GEPA Open-Source Prompt Optimizer Hits 6.3k Stars, but Users Flag Missing Audit Step

cto_junior · x · 2026-09-02

A practitioner thread on GEPA (Genetic-Pareto), the open-source framework (6.3k GitHub stars) for optimizing prompts, code, agent architectures and configs, notes one gap: it iterates well but lacks a GEPA-like audit to catch missing key facts or misplaced instructions in a prompt.

How GEPA works: unlike RL or gradient methods that collapse execution traces into a scalar reward, GEPA has an LLM read full traces — errors, profiling data, reasoning logs — to diagnose failures, then applies reflective optimization with Pareto-efficient evolutionary search. Ships with Claude skills and examples, fully open source.

Original post →

More from coding & agent

coding & agent channel →