best-practices-self-improvement-loop

Deterministic coded self-improvement loops for *-lab skills. The agent will NOT do this reliably — loops MUST be in code. Proven pattern from Classifier Lab: data_enrichment.py + training_loop.py.