bhavyaku07/dropshipping-operations-env

📦 Dropshipping Operations Manager (OpenEnv)

Overview

The Dropshipping Operations Manager is a text-based, multi-step simulation environment built for the OpenEnv framework. In this simulation, an LLM agent acts as the operations lead for an e-commerce storefront. The agent must parse incoming data from multiple streams (suppliers, customers, and competitors) and execute precise business actions to maintain profitability and customer satisfaction.

Observation Space (State)

The environment state is returned as a strict JSON dictionary containing four primary keys:

• —inventory: Current product catalog, stock levels, and retail pricing.
• —supplier_inbox: Automated emails from suppliers regarding stock shortages or shipments.
• —customer_tickets: Support tickets from buyers requesting refunds or complaining about delays.
• —competitor_data: Web-scraped pricing metrics from rival storefronts.

Action Space

The agent must output exactly one of the following five actions as a string per step. Any malformed action or hallucinated parameter will be gracefully caught and converted to a noop().

1. 1.update_inventory("PROD_ID", new_stock_int): Updates the internal stock ledger based on supplier emails.
2. 2.issue_refund("TICK_ID", percentage_int): Processes a financial refund to resolve a customer complaint (e.g., 15 for a 15% refund).
3. 3.update_price("PROD_ID", new_price_float): Adjusts the retail price of a product to maintain a competitive edge.
4. 4.reply_ticket("TICK_ID"): Sends a generic confirmation reply to close a customer support ticket.
5. 5.noop(): Takes no action and advances the environment step.

Evaluation & Graders

The environment evaluates the agent's performance across 5 maximum steps using strict, deterministic Python graders:

• —Task 1 (Easy - Inventory Sync): The agent must read a supplier email stating a product is out of stock and update PROD_001 stock to 0.
• —Task 2 (Medium - Customer Resolution): The agent must read a customer complaint (TICK_882), issue exactly a 15% refund, and reply to the ticket.
• —Task 3 (Hard - Dynamic Pricing): The agent must read competitor data showing a rival priced at $28.00 and undercut them by exactly 5% (updating PROD_003 to $26.60).