openenv.yaml

spec_version: 1
name: shopOps
description: >
  Tool-driven customer-operations environment for e-commerce support teams.
  Agents must inspect policies and order facts, manage scarce replacement
  inventory, juggle SLA pressure across a queue, and avoid fraud or premature
  closure using the standard OpenEnv step/reset/state interface.
type: space
runtime: fastapi
app: server.app:app
port: 8000

tasks:
  - name: refund_policy_recovery
    description: >
      Single-case recovery task. The agent must inspect order and policy data,
      choose a compliant partial refund, add the required note, and close the
      case cleanly.
    difficulty: easy
    grader: graders.RefundPolicyRecoveryGrader
    reset_parameters:
      task: refund_policy_recovery

  - name: sla_queue_juggle
    description: >
      Multi-case operations queue. The agent must switch between concurrent
      cases, prioritize SLA-critical work, inspect inventory and histories, and
      close five cases without wasting time or budget.
    difficulty: medium
    grader: graders.SlaQueueJuggleGrader
    reset_parameters:
      task: sla_queue_juggle

  - name: fraud_stockout_cascade
    description: >
      High-coupling scenario with scarce inventory, suspicious refund behavior,
      delayed evidence, and cascading consequences from bad early decisions.
      Tests frontier-model planning quality.
    difficulty: hard
    grader: graders.FraudStockoutCascadeGrader
    reset_parameters:
      task: fraud_stockout_cascade