Agentic Separation Logic Specification Synthesis

A new agentic system called Spec-Agent can automatically synthesize formal specifications for 85% of target functions across large C++ codebases, with no false positives observed during validation, according to research published on arXiv [1]. The system outperforms Claude Code Opus 4.6 at a token cost roughly 10 times lower [1]. Specification synthesis — the task of inferring formal specifications from program implementations and natural language — is considered important for refactoring, transpilation, optimization, and verification [1]. The challenge has remained open for large C++ repositories because existing approaches that rely on large language models fail to scale, produce specifications expressive enough to capture systems-code features such as dynamic memory and heap-allocated data structures, and systematically validate candidates to rule out incorrect results [1]. Spec-Agent addresses these limitations by targeting a ladder of specification languages: propositional logic, first-order logic, propositional separation logic, and first-order separation logic [1]. For each function, the system uses static analysis and runtime heap tracing to select the appropriate target language, generalizes existing functional tests into fuzz harnesses, and iteratively refines LLM-generated candidates via counterexample-guided feedback [1]. The researchers evaluated Spec-Agent on open-source C++ codebases comprising millions of lines of code [1]. The system synthesized valid specifications for 85% of target functions, and no false positives were observed under fuzzing and expert validation [1]. In comparison, Claude Code Opus 4.6 was outperformed at a token cost roughly 10 times lower [1]. Formal specification languages such as separation logic extend Hoare logic to reason about programs that manipulate pointers and mutable data structures, making them particularly relevant for systems code written in C++ [5]. The actor model, a mathematical model of concurrent computation that treats actors as the basic building block and removes the need for lock-based synchronization, represents another formal framework that has influenced practical implementations of concurrent systems since its origin in 1973 [4]. While Spec-Agent does not target actor-based concurrency directly, its ability to handle heap-allocated structures addresses a related class of verification challenges [1][4].