File size: 1,890 Bytes
08c0cf7
 
26f67bb
08c0cf7
26f67bb
 
 
 
08c0cf7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
26f67bb
08c0cf7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
name: nexus-incident-investigation
version: "1.0.0"
tags: ["openenv", "multi-agent"]
description: >
  NEXUS  Multi-Agent Incident Investigation Environment.
  Multiple AI agents (up to 10) collaborate to investigate real-world system incidents.
  Agents can take different roles: Investigator, Validator, Forensic Analyst,
  Network Engineer, System Admin, Security Architect, and Compliance Officer.
  Together they identify root causes across software, business-process,
  and cascade-system failure scenarios.

tasks:
  - name: software-incident
    description: Single-service software bug causing user-facing errors
    difficulty: easy
    max_steps: 8
    grader: scenarios/graders/easy_grader.py

  - name: business-process-failure
    description: Multi-team process breakdown with misleading red-herrings
    difficulty: medium
    max_steps: 8
    grader: scenarios/graders/medium_grader.py

  - name: cascade-system-failure
    description: Multi-system cascade failure with misleading logs
    difficulty: hard
    max_steps: 8
    grader: scenarios/graders/hard_grader.py

action_space:
  type: text
  description: "Free-form natural language message with optional TOOL: calls"

observation_space:
  type: structured
  fields:
    scenario_description: string
    scenario_context: string
    partner_message: string
    tool_results: list
    clues_found: list
    investigation_stage: string
    round: integer
    available_tools: list

reward_range: [0.0, 1.0]
reward_description: >
  Dynamically computed from semantic similarity of hypothesis to root-cause,
  tool quality, fix correctness, and investigation efficiency.

inference_script: inference.py
entry_point: backend/main.py
docker_port: 7860

baseline_scores:
  software-incident: 0.88
  business-process-failure: 0.72
  cascade-system-failure: 0.48