@inproceedings{f19e7f0f676b43e9882464202a62bbb2,
title = "Framework-Adaptive Agentic Pipelines for API Knowledge Extraction and Test Generation",
abstract = "Automated API documentation and test generation for Python web applications remain challenging due to the fragmentation of the framework ecosystem and the limitations of existing single-framework tools. We present a framework-adaptive multi-agent pipeline supporting seven Python web frameworks — Django, FastAPI, Flask, Bottle, Tornado, Pyramid, and Sanic — through dedicated static parsers and framework-specific prompt templates. A rule-based parsing layer extracts per-endpoint context without invoking any language model. A Generation Agent then produces structured parameter documentation, which a Validation Agent independently verifies through up to three revision rounds; endpoints that fail verification are flagged for human review rather than passed forward silently. A Consistency Agent subsequently identifies cross-endpoint inconsistencies across the complete document set. A Test Generation Agent further produces test intents and pytest skeletons for each confirmed endpoint. Evaluation across all seven frameworks demonstrates robust detection and extraction performance, and analysis of validation false positives reveals four distinct failure categories with targeted remediation strategies identified for future refinement.",
keywords = "Codes, Large language models, Application programming interfaces, Modeling, Documentation, Tornadoes, Printing, Testing, Pipelines, Conferences, API Documentation, Multi-Agent Systems, Python Web Frameworks, Test Generation",
author = "Peng Zhao and John Blake and Evgeny Pyshkin",
note = "This is an accepted manuscript of an article published in: P. Zhao, J. Blake and E. Pyshkin, {"}Framework-Adaptive Agentic Pipelines for API Knowledge Extraction and Test Generation,{"} 2026 IEEE International Conference on Industry 4.0, Artificial Intelligence, and Communications Technology (IAICT), Bali, Indonesia, 2026, pp. 185-192, doi: 10.1109/IAICT71158.2026.11620949. For the purposes of open access the author/s has/ve applied a Creative Commons Attribution (CC BY) licence to any Author Accepted Manuscript (AAM) version arising from this submission. ",
year = "2026",
month = jul,
day = "28",
doi = "10.1109/IAICT71158.2026.11620949",
language = "English",
series = "Proceedings - International Conference on Industry 4.0, Artificial Intelligence, and Communications Technology (IAICT)",
publisher = "IEEE",
pages = "185--192",
booktitle = "2026 IEEE International Conference on Industry 4.0, Artificial Intelligence, and Communications Technology (IAICT)",
address = "United States",
}