iterative-refinement

Iterative refinement automates test-fix cycles by executing repeated action-validate-improve loops with convergence detection. Define success criteria and max iterations in configuration, then the system runs cycles that execute fixes, validate outcomes, record progress, and terminate when criteria are met or convergence is detected.

Iterative refinement supports four termination modes: fixed iteration count, criteria-based (stop when conditions met), convergence-based (stop when improvements plateau), and hybrid combinations. Choose the mode that fits your workflow—fixed for predictable tasks, criteria-based for goal-driven work, convergence for performance optimization.

Yes, iterative refinement applies to test-fix cycles, code quality improvement, performance optimization, and progressive enhancement. It preserves context between iterations and supports language-specific validators, making it suitable for repeated action-validate-improve workflows across software projects.

Iterative refinement uses a modular, multi-file structure with progressive disclosure of tooling to preserve context between iterations while minimizing token consumption. This design allows complex workflows to run efficiently by organizing related actions and validators across multiple files.

Use iterative refinement when you need automated, repeatable validation loops with intelligent termination and progress tracking. It's ideal for scenarios where manual cycles would be tedious—test-fix patterns, performance tuning, and quality improvements that require consistent validation at each step.