Coding agents are remarkably capable, and yet how do we treat them? Sometimes by giving them stream-of-consciousness descriptions of half-baked ideas: vague feature requests, underspecified tickets, or maybe just good old vibes. We've got better implementation assistants than we ever could have imagined, but we're not always doing a great job thinking through what we want them to build. Would a lightweight spec document be so bad?

OpenSpec doesn't think so. Born from the recognition we've never really come to consensus on how to capture features—some do PRDs in Notion, some make epics in Jira, some still write user stories on index cards—it uses some Agent Skills and a CLI to guide you through structured feature exploration for greenfield and brownfield projects alike, producing detailed designs and todo lists precise enough to direct a coding agent even in the presence of some complexity. We'll trace how it emerged, see how it positions itself against approaches like Spec Kit and traditional design tooling, and examine the recent developments that have extended its capabilities.

The specification problem is older than agents, having bedeviled us since the first time someone handed a one of our grandparents a napkin sketch and called it a design doc. OpenSpec is surely not the final answer, but it's the right one for agentic engineering right now.