What is designer-led-agent-starter?

Designer-Led Agent Starter is a workflow template for product designers and non-engineers who want to use AI coding assistants to build a minimum viable product from rough design ideas. The core problem it solves is that AI coding agents tend to jump straight into writing code before they fully understand what the product is supposed to do, producing something that looks finished but is missing key logic, edge cases, or constraints. The template addresses this by giving the AI agent a structured set of source-of-truth documents to fill out before any coding begins. These cover product rules, user roles, feature flows, architecture decisions, environment differences between local and production, risk concerns, and acceptance criteria. The agent is instructed to read these docs, ask clarifying questions like a product manager, write discoveries back into the docs, and only then implement one capability at a time. Documentation is not a one-time setup step: every build pass may reveal missing rules or undefined states, and the agent keeps the source docs updated throughout. The template is bilingual, offered in English and Traditional Chinese, and requires no slash commands. You work with plain natural-language instructions. It includes eighteen numbered document types covering everything from raw design notes to a contradiction audit, plus an AGENTS.md file that sets the agent's operating rules. The project grew out of roughly three months of real product work on a creator platform called HAIBUNKA, which produced an 80,000 word design document, over 200 specification and audit files, and more than 500 implementation files, so the workflow reflects lessons from an actual build rather than a theoretical outline. This would be useful to product designers, PMs, or founders who want to build an MVP with AI assistance but lack an engineering background, and want a process that keeps the AI from getting ahead of the product definition. It is released under the MIT license.