nelsonwerd/idea-to-ship-skills

idea-to-ship is a collection of reusable agent skills for Claude and OpenAI Codex that help solo builders and small teams work through the parts of building software that come before and after writing code. The skills are designed for people who tend to start coding on a new idea and later find the idea was not worth building, or that a build got partway and stalled. The collection has five skills. The first, ideate, takes a vague idea and turns it into a written concept document that includes a success metric and a kill criterion, meaning the condition under which you would stop working on it. It refuses to produce a build plan until the idea survives an honest pressure test. The second, deep-dive, is an investigative analysis tool: it runs parallel specialist agents on a question, synthesizes the findings, puts them through an adversarial review, and produces a plain-English verdict with a confidence score. It works on code audits, strategy questions, and design reviews, not just new ideas. The third skill, prompt-pack, converts a settled concept into a sequence of self-contained build prompts. Each prompt covers one unit of work, specifies what to verify when done, and is designed to leave the app working before the next prompt runs. The fourth, build-loop, drives a build past the point where it compiles by taking screenshots of the running app, critiquing the output, rebuilding, and repeating until defined acceptance criteria pass or a stop condition fires to prevent infinite retries. The fifth, autopilot, chains the other four skills and runs the full pipeline from idea to first-draft shipped app, operating autonomously and handing back a ledger of what still needs human judgment or market validation. The skills are composable: you can run any one on its own, run them in sequence yourself, or let autopilot fly the whole pipeline. The README includes a worked example showing excerpts of real output from each skill.