memory-distill

Distilling episodic memory involves reading failures.jsonl and decisions.jsonl to identify unpromoted lessons, mapping them by component and category to target skills, and appending the promoted entries to update the knowledge base.

Distill failures and decisions periodically after collecting several entries. The process identifies unpromoted items in your jsonl files and maps them by component, category, and ref to target skills or rules for permanent knowledge integration.

The mapping mechanism identifies unpromoted items in memory files, categorizes them by component, category, and ref, and appends distilled entries to corresponding jsonl files. It avoids duplicates by checking and updating the promoted flag for each processed item.

Memory distillation requires no dependencies or external components. It autonomously reads failures.jsonl and decisions.jsonl, identifies unpromoted lessons, maps them to target skills or rules, and updates the promoted flag to reflect permanent knowledge integration.

To avoid duplicate lessons when promoting episodic memory, the distillation process checks the promoted flag in failures.jsonl and decisions.jsonl before appending. It identifies unpromoted items, maps them to target skills or rules, and updates the flag to mark them as promoted.

Episodic memory lessons remain unreflected in skills and rules because the promoted flag in failures.jsonl and decisions.jsonl has not been updated. Running the distillation process identifies unpromoted items, maps them to target skills or rules, and updates the flag to mark them as promoted.