关于
The learn-from-session skill provides an automated retrospective layer for Claude Code workflows by parsing interaction logs to identify points of friction. By examining session JSONL files, it detects patterns such as repeated iterations, manual user corrections, and skipped planning phases. The skill applies a rigorous '3/3 counterfactual test' to ensure all recommendations are high-signal, meaning the suggested changes are guaranteed to have prevented specific rework if they had been active during the original session. This creates a continuous improvement loop for developers looking to refine their custom agentic workflows.
