compound-learnings

Transform ephemeral session learnings into permanent, compounding capabilities. Read the most recent 5-10 files (or specify a date range). For each learnings file, extract entries from these specific sections: Build a frequency table as you go: Before counting, merge patterns that express the same principle: Example consolidation: Use the most general formulation. Update the frequency table. Critical step: Look at what the learnings cluster around. If >50% of patterns relate to one topic (e.g., "hooks", "tracing", "async"): → That topic may need a dedicated skill rather than multiple rules → One skill compounds better than five rules Ask yourself: "Is there a skill that would make all these rules unnecessary?" For each pattern, determine artifact type: Artifact Type Examples: Present each proposal in this format: Use AskUserQuestion to get approval for each artifact (or batch approval). Create .claude/skills/ /SKILL.md with: Add triggers to skill-rules.json if appropriate. Create shell wrapper + TypeScript handler: Then create src/ .ts , build with esbuild, and register in settings.json : Edit existing agent in .claude/agents/ .md to add the learned capability.