dataset-task-manager

Feature-evolution dataset maintained as part of isselab/agentic-feature-traced-datasets.

Dataset

This dataset captures controlled feature evolution for a desktop task management application. The seed project in project/ is a Maven-based JavaFX application that starts from a compact task manager and includes feature-tracing support through .feature-model, .feature-to-file, and embedded feature annotations.

The benchmark defines 14 sequential evolution steps from v000 to v014. The scenarios cover additions, modifications, removals, renames, relocations, merges, splits, and feature interactions. Each step under benchmark/steps/step-*/ contains the user prompts that drove the change, expected feature-level changes, and manually curated ground-truth traceability artifacts. Ground truth is represented as:

• .feature-model for the expected hierarchical feature model after the step.
• .feature-to-file for file-level feature ownership mappings.
• .feature-to-folder for folder-level mappings when applicable.

The subject project also uses inline annotations in source comments with &begin[FeatureName] / &end[FeatureName] blocks or &line[FeatureName] markers for smaller feature fragments. Feature names are PascalCase and refer to local feature names from the feature model.

Limitations: the dataset focuses on a small JavaFX application and its traceability artifacts, not on broad production task-management behavior.

Contents

• project/ is the initial project supplied to Agent-HAnS.
• benchmark/steps.json orders the feature-evolution steps.
• Each step contains exact prompts, expected changes, and expected traceability artifacts.
• docs/ contains dataset-specific documentation.
• schema/ defines the machine-readable formats.

Validate

python scripts/validate_dataset.py

The validator checks dataset structure. it does not evaluate generated output.

License

See LICENSE and the license field in dataset.json.