When a task feels too big to estimate or you're paralyzed by where to start, this breaks it down into manageable pieces using vertical slicing, walking skeletons, and dependency analysis. It's grounded in cognitive limits (working memory caps at 7 concepts, tasks over 3 days have 60% completion rates vs 95% for sub-2-hour chunks) and includes diagnostic states like "no clear entry point" or "scope creep" with specific interventions for each. The checklist is solid, the anti-patterns section calls out real problems like unbounded research, and it pushes you toward shipping smallest viable versions. Pairs well with github-agile for actually tracking the decomposed work. Best used before estimation, not during implementation.
npx skills add https://github.com/jwynia/agent-skills --skill task-decomposition