Ghaythfd/gherkin-scenarios

Gherkin BDD Test Scenarios Dataset

A dataset of Gherkin/Cucumber BDD test scenarios for fine-tuning language models to generate automated test cases.

Dataset Description

This dataset contains instruction-response pairs for training models to generate Gherkin test scenarios. The scenarios cover data management operations in enterprise software systems.

Domains Covered

• Data Management: CSV/JSON/Excel import and export
• Navigation: UI screens, search, filters
• API Testing: REST and SOAP web services
• Job Scheduling: Task execution and reporting
• IBOR: Investment Book of Record operations

Dataset Structure

Format

ShareGPT conversational format:

{
  "conversations": [
    {"from": "human", "value": "Write a Gherkin scenario for testing CSV file import"},
    {"from": "gpt", "value": "Scenario Outline: Import CSV...\n    Given I am logged..."}
  ]
}

Statistics

• Total Examples: 726
• Training Split: 653 (90%)
• Validation Split: 73 (10%)
• Source Files: 88 Gherkin feature files

Example

{
  "conversations": [
    {
      "from": "human",
      "value": "Create a BDD test scenario for advanced search using text filter"
    },
    {
      "from": "gpt", 
      "value": "Scenario Outline: Advanced Filter using one text field\n    Given I am logged as TAV_standard.user on <screen_name>\n    And I open advanced search tool\n    When I set an advanced text filter <field_name> = <field_value>\n    And I click on Search button\n    Then The query grid shows all rows having <field_name> equal to <field_value>\n\n    Examples:\n      | screen_name   | field_name | field_value |\n      | My Securities | Name       | Test_Value  |"
    }
  ]
}

Usage

Load with Datasets Library

from datasets import load_dataset

dataset = load_dataset("Ghaythfd/gherkin-scenarios")

For Fine-tuning

from datasets import load_dataset

dataset = load_dataset("Ghaythfd/gherkin-scenarios", split="train")

# Format for training
def format_prompt(example):
    conversation = example["conversations"]
    text = f"### Instruction:\n{conversation[0]['value']}\n\n### Response:\n{conversation[1]['value']}"
    return {"text": text}

dataset = dataset.map(format_prompt)

Creation Process

1. Collected 88 Gherkin .feature files from test automation suites
2. Parsed scenarios using regex extraction
3. Generated instruction variations for each scenario
4. Created completion examples from partial scenarios
5. Formatted into ShareGPT conversational structure

Intended Use

• Fine-tuning LLMs for Gherkin scenario generation
• Training test automation assistants
• BDD/TDD tooling development

License

Apache 2.0