OpenPlanr overview
What OpenPlanr is, how its plan, design, build, review, and operate loop works, and where to start in Claude Code, Codex, or Cursor.
OpenPlanr is the shared delivery loop for product teams and their AI agents. You plan, design, build, review, and operate from durable context in your repository, with 25+ skills for Claude Code, Codex, and Cursor, plus the deterministic openplanr CLI.
What it is
- Skills, not a second model. Every skill runs inside the coding agent you already use. Skills and the openplanr CLI add no model calls or telemetry; the optional design engine calls OpenAI only when you select its OpenAI provider and supply your own key.
- Plans are files. Specifications, user stories, tasks, and provenance live under
.planr/in your repository, reviewed and versioned like code. See Plans are files. - A deterministic CLI.
openplanrstores and validates planning files, renders diagrams and reports, installs the skills into each agent, diagnoses installations, and syncs with GitHub Issues and Linear. - Open source under MIT. One repository, published as 3 npm packages:
openplanr(the CLI and host packages),planr-pipeline(the delivery pipeline), and@openplanr/protocol(schemas and registries).
OpenPlanr is not a hosted project tracker and does not replace your issue tracker. It closes the loop: your agent reads the plan before it changes code, and what it ships flows back as evidence for the next review.
How it works
Reasoning stays in your agent. The CLI never calls a model; it validates what the agent wrote, renders it, and keeps trackers in step.
Every step is a skill your agent selects from its description, or one you invoke by name:
| Stage | What you get | Guide |
|---|---|---|
| Plan | A spec with measurable acceptance criteria, then user stories and tasks with exact file scope | Plan a feature |
| Design | One design direction in a review studio, and a design spec that Plan reads | Design before you build |
| Build | One task at a time, verified and small enough to review as one pull request | Build one task at a time |
| Review | A plan review before building, browser QA after it, and a readiness check before merge | Review before it lands |
| Operate | Delivery status from the plan, and leadership reviews that end in decisions | Track delivery and run operating reviews |
Plan and ship stay separate steps that you invoke. The agent never chains them on its own.
Start here
- Install and set up OpenPlanr: the CLI, the skills for your agent, and the planning folder in your project. To let your agent do it for you, see Set up from your agent.
- Write your first spec.
- Read the page for your agent: Claude Code, Codex, or Cursor.
- Copy a prompt from the prompt library, starting with the ones marked Start here.