wa:learn

To capture session learnings into durable memory, you provide a learning observation and the Skill classifies it into categories like workflow-tactic or tool-pattern, then persists it with deduplicated writes. This ensures future evolve analysis can route and reuse the insights.

You can store error patterns and heuristics for workflow routing, tool patterns, and context issues. The Skill includes a correction mode that updates the error catalog and creates a heuristic with a baseline confidence for future reference.

To avoid duplicate memory entries when appending new learning observations, the Skill applies staged parsing, classification, and routing, followed by a confirmation step. It performs deduplicated writes by appending or updating topic entries under memory directories.

You should update the system-wide memory file with high-impact learnings when a lesson has broad applicability across evolving workflows. The Skill optionally writes these significant observations to memory/MEMORY.md so they remain accessible for future analysis.

Yes, learning capture requires YAML-frontmatter metadata for proper discovery and classification. The Skill uses this structured metadata during its staged parsing process to route observations into the correct categorized memory locations.

The best way to record a tool-pattern observation is to run the capture command with your learning text. The Skill extracts the learning, classifies it as a tool-pattern, and performs a deduplicated write to persist it for future evolve analysis.