Project Setup: App Idea to PRD, MVP Backlog & Build Plan

Tell your coding agent your app idea, or point it at an existing repo, and get a PRD, a UX/UI spec, an MVP backlog of epics and user stories, and agent rules before any product code is written. ## What it does **The problem.** AI coding agents start building before anyone decided what the product is, which feature comes first, what "done" means, or which assumption could sink the whole thing. You end up with a vague task list, half-built features in the wrong order, and an agent that says "done" when nothing was checked. **What you get.** Project Setup works like a product manager, tech lead and QA lead in one pass: - **Product plan (PRD):** problem, target users, MVP scope and non-goals, success metrics, risks and open questions. - **UX/UI design spec** for apps with a screen: information architecture, key screens, components, accessibility, and loading, empty and error states, tailored to web, SaaS, mobile, desktop or browser extension. - **System design basics:** architecture map, stack constraints, security rules and performance budgets in an engineering guide. - **MVP backlog by epic:** user stories with IDs (APP-001, APP-002...), priority, size, dependencies, and acceptance criteria your agent can check with a command. - **Riskiest assumption first:** the one thing that could kill the project becomes story 001, so you find out in week one, not month three. - **Agent rules** for Claude Code, Cursor, Codex, Gemini CLI and Antigravity from one `AGENTS.md`, plus `scripts/verify-project.sh`, a test script that fails until real checks exist. **Why not just prompt for it.** The skill ships 8 templates, 4 reference playbooks and 3 scripts: a bootstrap with a dry run that never overwrites a file, a validator that fails while any template placeholder is left unfilled, and a verifier scaffold that detects Rust, Node.js, Go, Java, Python and .NET and writes a failing `TODO` when it cannot recognize your stack. An unconfigured project never reports a fake pass. **Use it when** you start a new product, turn an app idea into a build plan, or adopt an existing repository for AI-assisted delivery. **Known limits.** - Software projects only. - It writes a UX/UI spec, not mockups or images. - "System design" here means architecture, constraints and budgets, not a full distributed-systems design document. - Generated verifier commands are a starting point; review them before relying on them. ## Example: a new app idea ~~~text You: I want to build a habit tracker for runners. Set up this project. Story prefix RUN. Agent: Classification: greenfield, mobile + web app, MVP, solo builder, single-agent delivery. Riskiest assumption: we can read run data from Strava and Apple Health without a paid API tier. RUN-001 proves it before anything else is built. Epics: E1 Onboarding & accounts, E2 Run sync, E3 Habit streaks, E4 Reminders. Backlog: 14 user stories, each with a command or a screenshot that proves it done. Files to create (nothing written yet): docs/PROJECT_PROFILE.md, docs/PRODUCT_PLAN.md, docs/DESIGN_SPEC.md, docs/BACKLOG.md, docs/ENGINEERING_GUIDE.md, specs/_TEMPLATE.md, notes/_TEMPLATE.md, AGENTS.md, CLAUDE.md, GEMINI.md, scripts/verify-project.sh. Approve? ~~~ ## Where it fits 1. Project Setup **(this skill)**, once per repository: PRD, epic backlog, verifier 2. Multi-Agent Story Flow or Dual-Session Story Flow (free), once per story: build, verify, independent review 3. You: read the diff, push, open the PR, merge 4. Merged Worktree Cleanup (free), after the merge: remove the worktree and branch ## How to use 1. Connect your agent to the Agensi MCP and ask for **Project Setup**. Keep one shared copy in `~/.agents/skills/project-setup`: Codex CLI and Gemini CLI read it directly. Point other agents at the same folder instead of making a second copy: - Claude Code: `ln -sfn ~/.agents/skills/project-setup ~/.claude/skills/project-setup` - Cursor: `ln -sfn ~/.agents/skills/project-setup ~/.cursor/skills/project-setup` - Antigravity: `ln -sfn ~/.agents/skills/project-setup ~/.gemini/config/skills/project-setup` - Windows: `mklink /J "%USERPROFILE%\.claude\skills\project-setup" "%USERPROFILE%\.agents\skills\project-setup"` Have the ZIP instead? `unzip -o project-setup-1.2.0.zip -d ~/.agents/skills/project-setup`. Keep only this one copy: a second install of the same skill can make the agent pick the wrong one. 2. Say `set up this project`, `write a PRD and backlog for my app idea`, or `set up this repo for the story flow`. Add a story prefix if you have one (`prefix CAN`). 3. Review the classification, the riskiest assumption and the dry-run file list, then approve. Nothing is written before you approve. ## Requirements Python 3.7+ (standard library only), bash and git, plus any coding agent that loads SKILL.md skills: Claude Code, Codex CLI, Cursor, Google Antigravity, Gemini CLI, Windsurf, GitHub Copilot or OpenCode. On Windows, run the scripts from Git Bash or WSL; the files written to your repo contain no symlinks, so the repo itself is Windows-safe. ## Changelog **1.2.0** - `--instructions gemini` and `--instructions all`: one `AGENTS.md` holds the rules, and `CLAUDE.md` and `GEMINI.md` import it with `@AGENTS.md`. If `CLAUDE.md` or `GEMINI.md` already holds rules and there is no `AGENTS.md`, setup keeps that file whole and advises merging instead of writing a second rules file. - `--story-prefix` writes your prefix (for example `CAN-001`) into the profile and backlog; story 001 is reserved for the riskiest assumption. - The bootstrap never writes through a symlink and creates files atomically. - The generated verifier runs unittest with `-v`, works before the first commit and outside Git, and the scaffold never writes through a symlinked `scripts/verify-project.sh`. - The engineering guide's reusable story prompt no longer trips setup validation. **1.1.0** - CLAUDE.md imports AGENTS.md with `@AGENTS.md` instead of a symlink, so generated repos are Windows-safe; portable across SKILL.md agents.