task-memory-consolidation

Consolidation parses memory.yaml to extract learnings, groups similar ones by topic, identifies clusters with 3+ learnings, maps them to target skills, and generates updates including new sections and anti-patterns. This reduces cognitive load and accelerates future work by converting accumulated experience into durable skill improvements.

Consolidate during RSID Ideate, when memory.yaml grows beyond approximately 10 tasks, or when learnings recur across multiple tasks. The six-step pipeline—parse, group, threshold, map, generate, and archive—automates progression from memory to integrated skills.

Consolidated memories are archived to memory-archive.yaml, separating active learnings from historical patterns. This keeps memory.yaml focused on recent work while preserving the full learning history for reference and future pattern detection.

The threshold applies a 3+ learnings requirement to grouped patterns. Only clusters meeting this threshold are mapped to target skills and generate updates, ensuring that only well-established recurring patterns become skill improvements rather than isolated observations.

Yes. The generate step produces skill updates that include new sections capturing consolidated patterns and anti-patterns documenting common pitfalls or mistakes. These updates integrate directly into target skills based on cluster-to-skill mapping.

Consolidation implements a flat Learning data model with explicit type taxonomy, simplifying parsing and grouping logic. This structure supports efficient clustering of similar learnings and straightforward mapping to skill updates.