wtf is vibe-coding-prd-skill?

This repository is a Codex skill, not a standalone application. It is designed to be dropped into a coding agent's skills folder so that Codex can turn messy product materials into a clean, coding ready product requirements document, often called a PRD. The problem it addresses is that raw product inputs are usually a poor fit for handing directly to a coding agent: requirements are scattered and missing details about users and scope, original PRDs are bloated with business context that a coding agent does not need, competitor research is often just screenshots without concrete features, and acceptance criteria are too vague for an agent to know when a task is actually done. A user triggers the skill with a plain request, such as asking it to write a PRD for vibe coding, or to turn a messy internal document into something that can be sent to a coding agent. The skill first figures out what kind of input it has received: a clear requirement, a raw engineering PRD, a competitor analysis, a document link from Feishu or Lark, or an unclear request that needs clarifying questions before it can proceed. Depending on the input type, it either confirms the requirement definition, strips business context from a heavy PRD down to what an engineer needs, extracts reusable features and interaction patterns from competitor research, or asks the user multiple choice style questions to pin down what they actually want. Whichever path it takes, the skill converges on the same output structure: a feature list with priorities, a recommended technology stack, a rough design for the core pages, detailed feature and agent workflow descriptions, data model and API design notes, edge cases and fallback behavior, and clear acceptance criteria, all written as a single Markdown PRD ready to hand to a coding agent like Claude Code or Codex. Installing it is a matter of cloning the repository into the local skills directory and restarting Codex. The repository itself is small, containing just a SKILL.md definition file and an agent configuration file. The README suggests giving the skill as much upfront context as possible, including the target users, the product form, whether an MVP is wanted, any technology stack constraints, and whether AI or agent features are needed, and notes that missing details will be asked about rather than assumed.