loop-agent

Iterative refinement loops orchestrate repeated cycles of improvement by defining a loop plan with max iterations and a success criterion. Configure termination modes—fixed iteration count, criteria-based checks, convergence detection, or hybrid—then run cycles until your target quality is met or max iterations reached, with progress tracking throughout.

Termination modes control when loops stop: fixed iteration runs a set number of cycles regardless of outcome; criteria-based halts when specific success conditions are met; convergence stops when improvements plateau; hybrid combines multiple modes. Choose based on whether your goal is predictable cycles, quality thresholds, or detecting diminishing returns.

Yes, iterative refinement applies to code refinement, testing validation, performance tuning, and documentation analysis. Each iteration applies fix–validate cycles, monitors progress against success criteria, and detects convergence, enabling systematic quality improvement across any workflow requiring repeated cycles and measurable advancement.

Progress tracking monitors iteration counts, success criteria achievement, and improvement metrics across cycles. Convergence checks detect when further iterations yield diminishing returns. Combined with your chosen termination mode—criteria-based thresholds, fixed limits, or convergence signals—you gain visibility into when quality targets are reached.

Define workflow agents to execute within each iteration cycle; the loop plan orchestrates their sequencing per iteration. No predefined dependencies or components are required—you configure agents specific to your refinement task, whether code analysis, test validation, or performance measurement, and the loop manages their execution until termination.

Avoid iterative refinement when tasks require single-pass execution with no feedback loops, outcomes are unpredictable across cycles, or convergence is impossible to define. Iterative refinement excels at progressive improvement but demands measurable success criteria and agent cycles capable of incremental advancement toward quality targets.