This runs ahead of brainstorming to answer "what's worth exploring?" rather than "how should we build it?" It grounds in your actual codebase first, generates a batch of ideas, then explicitly critiques and rejects most of them with reasons instead of just ranking everything optimistically. The output is a ranked artifact in docs/ideation/ that routes into ce-brainstorm for the detailed work. It can resume recent ideation sessions if you left something open, and it has a surprisingly careful scope gate that asks whether you actually named a subject before spinning up the machinery. The questioning discipline is strict: it only asks what's needed to identify a topic, never solution constraints or audience, because those belong in brainstorming. Honest take: the rejection mechanism is the real quality lever here, not the generation volume.
npx -y skills add everyinc/compound-engineering-plugin --skill ce-ideate --agent claude-codeInstalls into .claude/skills of the current project.
Select a file.
juliusbrussee/caveman
mattpocock/skills
shadcn/improve
obra/superpowers
forrestchang/andrej-karpathy-skills
vercel-labs/skills