xXArjunXx/MetaHackathon

Email Triage OpenEnv

A real-world environment for training and evaluating AI agents on email triage tasks. Implements the full OpenEnv specification for agent learning.

Overview

This environment simulates the task of triaging and organizing emails into appropriate categories and actions. An AI agent must learn to:

• —Classify emails by priority and type
• —Identify spam and suspicious emails
• —Execute appropriate actions (archive, delete, reply, forward)
• —Handle complex, contextual decision-making

Perfect for training agents on realistic email management workflows.

Environment Specification

Tasks

1. 1.Easy - Basic Email Management
2. 2.Sort 5 emails into appropriate categories
3. 3.Actions: Read, Archive, Delete
4. 4.Expected difficulty: Straightforward classification

1. 1.Medium - Smart Email Organization
2. 2.Triage 8 emails with attachment handling
3. 3.Actions: Read, Archive, Mark Important, Flag
4. 4.Expected difficulty: Requires contextual understanding

1. 1.Hard - Expert Email Management
2. 2.Manage 10 complex emails requiring nuanced decisions
3. 3.Actions: All above + Reply, Forward, Report Spam
4. 4.Expected difficulty: Requires deep contextual understanding

Observation Space

{
  "task": str,
  "step_count": int,
  "cumulative_reward": float,
  "normalized_score": float,
  "actions_taken": dict,
  "progress": str,
  "done": bool
}

Action Space

Valid actions: read, archive, delete, flag, mark_spam, reply, forward

Reward Function

• —Correct action: +0.1 to +0.3
• —Incorrect action: -0.1 to -0.2
• —Episode bonus: +0.5 for 100% accuracy

Setup & Usage

Installation

pip install -r server/requirements.txt
pip install openenv-core

Running Locally

cd server
python main.py

Docker

docker build -t email-triage-openenv .
docker run -p 8000:8000 email-triage-openenv

Baseline Performance

API Endpoints

• —POST /reset - Start new episode
• —POST /step - Execute action
• —GET /state - Get current state
• —GET /grade - Get scores

OpenEnv Compliance

✅ Full OpenEnv spec ✅ Typed Pydantic models ✅ Deterministic graders ✅ Meaningful rewards

openenv validate

License

MIT