Email_AI_Automation/openenv.yaml at main · Sam103008/Email_AI_Automation · GitHub

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
94
95
96
97
98
99
100
101
102
103
104
105
106
name: email-triage-openenv
version: "1.0.0"
description: >
  An OpenEnv-compliant environment for training AI agents to manage a business
  email inbox. The agent labels emails by priority, drafts replies, routes
  messages to the appropriate department, and escalates critical issues.

tags:
  - openenv
  - email
  - triage
  - real-world
  - nlp

tasks:
  - id: task1
    name: Priority Labeling
    difficulty: easy
    max_steps: 10
    description: >
      Label 5 emails with the correct priority: urgent / high / normal / low / spam.
      Urgent emails (production outage, security alert) carry 2× weight in scoring.
    scoring:
      method: weighted_label_accuracy
      range: [0.0, 1.0]

  - id: task2
    name: Triage and Reply
    difficulty: medium
    max_steps: 25
    description: >
      Label 10 emails by priority and draft replies to emails that require a response.
      Replies must be at least 20 characters. Unnecessary replies incur a 0.05 penalty.
    scoring:
      method: weighted_label_plus_reply_coverage
      formula: "0.6 * label_score + 0.4 * reply_score - 0.05 * unnecessary_replies"
      range: [0.0, 1.0]

  - id: task3
    name: Full Inbox Management
    difficulty: hard
    max_steps: 40
    description: >
      Manage a 15-email inbox: label by priority, route to the correct department
      (devops / legal / engineering / management), and escalate time-critical or
      compliance issues. Critical emails carry 3× label weight.
    scoring:
      method: weighted_multi_action
      formula: "0.40 * label + 0.35 * routing + 0.25 * escalation + reply_bonus"
      range: [0.0, 1.0]

observation_space:
  type: object
  fields:
    task_id: string
    step: integer
    emails:
      type: array
      items:
        id: string
        sender: string
        subject: string
        body: string
        received_at: string (ISO 8601)
        has_attachment: boolean
        thread_id: string (nullable)
    inbox_size: integer
    done: boolean
    message: string
    metadata: object

action_space:
  type: object
  fields:
    action_type:
      type: enum
      values: [label, reply, archive, escalate, route, noop]
    email_id: string (nullable)
    label:
      type: enum
      values: [urgent, high, normal, low, spam]
      nullable: true
    reply_text: string (nullable, min 20 chars for reward)
    route_to: string (nullable)
    reason: string (nullable)

reward:
  type: object
  fields:
    value: float (0.0 – 1.0)
    breakdown: object
    message: string
  intermediate_rewards: true
  terminal_reward: true

api:
  reset: POST /reset
  step:  POST /step
  state: GET  /state
  tasks: GET  /tasks
  grade: POST /grade

runtime:
  framework: fastapi
  python: ">=3.11"
  port: 7860