FAPO: Fully Autonomous Prompt Optimization of Multi-Step LLM Pipelines

FAPO uses Claude Code as an autonomous optimizer for multi-step LLM pipelines, iterating through prompt edits and structural chain changes based on failure attribution across intermediate steps. Tested across six benchmarks and three task models, it outperforms the GEPA baseline in 15 of 18 comparisons by targeting bottlenecks at the retrieval, reasoning, or formatting layer rather than optimizing prompts globally.