darwinian-evolver

What problem does it solve?

It automates iterative optimization of an LLM prompt, regex, SQL query, or small code snippet by using an evolutionary search loop driven by a fitness score.

Core Features & Use Cases

• Fitness-driven evolution: Improves a starting “organism” using an evaluator that scores outputs from measurable criteria.
• Failure-case guided mutation: Uses trainable failure cases to prompt the mutator to propose better artifacts.
• Overfitting detection: Supports holdout failure cases to catch solutions that only work on training inputs.
• Common use cases: Optimize prompt instructions for reliability, evolve regex patterns for extraction, tune SQL queries for correctness, or refine small code snippets validated by tests or benchmarks.
• Wrapper around upstream tool: Drives Imbue’s darwinian-evolver via CLI/subprocess so Titan does not import AGPL upstream code into its core.

Quick Start

Ask the skill to optimize a prompt by stating your target behavior, providing a scoring rule (or test cases), and giving a starting template like “Say {{ phrase }}”.