bz-lab/AUITestAgent

Decouples GUI interaction and function verification into separate LLM-driven modules with dynamically organized agents for parsing diverse natural language requirements, then applies multi-dimensional data extraction to verify UI responses against expected outcomes. Evaluated on 8 commercial Android/iOS apps (Meituan, Facebook, Gmail, etc.) with tiered difficulty benchmarks, achieving 100% success on simple tasks, 80% on moderate complexity, and 90% bug detection recall with 4.5% false positive rate. Integrates with mobile app testing pipelines and demonstrates practical bug-finding capability in production applications.

289 stars. No commits in the last 6 months.