Search papers, labs, and topics across Lattice.
This paper introduces CodeSpec, a dual executable specification method designed to enhance the reliability of feature development in LLM-based code agents by ensuring coherent functional chains within existing architectures. By pairing sub-requirement semantics with repository architectures, CodeSpec generates complementary specifications that verify the completeness and correctness of these chains, thereby maintaining design-implementation consistency over long-horizon interactions. The method achieves significant performance improvements on FeatureBench, with pass rates of 70.7%, 55.0%, and 49.9% under DeepSeek-V4-Pro, surpassing existing baselines like Claude Code.
CodeSpec transforms feature development by ensuring that LLM-based code agents produce reliable and verifiable functional chains, achieving up to 70.7% pass rates on complex benchmarks.
LLM-based code agents have advanced repository-level software development through iterative interaction with codebases and tools. However, feature development requires integrating new behaviors into existing architectures through coherent cross-component functional chains. Existing agents typically derive such chains through free-form reasoning, often producing unreliable feature designs with incomplete functional chains. Moreover, textual designs are difficult to verify and enforce, making it challenging to maintain design-implementation consistency throughout long-horizon development. We propose CodeSpec, a dual executable specification method for repository-level feature development. It builds reliable functional chains from evidence pairing sub-requirement semantics with repository architectures, then compiles them into complementary architecture and behavior specifications that check chain completeness and correctness while preserving design-implementation consistency over long interactions. On FeatureBench, which targets feature development in existing repositories, CodeSpec achieves 70.7%, 55.0%, and 49.9% pass rates under DeepSeek-V4-Pro, outperforming representative baselines such as Claude Code. Results on the repository generation benchmark NL2Repo-Bench further demonstrate its generalizability.
