compound-learnings

Convert session learnings into reusable skills by extracting patterns, building frequency tables to identify recurring themes, consolidating similar patterns, and categorizing them as rules, skills, or agent updates based on evidence thresholds. The workflow reads learnings, applies signal detection, and outputs approved artifacts with frontmatter and examples to appropriate directories.

Extract patterns by identifying sections in learnings marked as Patterns, Takeaway, What Worked, What Failed, and Key Decisions. Build a frequency table to surface recurring themes, then consolidate similar patterns and detect clustering needs. High-frequency, well-evidenced patterns become candidates for skills; lower-signal items become rules or documentation.

Yes. The workflow reads learnings from sessions, automates pattern extraction, builds frequency tables, consolidates duplicates, detects clustering, applies thresholds, and outputs approved artifacts. This enables you to systematically convert ephemeral session insights into permanent, reusable capabilities across Claude projects without manual compilation.

The workflow categorizes artifacts by signal strength and complexity. Learnings with high frequency, broad applicability, and multi-step reasoning become skills. Simple, frequently-applied insights become rules. The process applies thresholds to distinguish between artifact types, ensuring only well-evidenced patterns are promoted to skills.

Automated compilation applies frequency analysis, clustering detection, and signal thresholds to identify which patterns deserve codification, reducing manual judgment and bias. Manual documentation captures ad-hoc insights; compilation ensures only high-signal, recurring patterns become permanent capabilities, improving reuse across projects.