in range(20):
+        tickets = obs.tasks
+
+        if not tickets:
+            break
+
+        # 🔥 same logic as inference
+        def score(t):
+            type_score = {"bug": 3, "feature": 2, "enhancement": 1}
+            return (
+                type_score.get(t["type"], 0),
+                t["priority"],
+                -t["deadline"]
+            )
+
+        best = sorted(tickets, key=score, reverse=True)[0]
+
+        obs = env.step(type("obj", (), {"task_id": best["id"]}))
+
+        total_reward += obs.reward
+
+        steps.append({
+            "chosen": best,
+            "remaining": obs.tasks,
+            "reward": obs.reward,
+            "time": obs.current_time
+        })
+
+        if obs.done:
+            break
+
+    # 🔥 normalize score
+    score = max(0.0, min(1.0, total_reward / 200))
+
+    return {
+        "steps": steps,
+        "final_score": score
+    }
+
+
+app.include_router(router)
+
+# ================= MAIN =================
+
+def main(host: str = "0.0.0.0", port: int = 8000):
+    import uvicorn
+    uvicorn.run(app, host=host, port=port)
+
+
+if __name__ == "__main__":
+    main()

server/requirements.txt

@@ -0,0 +1,6 @@
+openenv[core]>=0.2.0
+fastapi>=0.115.0
+uvicorn>=0.24.0
+
+
+

server/taskmanager_environment.py

@@ -0,0 +1,169 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+
+"""
+AI Ticket Prioritization Environment (Jira-like)
+
+Simulates a real-world engineering workflow:
+- Bug fixes (critical)
+- Feature development
+- UI enhancements
+
+Agent must prioritize tickets to maximize business impact and avoid SLA violations.
+"""
+
+from dataclasses import dataclass
+from uuid import uuid4
+import random
+
+try:
+    from openenv.core.env_server.interfaces import Environment
+    from openenv.core.env_server.types import State
+except ImportError:
+
+    class Environment:
+        pass
+
+    @dataclass
+    class State:
+        episode_id: str
+        step_count: int
+
+
+try:
+    from ..models import TaskmanagerAction, TaskmanagerObservation
+except ImportError:
+    from models import TaskmanagerAction, TaskmanagerObservation
+
+
+class TaskmanagerEnvironment(Environment):
+    SUPPORTS_CONCURRENT_SESSIONS: bool = True
+
+    def __init__(self):
+        self._state = State(episode_id=str(uuid4()), step_count=0)
+        self.current_time = 0
+        self.tickets = []
+        self.max_steps = 20
+
+        self.total_reward = 0
+        self.episode_count = 0
+
+    # ================= TICKET GENERATOR =================
+
+    def generate_tickets(self, num_tickets):
+        tickets = []
+        current_time = 0
+
+        for i in range(num_tickets):
+            ticket_type = random.choice(["bug", "feature", "enhancement"])
+
+            effort = random.randint(1, 3)
+
+            # 🔥 ensure feasible deadline
+            slack = random.randint(3, 8)
+            deadline = current_time + effort + slack
+
+            priority = random.randint(1, 5)
+
+            ticket = {
+                "id": i + 1,
+                "deadline": deadline,
+                "priority": priority,
+                "effort": effort,
+                "type": ticket_type,
+            }
+
+            tickets.append(ticket)
+
+            # update time so sequence is solvable
+            current_time += effort
+
+        # 🔥 shuffle so agent must think
+        random.shuffle(tickets)
+
+        return tickets
+
+    # ================= RESET =================
+
+    def reset(self) -> TaskmanagerObservation:
+        self._state = State(episode_id=str(uuid4()), step_count=0)
+        self.current_time = 0
+
+        self.episode_count += 1
+
+        avg_reward = self.total_reward / max(1, self.episode_count)
+
+        if avg_reward < 5:
+            num_tickets = 5
+        elif avg_reward < 15:
+            num_tickets = 8
+        else:
+            num_tickets = 12
+
+        self.tickets = self.generate_tickets(num_tickets)
+
+        print(
+            f"Episode {self.episode_count} | Tickets: {num_tickets} | Avg reward: {avg_reward:.2f}"
+        )
+
+        return TaskmanagerObservation(
+            tasks=self.tickets,  # ⚠️ keep 'tasks' for compatibility
+            current_time=self.current_time,
+            steps_left=self.max_steps,
+            reward=0.0,
+            done=False,
+        )
+
+    # ================= STEP =================
+
+    def step(self, action: TaskmanagerAction) -> TaskmanagerObservation:
+        self._state.step_count += 1
+
+        reward = 0
+
+        ticket = next((t for t in self.tickets if t["id"] == action.task_id), None)
+
+        if ticket:
+            self.current_time += ticket["effort"]
+
+            # 🎯 BASE REWARD
+            if self.current_time <= ticket["deadline"]:
+                reward = ticket["priority"] * 3  # boosted reward for being on time
+            else:
+                delay = self.current_time - ticket["deadline"]
+                # Soft penalty: base priority minus a small delay fraction (can still be positive if slightly late)
+                reward = max(-2, ticket["priority"] - (delay * 0.5))
+
+            # 🔥 TYPE MULTIPLIER (REAL-WORLD LOGIC)
+            if ticket["type"] == "bug":
+                reward *= 2  # critical
+            elif ticket["type"] == "feature":
+                reward *= 1.5
+            else:  # enhancement
+                reward *= 1
+
+            # remove ticket
+            self.tickets = [t for t in self.tickets if t["id"] != action.task_id]
+
+        else:
+            reward = -1
+
+        self.total_reward += reward
+
+        done = len(self.tickets) == 0 or self._state.step_count >= self.max_steps
+
+        return TaskmanagerObservation(
+            tasks=self.tickets,
+            current_time=self.current_time,
+            steps_left=self.max_steps - self._state.step_count,
+            reward=reward,
+            done=done,
+            metadata={
+                "step": self._state.step_count,
+                "remaining_tickets": len(self.tickets),
+            },
+        )
+
+    @property
+    def state(self) -> State:
+        return self._state