When you need to solve something complex and aren't sure which approach is best, this runs three agents in parallel to propose different solutions, then uses meta-judges to create custom evaluation criteria before voting on which proposals to pursue. The strong bit is the adaptive strategy: if there's a clear winner it polishes that solution, if everything scores poorly it redesigns from scratch, and if opinions split it synthesizes the best parts. It's structured around eight phases with proper pruning and expansion, saving all the research artifacts to `.specs/research` so you can see how it arrived at the answer. Probably overkill for straightforward tasks, but useful when the problem space is genuinely unclear and you want systematic exploration rather than going with your first instinct.
npx skills add https://github.com/neolabhq/context-engineering-kit --skill tree-of-thoughts