Hybrid Tool Combines Mechanical Analysis and AI for Move Specification Inference
Researchers have introduced a new specification inference tool designed for the Move Prover, aiming to reduce the manual effort required in writing formal specifications for Move bytecode. This innovative approach integrates weakest-precondition (WP) analysis with an agentic coding CLI, such as Claude Code, via a Model Context Protocol (MCP) service. The WP analysis provides a sound, mechanical baseline, while the AI agent addresses areas where WP is less effective, such as generating loop invariants and high-level idiomatic specifications like monotonicity and conservation. The Move Prover acts as an oracle to validate the generated specifications, allowing the agent to refine proof hints until verification succeeds. The tool has been successfully tested on a corpus of canonical Move code, including complex features like higher-order functions, dynamic dispatch, and global state. This development represents a significant step in automating formal verification processes, combining rigorous mechanical analysis with the flexibility of artificial intelligence to enhance software reliability and developer productivity in blockchain and secure programming environments.
Wire timeline
Hybrid Tool Combines Mechanical Analysis and AI for Move Specification Inference
Researchers have introduced a new specification inference tool designed for the Move Prover, aiming to reduce the manual effort required in writing formal specifications for Move bytecode. This innovative approach integrates weakest-precondition (WP) analysis with an agentic coding CLI, such as Claude Code, via a Model Context Protocol (MCP) service. The WP analysis provides a sound, mechanical baseline, while the AI agent addresses areas where WP is less effective, such as generating loop invariants and high-level idiomatic specifications like monotonicity and conservation. The Move Prover acts as an oracle to validate the generated specifications, allowing the agent to refine proof hints until verification succeeds. The tool has been successfully tested on a corpus of canonical Move code, including complex features like higher-order functions, dynamic dispatch, and global state. This development represents a significant step in automating formal verification processes, combining rigorous mechanical analysis with the flexibility of artificial intelligence to enhance software reliability and developer productivity in blockchain and secure programming environments.
cs.AI updates on arXiv.org