Super Library Agent: Joint Generation and Maintenance of Multiple Applications Beyond the Single Codebase
The paper argues coding agents need a shared-library workflow when building related app portfolios.
The authors define “Super Library Agent” as an agent that sequentially creates multiple related applications while maintaining reusable cross-application components. They say naive per-app generation duplicates logic, while naive library extraction misses shared code and makes migrations fragile. Their method adds candidate-guided extraction, pre-extraction consolidation, and migration informed by traces and call graphs. On WebGen-Bench and PaperBench, they report preserved functionality with less redundancy, shorter token footprint, and reduced structural erosion versus baselines. HF Daily Papers' note
The authors define “Super Library Agent” as an agent that sequentially creates multiple related applications while maintaining reusable cross-application components. They say naive per-app generation duplicates logic, while naive library extraction misses shared code and makes migrations fragile. Their method adds candidate-guided extraction, pre-extraction consolidation, and migration informed by traces and call graphs. On WebGen-Bench and PaperBench, they report preserved functionality with less redundancy, shorter token footprint, and reduced structural erosion versus baselines. HF Daily Papers' note
score 4