Published November 10, 2026 | Version v1.0.0
Software Open

Replication Package for "How Far Can LLMs Go in Generating Android Lint Checks from Natural Language?"

Authors/Creators

  • 1. Louisiana State University
  • 2. University of Kentucky
  • 3. University of California, Riverside

Description

Artifact for the paper "How Far Can LLMs Go in Generating Android Lint Checks from Natural Language?". LintBench is a 148-instance benchmark (113-instance human-verified subset) for evaluating LLMs on generating Android Lint detectors from natural-language issue specifications, with a Dockerised Lint test harness. It uses a retrieval-augmented, generation and repair pipeline evaluated on the same instances.

Notes

If you use LintBench, please cite it as below.

Files

Fakorede/lintbench-v1.0.0.zip

Files (203.8 MB)

Name Size
md5:59738869a5f50a810aaa2b13351c9cb7
203.8 MB Preview Download

Additional details

Related works

Is supplement to
Software: https://github.com/Fakorede/lintbench/tree/v1.0.0 (URL)

Software

Repository URL
https://github.com/Fakorede/lintbench