A planning-only repo. No code gets written here.
Open it in Claude Code, Codex, or Gemini CLI. Run /plan (Claude Code) or just describe
what you want to build. Walk through a short interview. Walk away with a /outputs folder
ready to drop into your next repo.
outputs/
├── PRD.md ← what you're building and why
├── STACK.md ← tech decisions with rationale
├── TASKS.md ← first sprint
└── [TOOL].md ← context file for your next repo (named for your AI tool)
The context file is named for the tool you'll build with: CLAUDE.md for Claude Code,
AGENTS.md for everything else (Codex, Gemini CLI, Cursor, and most coding agents now
read AGENTS.md natively — it's become the cross-tool standard).
Drop outputs/ into your build repo. Your AI assistant has full context from the
first message. No re-explaining.
Using mvp-builder? Rename the context file to
CONTEXT.mdbefore dropping it in.
Companion template:mvp-builder — picks up where this repo leaves off.
Option 1 — Use as a GitHub template (recommended): Click the "Use this template" button on GitHub to create your own copy.
Option 2 — Clone directly:
git clone https://github.com/ContractorKeith/project-planner.git my-project-planner
cd my-project-plannerThe core logic is stored in skills/orchestrator.md. But you don't need to do anything manually because we provide proxy files out of the box for the most common tools. Each AI tool will read its respective proxy file automatically:
| Tool | Setup |
|---|---|
| Claude Code | Works out of the box — reads CLAUDE.md automatically, plus a /plan command |
| OpenAI Codex | Works out of the box — reads AGENTS.md automatically |
| Gemini CLI | Works out of the box — reads GEMINI.md automatically |
| Cursor | Works out of the box — .cursor/rules/ (legacy .cursorrules also included) |
| Windsurf | Works out of the box — .windsurf/rules/ (legacy .windsurfrules also included) |
Start a session:
/plan
Or just describe what you want to build. The orchestrator will take it from there.
Heads up: don't use
/initin Claude Code — that's a built-in command that generates a CLAUDE.md, not the planning entry point.
project-planner/
├── .claude/commands/plan.md ← /plan command for Claude Code
├── .cursor/rules/ ← IDE instructions for Cursor (+ legacy .cursorrules)
├── .windsurf/rules/ ← IDE instructions for Windsurf (+ legacy .windsurfrules)
├── CLAUDE.md ← proxy to orchestrator
├── GEMINI.md ← proxy to orchestrator
├── AGENTS.md ← proxy to orchestrator
├── README.md
├── LICENSE
├── CONTRIBUTING.md
│
├── skills/
│ ├── orchestrator.md ← core orchestrator logic
│ ├── skill-creator.md ← create new skills mid-session
│ ├── problem-framing.md ← extracts the problem
│ ├── stack-selection.md ← defines the technical shape
│ ├── prd-writing.md ← generates /outputs/
│ ├── plan-refiner.md ← refines an existing plan
│ ├── plan-eng-review.md ← engineering stress-test of the plan
│ ├── phase-gates.md ← final checklist before handoff (community)
│ └── multi-model-review.md ← second-model plan review (community)
│
├── examples/ ← sample outputs from a fictional project
│ ├── README.md
│ ├── PRD.md
│ ├── STACK.md
│ ├── TASKS.md
│ └── CLAUDE.md
│
└── outputs/ ← generated per session, gitignored by default
The examples/ folder contains sample outputs from a fictional project (TaskTrail — a CLI task tracker).
Browse them to see what your own /outputs/ folder will look like after a session.
If a planning need comes up that isn't covered by an existing skill, ask the AI to
create one using skills/skill-creator.md. It lands in /skills/ and is available
immediately and in every future session.
project-planner → outputs/ → build repo → ship
Each repo has one job. This one's job is planning.
See mvp-builder for the next link in the chain.
MIT