dk-close-the-loop

Automate iterative refinement by enforcing a structured observe, hypothesize, change, and rerun loop. This skill systematically processes prompts, plans, or model results while preserving all intermediate states to improve outcomes.

A structured loop for debugging AI pipelines applies reproducible phases including Observe, Hypothesize, Change, Run, Compare, and Verdict. This process systematically critiques artifacts and stores state on disk to prevent context bleed.

Keep context organized during prompt refinement by storing state in a dedicated scratch directory. This workflow preserves all intermediate artifacts on disk, preventing context bleed and maintaining an auditable history of changes.

Yes, you can use a scratch-directory workflow to compare model results. The process evaluates changes by running new iterations against baselines, applying a structured verdict phase to determine if the refinement improved the output.

Systematically critique and improve AI-generated plans by applying a phase-driven iteration process. You observe the baseline, hypothesize improvements, implement changes, and rerun to compare results, ensuring reproducible refinement.

Iterative AI refinement requires a scratch-directory workflow to manage loop state rather than a specific framework. The process stores intermediate phases and artifacts on disk to ensure reproducibility and prevent context bleed.