AI marketing skills, scored

Each one benchmarked on real briefs, against the bare model. How scoring works

Point Updraft at a piece of AI-generated capability — a skill folder, an agent bundle, a prototype, a knowledge base entry — and it decides whether the thing is worth adopting, worth tuning, or worth rejecting, by assessing it against the way the job is done today: your rubric, your golden examples, your registry of adopted capabilities, your knowledge base. When a candidate earns adoption, Updraft integrates it: registry entry created, changelog written, knowledge base proposal raised for human ratification.

Every assessment also scores the bare model with no capability attached at all, so a result can be compared against what you get by just asking. One engine and one Postgres database sit behind three surfaces — an MCP server, a CLI and this review app — and all three call the same engine code.