Analyse recorded sessions for recurring failure patterns and turn them into correction rules that can be applied back, with a status view (scan, apply, status). Use when the same mistake keeps coming back and you want a learned, repeatable correction rather than another one-off fix. Not for distilling a whole loop into reusable knowledge — use agent-intel (loop-distill); not for detecting structural code smells — use the code_* detectors; not for governed long-term memory — use memory-hub.
