Spaces:
Sleeping
Sleeping
File size: 2,259 Bytes
be63ecf d04cd46 be63ecf 5333423 d04cd46 e15ba9b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 | ---
title: Mail Triage Agent
emoji: 👀
colorFrom: blue
colorTo: green
sdk: docker
pinned: false
license: mit
app_port: 8000
---
Mail Triage Agent v4 (Security Evaluation)
This repository contains a high-fidelity environment and an autonomous agent designed for Email Security Triage. The project focuses on detecting sophisticated threats like "Digital Seduction" (Phishing), typo-squatted domains, and malicious URL redirections.
🚀 Overview
The system consists of two primary components:
Environment (env.py): A FastAPI-based server implementing the OpenEnv specification. it serves a dataset of 15 email scenarios categorized by difficulty (1 to 3).
Agent Logic (inference.py): An LLM-powered agent (using gemini-2.0-flash) that analyzes email metadata, headers, and URLs to make triage decisions.
🛠 Project Structure
env.py: The core environment logic. Includes the dataset and scoring metrics.
inference.py: The agent's decision-making loop.
models.py: Pydantic models defining the Observation and Action spaces.
openenv.yaml: Metadata for the OpenEnv benchmark framework.
Dockerfile: Containerization setup for deployment.
requirements.txt: Python dependencies.
🧪 Scoring Logic
The environment uses a sophisticated reward system:
Perfect Classification: 1.0 + (difficulty * 0.1)
Partial Credit: 0.4 (e.g., classifying Phishing as Spam).
Dangerous Failure: -1.5 (e.g., letting Phishing into the INBOX).
False Positive: -0.5 (e.g., blocking legitimate mail).
Reasoning Bonus: +0.05 for providing detailed justifications.
⚙️ Setup & Installation
Prerequisites
Docker (optional)
Python 3.10+
A Google Gemini API Key
Local Execution
Install dependencies:
pip install -r requirements.txt
Set your environment variables:
export GEMINI_API_KEY="your_api_key_here"
Run the environment server:
uvicorn env:app --host 0.0.0.0 --port 8000
In a separate terminal, run the agent:
python inference.py
🛡 Security Scenarios Covered
Clean: Official Manipal or Amazon communications with valid SPF/DKIM.
Spam: Marketing mail from Swiggy or Internshala.
Phishing: Typo-squatted domains (e.g., manipal-edu.in vs manipal.edu) and shortened URLs.
Credential Theft: Fake security alerts from bank/Google look-alikes.
|