openenv-support / openenv.yaml
Mahir
Fix inference.py: sync heuristic fallback, clamp all scores--2
687d4d6
Raw
History Blame Contribute Delete
2.63 kB
name: multi-domain-ai-agent-environment
version: "2.0.0"
description: >
A real-world OpenEnv environment testing AI agents across four professional
domains: Customer Support Triage, Legal Document Review, Clinical Patient
Triage, and Software Engineering PR Review. Each domain contains 3 tasks
progressing from easy classification to hard open-ended generation.
author: OpenEnv Contributor
tags:
- customer-support
- legal
- clinical
- engineering
- nlp
- multi-domain
- real-world
- triage
tasks:
- id: ticket_classification
name: Ticket Classification & Routing
difficulty: easy
score_range: [0.001, 0.999]
domain: customer_support
- id: response_drafting
name: Response Drafting & Quality
difficulty: medium
score_range: [0.001, 0.999]
domain: customer_support
- id: queue_management
name: SLA Queue Management
difficulty: hard
score_range: [0.001, 0.999]
domain: customer_support
- id: multi_turn_conversation
name: Multi-Turn De-escalation
difficulty: very_hard
score_range: [0.001, 0.999]
domain: customer_support
- id: legal_clause_identification
name: Legal Clause Identification
difficulty: easy
score_range: [0.001, 0.999]
domain: legal
- id: legal_risk_flagging
name: Legal Risk Flagging
difficulty: medium
score_range: [0.001, 0.999]
domain: legal
- id: legal_clause_redlining
name: Legal Clause Redlining
difficulty: hard
score_range: [0.001, 0.999]
domain: legal
- id: clinical_triage_classification
name: Clinical Triage Body System
difficulty: easy
score_range: [0.001, 0.999]
domain: clinical
- id: clinical_esi_assignment
name: Clinical ESI Assignment
difficulty: medium
score_range: [0.001, 0.999]
domain: clinical
- id: clinical_triage_note
name: Clinical Triage Note Generation
difficulty: hard
score_range: [0.001, 0.999]
domain: clinical
- id: pr_type_classification
name: PR Type Classification
difficulty: easy
score_range: [0.001, 0.999]
domain: engineering
- id: pr_bug_identification
name: PR Bug Identification
difficulty: medium
score_range: [0.001, 0.999]
domain: engineering
- id: pr_review_comment
name: PR Code Review
difficulty: hard
score_range: [0.001, 0.999]
domain: engineering
endpoints:
reset: POST /reset
step: POST /step
state: GET /state
tasks: GET /tasks
grader: POST /grader
baseline: POST /baseline
observation_schema: app.models.Observation
action_schema: app.models.Action
reward_schema: app.models.Reward