Search papers, labs, and topics across Lattice.
The paper introduces CLI-Tool-Bench, a new benchmark for evaluating LLMs' ability to generate complete CLI tools from scratch without scaffolds. This benchmark uses black-box differential testing against human-written oracles to assess end-to-end behavior and system side effects. Experiments on seven LLMs show that even the best models achieve less than 43% success, indicating significant room for improvement in 0-to-1 software generation.
LLMs struggle to build complete software from scratch, with even the best models failing more than half the time on a new CLI tool generation benchmark.
Large Language Models (LLMs) are driving a shift towards intent-driven development, where agents build complete software from scratch. However, existing benchmarks fail to assess this 0-to-1 generation capability due to two limitations: reliance on predefined scaffolds that ignore repository structure planning, and rigid white-box unit testing that lacks end-to-end behavioral validation. To bridge this gap, we introduce CLI-Tool-Bench, a structure-agnostic benchmark for evaluating the ground-up generation of Command-Line Interface (CLI) tools. It features 100 diverse real-world repositories evaluated via a black-box differential testing framework. Agent-generated software is executed in sandboxes, comparing system side effects and terminal outputs against human-written oracles using multi-tiered equivalence metrics. Evaluating seven state-of-the-art LLMs, we reveal that top models achieve under 43% success, highlighting the ongoing challenge of 0-to-1 generation. Furthermore, higher token consumption does not guarantee better performance, and agents tend to generate monolithic code.