Ambig-IaC: Multi-level Disambiguation for Interactive Cloud Infrastructure-as-Code Synthesis
Paper • 2604.02382 • Published • 1
id int32 0 299 | prompt stringlengths 93 303 | prompt_original stringlengths 21 1.39k | intent stringlengths 25 2.17k | main_tf stringlengths 109 8.28k | checks_rego stringlengths 171 6.45k | plan_json stringlengths 1.41k 174k |
|---|---|---|---|---|---|---|
0 | I need a way to track which lookups are being made against our domain so we can review that activity in a central place, and it should be dependable enough to use for troubleshooting later. | Configure a query log that can create a log stream and put log events using Route 53 resources. Name the zone "primary", the cloudwatch log group "aws_route53_example_com", and the cloudwatch log resource policy "route53-query-logging-policy" | Has one "aws_route53_zone" resource
with "name"
Has one "aws_cloudwatch_log_group" resource
Has one "aws_cloudwatch_log_resource_policy" resource
with "policy_document" enabling "logs:CreateLogStream" and "logs:PutLogEvents"
with "policy_name"
Has one "aws_route53_query_log" resource
with "cloudwatch... | terraform {
required_providers {
aws = {
source = "hashicorp/aws"
version = ">= 5.75"
}
}
required_version = ">= 1.9.0"
}
provider "aws" {
region = "us-east-1"
}
resource "aws_route53_zone" "primary" {
name = "example53.com"
}
resource "aws_cloudwatch_log_group" "aws_route53_example... | package terraform.validation
default is_configuration_valid = false
default is_valid_r53_zone = false
default is_valid_cloudwatch_log_group = false
default is_valid_cloudwatch_log_resource_policy = false
default is_valid_route53_query_log = false
# Validate aws_route53_zone resource
is_valid_r53_zone {
some i... | {"format_version":"1.2","terraform_version":"1.14.3","planned_values":{"root_module":{"resources":[{"address":"aws_cloudwatch_log_group.aws_route53_example_com","mode":"managed","type":"aws_cloudwatch_log_group","name":"aws_route53_example_com","provider_name":"registry.terraform.io/hashicorp/aws","schema_version":0,"v... |
1 | I need to make sure one of our internal name‑lookup setups can be used from more than one private network, and it should work even as we add or remove networks over time. | Configure a valid Route 53 zone association resource | Has multiple "aws_vpc" resource
with a specified "cidr_block"
with "enable_dns_hostnames" equal to "true"
with "enable_dns_support" equal to "true"
Has one "aws_route53_zone" resource
with "name"
with a "vpc" block that contains "vpc_id" that references the "aws_vpc" resource
Has one "aws_route53_... | provider "aws" {
region = "us-east-1"
}
resource "aws_vpc" "example" {
cidr_block = "10.6.0.0/16"
enable_dns_hostnames = true
enable_dns_support = true
}
resource "aws_route53_zone" "example" {
name = "example.com"
vpc {
vpc_id = aws_vpc.example.id
}
}
resource "aws_vpc" "alternate" ... | package terraform.validation
default is_configuration_valid = false
default is_valid_r53_zone = false
default is_valid_vpc = false
default is_valid_zone_association = false
default is_valid_vpc_association = false
# Validate aws_route53_zone resource
is_valid_r53_zone {
some i
resource := input.configurat... | {"format_version":"1.2","terraform_version":"1.14.3","planned_values":{"root_module":{"resources":[{"address":"aws_route53_vpc_association_authorization.example","mode":"managed","type":"aws_route53_vpc_association_authorization","name":"example","provider_name":"registry.terraform.io/hashicorp/aws","schema_version":0,... |
2 | I need our site to show up at our main domain and automatically direct visitors to whatever is handling the traffic behind it, and it should stay reliable even if demand spikes. | Configure a Route 53 record with an Elastic Load Balancer resource. Call the zone "primary" and the elb "main" | Has one "aws_route53_zone" resource
with "name"
Has a "aws_elb" resource
with "name"
with "availability_zones"
with a "listener" block with "instance_port", "instance_protocol", "lb_port", "lb_protocol"
Has a "aws_route53_record" resource
with "name"
with "type"
with "ttl"
with "zone_i... | provider "aws" {
region = "us-east-1"
}
resource "aws_route53_zone" "primary" {
name = "example53.com"
}
resource "aws_elb" "main" {
name = "sample"
availability_zones = ["us-east-1c"]
listener {
instance_port = 80
instance_protocol = "http"
lb_port = 80
lb_proto... | package terraform.validation
default is_configuration_valid = false
default is_valid_r53_zone = false
default is_valid_r53_record = false
default is_valid_elb = false
# Validate aws_route53_zone resource
is_valid_r53_zone {
some i
resource := input.configuration.root_module.resources[i]
resou... | {"format_version":"1.2","terraform_version":"1.14.3","planned_values":{"root_module":{"resources":[{"address":"aws_elb.main","mode":"managed","type":"aws_elb","name":"main","provider_name":"registry.terraform.io/hashicorp/aws","schema_version":0,"values":{"access_logs":[],"availability_zones":["us-east-1c"],"connection... |
3 | Our main data store is too slow for people overseas, so I want to spread read traffic across a few copies in different parts of the world while keeping things reliable. | Using route 53 resources, configure a weighted routing policy that splits users between three db_instances that are replicas of a main db_instance. Provision the three replica instances in "us-east-1", "eu-central-1", and "ap-southeast-1". Provision the zone and main db_instance in "us-west-1". Call the zone "main", th... | Has one "aws_route53_zone" resource
with "name"
Has a "aws_db_instance" resource
with "identifier" set to "master"
with "instance_class"
with "allocated_storage"
with "engine"
with "username"
with "password"
with "backup_retention_period" set to a non-zero number
Has multiple "aws_db_i... | terraform {
required_providers {
aws = {
source = "hashicorp/aws"
version = ">= 5.75"
configuration_aliases = [ aws.main, aws.us_east, aws.eu_central, aws.ap_southeast ]
}
}
required_version = ">= 1.9.0"
}
provider "aws" {
alias = "main"
region = "us-west-2"
}
provider "aws" {
... | package terraform.validation
default is_configuration_valid = false
default is_valid_r53_zone = false
default is_valid_r53_record_us = false
default is_valid_r53_record_eu = false
default is_valid_r53_record_ap = false
default is_valid_db_instance_main = false
default is_valid_db_instance_replicaus = false
defa... | {"format_version":"1.2","terraform_version":"1.14.3","planned_values":{"root_module":{"resources":[{"address":"aws_db_instance.primary","mode":"managed","type":"aws_db_instance","name":"primary","provider_name":"registry.terraform.io/hashicorp/aws","schema_version":2,"values":{"allocated_storage":20,"allow_major_versio... |
4 | I need a setup where people inside our company automatically connect to a different data store than outsiders, and everyone should be able to use simple, human‑friendly names to reach the right place without extra steps. | Configure Route 53 to return different database endpoints to internal versus external users. Internal users are routed to an internal RDS instance, while external users are routed to a publicly accessible one. Name the zones "private" and "public", the databases "internal" and "public", and the subnet group "main" | Has one "aws_vpc" resource
with a specified "cidr_block"
Has multiple "aws_subnet" resources
with "vpc_id" referencing the "aws_vpc" resource
with specified "cidr_block" for each
with "availability_zone" set
Has one "aws_db_subnet_group" resource
with "subnet_ids" referencing the IDs of "aws_subne... | terraform {
required_providers {
aws = {
source = "hashicorp/aws"
version = ">= 5.75"
}
}
required_version = ">= 1.9.0"
}
provider "aws" {
region = "us-east-1"
}
# Create a VPC for internal resources
resource "aws_vpc" "main" {
cidr_block = "10.0.0.0/24"
}
# RDS instances
resource ... | package terraform.validation
default is_configuration_valid = false
default is_valid_r53_zone = false
default is_valid_r53_record = false
default is_valid_db_instance = false
default is_valid_vpc = false
default is_valid_subnet = false
default is_valid_subnet_group = false
is_valid_r53_zone {
some i, j... | {"format_version":"1.2","terraform_version":"1.14.3","planned_values":{"root_module":{"resources":[{"address":"aws_db_instance.internal","mode":"managed","type":"aws_db_instance","name":"internal","provider_name":"registry.terraform.io/hashicorp/aws","schema_version":2,"values":{"allocated_storage":20,"allow_major_vers... |
5 | I need to spread read traffic across a couple of copies of our main database so things stay fast and reliable, and I want to be able to nudge more users toward one copy than the other. | Using route 53 resources, configure a weighted routing policy that splits users between two db_instances that are replicas of a main db_instance. Call the zone "main", the original db_instance "primary", and the two replicas "replica-1" and "replica-2" respectively | Has one "aws_route53_zone" resource
with "name"
Has a "aws_db_instance" resource
with "identifier" set to "primary"
with "instance_class"
with "allocated_storage"
with "engine"
with "username"
with "password"
with "backup_retention_period" set to a non-zero number
Has multiple "aws_db_... | terraform {
required_providers {
aws = {
source = "hashicorp/aws"
version = ">= 5.75"
}
}
required_version = ">= 1.9.0"
}
provider "aws" {
region = "us-east-1"
}
resource "aws_db_instance" "primary" {
identifier = "primary"
allocated_storage = 20
engine = "postgr... | package terraform.validation
default is_configuration_valid = false
default is_valid_r53_zone = false
default is_valid_r53_record_primary = false
default is_valid_r53_record_secondary = false
default is_valid_db_instance_main = false
default is_valid_db_instance_replica1 = false
default is_valid_db_instance_repl... | {"format_version":"1.2","terraform_version":"1.14.3","planned_values":{"root_module":{"resources":[{"address":"aws_db_instance.primary","mode":"managed","type":"aws_db_instance","name":"primary","provider_name":"registry.terraform.io/hashicorp/aws","schema_version":2,"values":{"allocated_storage":20,"allow_major_versio... |
6 | I need a simple way to run and manage a web app in the cloud without handling servers myself, and it should have the right permissions to operate smoothly while keeping the setup pretty minimal. | Create a basic elastic beanstalk environment with a elastic beanstalk application. Name the iam_role "eb_ec2_role", the instance profile "eb_ec2_profile", and the elastic beanstalk application "my_application" | Has one "aws_iam_role" resource
with "name"
with "assume_role_policy" and "Service" set to "ec2.amazonaws.com"
Has one "aws_iam_role_policy_attachment" resource
with "role" referencing the "aws_iam_role"
with "policy_arn" set to "arn:aws:iam::aws:policy/AWSElasticBeanstalkWebTier"
Has one "aws_iam_ins... | terraform {
required_providers {
aws = {
source = "hashicorp/aws"
version = ">= 5.75"
}
}
required_version = ">= 1.9.0"
}
provider "aws" {
region = "us-east-1"
}
# IAM role for Elastic Beanstalk EC2 instances
resource "aws_iam_role" "eb_ec2_role" {
name = "elastic_beanstalk_ec2_role_... | package terraform.validation
default is_configuration_valid = false
default is_valid_iam_instance_profile = false
default is_valid_iam_role = false
default is_valid_iam_role_policy_attachment = false
default is_valid_eb_app = false
default is_valid_eb_env = false
is_valid_iam_role {
some i
resource :... | {"format_version":"1.2","terraform_version":"1.14.3","planned_values":{"root_module":{"resources":[{"address":"aws_elastic_beanstalk_application.my_application","mode":"managed","type":"aws_elastic_beanstalk_application","name":"my_application","provider_name":"registry.terraform.io/hashicorp/aws","schema_version":0,"v... |
7 | "I want to be able to shift my app’s traffic gradually between two different versions so I can rol(...TRUNCATED) | "Configure Blue/Green deployment with Route53 resources to navigate traffic between Elastic Beanstal(...TRUNCATED) | "Has one \"aws_iam_role\" resource\n with \"name\"\n with \"assume_role_policy\" and \"Service(...TRUNCATED) | "terraform {\n required_providers {\n aws = {\n source = \"hashicorp/aws\"\n version (...TRUNCATED) | "package terraform.validation\n\ndefault is_configuration_valid = false\n\ndefault is_valid_iam_inst(...TRUNCATED) | "{\"format_version\":\"1.2\",\"terraform_version\":\"1.14.3\",\"planned_values\":{\"root_module\":{\(...TRUNCATED) |
8 | "I need our main site to show up when people type in our custom name, and it should lead them to the(...TRUNCATED) | Set up a record that maps a domain name to an IPv4 address using Route 53 resources | "Has one \"aws_route53_zone\" resource\n with \"name\" set to \"example53.com\"\n\nHas one \"aws_(...TRUNCATED) | "provider \"aws\" {\n region = \"us-east-1\"\n}\n\nresource \"aws_route53_zone\" \"example53\" {\n (...TRUNCATED) | "package terraform.validation\n\ndefault is_configuration_valid = false\n\ndefault is_valid_r53_zone(...TRUNCATED) | "{\"format_version\":\"1.2\",\"terraform_version\":\"1.14.3\",\"planned_values\":{\"root_module\":{\(...TRUNCATED) |
9 | "I want our main domain to be reachable using modern addressing so people on newer networks can acce(...TRUNCATED) | Set up a record that maps a domain name to an IPv6 address using Route 53 resources | "Has one \"aws_route53_zone\" resource\n with \"name\" set to \"example53.com\"\n\nHas one \"aws_(...TRUNCATED) | "provider \"aws\" {\n region = \"us-east-1\"\n}\n\nresource \"aws_route53_zone\" \"example53\" {\n (...TRUNCATED) | "package terraform.validation\n\ndefault is_configuration_valid = false\n\ndefault is_valid_r53_zone(...TRUNCATED) | "{\"format_version\":\"1.2\",\"terraform_version\":\"1.14.3\",\"planned_values\":{\"root_module\":{\(...TRUNCATED) |
End of preview. Expand in Data Studio
Ambig-IaC: Ambiguous Infrastructure-as-Code Benchmark
A benchmark dataset of 300 tasks for testing AI agents that generate Infrastructure-as-Code (Terraform) configurations from ambiguous natural language intents.
Project page: https://zyang37.github.io/ambig-iac.github.io/
Dataset Description
This dataset is sourced from IaC-Eval. We performed manual fixes to the original Terraform configurations and validated that all 300 tasks pass terraform plan. Each task also includes the corresponding plan output in JSON format (plan_json field), which is easier to parse and compare programmatically than raw HCL.
Given an ambiguous infrastructure request (e.g., "I need a way to track which lookups are being made against our domain"), the agent is expected to generate a valid Terraform configuration. Each task includes the fully specified intent, a reference Terraform implementation, and OPA/Rego validation policies for evaluating the agent's output.
This dataset is useful for studying:
- How well AI agents generate correct IaC configurations from underspecified requirements
- Disambiguation strategies in IaC generation
- Iterative refinement of infrastructure configurations
Fields
| Field | Type | Description |
|---|---|---|
id |
int | Task index (0-299) |
prompt |
string | Ambiguous natural language requirement (~60 words avg) |
prompt_original |
string | Original detailed requirement with specific resource names |
intent |
string | Structured specification listing exact resources and attributes |
main_tf |
string | Reference Terraform (HCL) configuration |
checks_rego |
string | OPA/Conftest Rego policy for validation |
plan_json |
string | Terraform plan JSON output |
Usage
from datasets import load_dataset
ds = load_dataset("znyang/ambig-iac")
# Access a task
task = ds["train"][0]
print(task["prompt"]) # Ambiguous requirement
print(task["intent"]) # Detailed specification
print(task["main_tf"]) # Reference Terraform code
Dataset Statistics
- Tasks: 300
- Avg prompt length: ~60 words
- Avg intent length: ~12 lines
- Avg main.tf length: ~98 lines
- Avg checks.rego length: ~69 lines
- Domain: AWS infrastructure (Route 53, CloudWatch, VPC, Kinesis, IAM, etc.)
Citation
If you find this work useful, please cite:
@misc{yang2026ambigiacmultileveldisambiguationinteractive,
title={Ambig-IaC: Multi-level Disambiguation for Interactive Cloud Infrastructure-as-Code Synthesis},
author={Zhenning Yang and Kaden Gruizenga and Tongyuan Miao and Patrick Tser Jern Kon and Hui Guan and Ang Chen},
year={2026},
eprint={2604.02382},
archivePrefix={arXiv},
primaryClass={cs.SE},
url={https://arxiv.org/abs/2604.02382},
}
License
MIT
- Downloads last month
- 109