What problem does it solve? Single-shot generation often produces outputs that miss quality bars for code, reports, and analysis. This Skill provides structured patterns for iterative self-evaluation and refinement so agent outputs converge on defined criteria before delivery. ## Core Features & Use Cases - Reflection loops: Generate, critique against criteria, and revise outputs for up to a bounded number of iterations with structured JSON feedback. - Evaluator-optimizer pipelines: Separate generation and scoring components with configurable thresholds (default 0.85) and weighted rubric dimensions like accuracy, clarity, and completeness. - Code-specific refinement: Test-driven loops that generate tests, run them, and revise code until the suite passes. - L2 evidence records: Normalize evaluator and reviewer output into JSON evidence records appended via the governed append script for downstream prompt-engineering refinement. - Use Case: A pipeline stage produces a design document; the evaluator scores it against the brief and governance rules, and the optimizer revises it until the score clears the threshold before the next stage proceeds. ## Quick Start Ask the agent to evaluate the current draft against a defined rubric and iteratively refine it until the overall score reaches 0.85 or the iteration limit is hit.
