MORE | Fall 2025

Echidna Agent

This study develops an agentic system that executes Application Programming Interface (API) tests, explains failures, learns constraints, and automatically repairs test scripts. Evaluations on mock and public services show that combining a structured request recorder with Artificial Intelligence (AI) large language models (LLMs) infers hidden rules and generates targeted patches, improving pass rates. The approach reduces maintenance effort requirements, accelerates Continuous Integration/Continuous Delivery (CI/CD), and increases reliability for widely used software. Future work will expand benchmarks, integrate abstract syntax tree (AST)-level repairs, and explore contract synthesis and drift detection.