openenv.yaml

name: trace
version: "0.1.0"
description: "OpenEnv incident response environment"
author: "Rajarshi Datta"

# Environment interface
interface:
  entry_point: "server.app:app"
  port: 7860

# Tasks/benchmarks — each must have a grader for validation
tasks:
  - id: easy_cpu_spike
    name: "CPU Spike Recovery"
    description: "Agent must scale workers to recover from traffic spike"
    max_steps: 5
    difficulty: easy
    grader: "trace.graders:grade_easy_cpu_spike"

  - id: medium_cascade
    name: "Cascading Queue Failure"
    description: "Agent must diagnose and restart queue service to prevent cascade"
    max_steps: 7
    difficulty: medium
    grader: "trace.graders:grade_medium_cascade"

  - id: hard_mixed
    name: "Multi-service incident"
    description: "DB + release regression; agent must diagnose and remediate"
    max_steps: 8
    difficulty: hard
    grader: "trace.graders:grade_hard_mixed"

# Reward specification
reward:
  type: float
  min: 0.0
  max: 1.0
  description: "Cumulative normalized reward [0, 1]"

# Action schema
action:
  type: object
  properties:
    action_type:
      type: string
      enum:
        - inspect_logs
        - inspect_metrics
        - inspect_alert
        - restart_service
        - scale_workers
        - restart_database
        - rollback_release
        - clear_queue
        - declare_healthy
        - declare_unfixable
    target:
      type: string
      nullable: true
    value:
      type: number
      nullable: true

# Observation schema
observation:
  type: object
  properties:
    timestamp:
      type: string
    cpu_usage_pct:
      type: number
      minimum: 0
      maximum: 100
    memory_usage_pct:
      type: number
      minimum: 0
      maximum: 100
    error_rate_pct:
      type: number
      minimum: 0
      maximum: 100
    api_latency_ms:
      type: number
      minimum: 0
    queue_depth:
      type: integer
      minimum: 0
    services:
      type: object
      additionalProperties:
        type: string
    active_alerts:
      type: array
      items:
        type: string
    last_inspection:
      type: object
      nullable: true

# Metadata
metadata:
  environment_type: "RL"
  problem_statement: "Incident response in production infrastructure"
  benchmark_suite: "OpenEnv"
  validation_level: "full"