§ Commands · The catalog
25 commands. Each opens a specific conversation with your AI tool.
Spruce is structured as five tiers. They aren’t levels of capability — they’re shapes of conversation. Picking the right tier matters more than picking the right command within it.
For what running the full catalog produces on a real product, see the Stillpoint case study.
Setup
One command, run before everything else. /spruce-up captures your project’s character — what you’re building, who uses it, what voice and visual direction you want — into a context file the rest of Spruce reads from. Every other command works better when this exists; many work poorly without it.
Discovery
Reasoning about who the design serves, before reasoning about how it should look. The Discovery tier produces the HCD artifacts — personas, jobs-to-be-done, journeys, scenarios — that downstream commands read from when they need to ground a decision in a real user doing a real thing. Each command runs in three modes: draft from context when no research exists, structure user-supplied research when it does, or pressure-test a finished artifact for assumptions. /audit is the diagnostic counterpart — the only diagnostic command that frames findings against named personas + jobs rather than against general principles. Discovery is optional; every other tier can run without it, and most do. But work that’s grounded in Discovery reads as work made for someone specifically rather than work made for the average of training data.
Generative
Producing new design work. The generative loop runs in sequence: /sketch establishes visual direction first; /foundations codifies it into tokens and primitives; /design produces a first-pass surface from that foundation; /remix surfaces alternative directions on that first pass; /decide guides specific decisions when you’re adjusting or extending — adding features, refining a section, ideating on top of the chosen direction. Each command can stand alone, but together they shape the path from a single context file to a designed surface and beyond.
Diagnostic
Review and explanation, no code changes. The lens you reach for when the work feels off and you need a structured read on why. Diagnostic commands produce findings — formal audit, narrative critique, fast pattern scan, decision walkthrough — that you decide what to do with. They don’t presume.
Corrective
Discipline within a single dimension. The pass you reach for when one layer has drifted: typography looks generic, the palette has gone purple-and-blue, components are inconsistent, the voice has slipped into “Submit / Get Started” SaaS slop. Each corrective addresses one layer cleanly and leaves the rest alone.