Reproduce “Agentic Specification Generator for Move Programs” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yu-Fu Fu, Meng Xu, Taesoo Kim
MSG introduces an agentic framework leveraging lightweight Large Language Models to automatically generate formal specifications for Move smart contracts. The tool achieved 84% verifiability for 357 functions in Aptos libraries, producing 39% more verifiable specification clauses in aggregate compared to expert-written counterparts.
Key insights
- Large Language Models exhibit strong generalizability for generating specifications in emergent, verification-oriented languages like Move, despite limited training data.
- An agentic, modular architecture, aligning with the compositional nature of formal specification languages, substantially improves the quality and verifiability of LLM-generated specifications.
- Direct feedback from formal verifiers, including counterexamples, is essential for an iterative refinement loop that significantly boosts the accuracy and comprehensiveness of the output.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.