Getting Started

Install the plugin in your assistant, choose your starting point, then run the workflow that fits your project.

Before you begin

Claude Code, to use the /plugin commands below
a target repository open in your assistant
GitHub access only when you want to work from issues

Node.js, APM, and a clone of the SKRAFT repository are not required to use the plugin. They belong to development of the plugin itself.

1. Install SKRAFT with `/plugin`

Enter these commands in the Claude Code conversation, not in a terminal:

/plugin marketplace add SebastienDegodez/skraft-plugin
/plugin install skraft

The first command adds the repository as a marketplace. The second installs the skraft plugin published by that marketplace. Then open /plugin and check that skraft appears under installed plugins.

Windows workaround

If /plugin marketplace add fails with fatal: unable to checkout working tree, clone SKRAFT to a short local path and add that checkout instead:

git clone https://github.com/SebastienDegodez/skraft-plugin C:\s\skraft

/plugin marketplace add C:\s\skraft
/plugin install skraft

Or enable Windows long paths, restart the shell, and retry:

New-ItemProperty -Path "HKLM:\SYSTEM\CurrentControlSet\Control\FileSystem" -Name LongPathsEnabled -Value 1 -PropertyType DWORD -Force
git config --global core.longpaths true

2. Choose your journey

Do not automatically run one global chain. The right starting point depends on what your repository already has.

Your situation	Start with	Then
A refined story already exists	`skraft-orchestrator`	engineering pipeline
Issues exist but are not prepared	`backlog-discoverer`, then `backlog-planner`	`skraft-orchestrator`
Code exists without product documentation	`brownfield-analyst`	PRD, issue creation, product preparation, then `skraft-orchestrator`
Legacy code is unsafe to change	`brownfield-harness-builder`, then `brownfield-refactorer`	return to a prepared story, then `skraft-orchestrator`

The Brownfield, DISCOVER, and DISCUSS workflows are invoked directly. They are not hidden phases of skraft-orchestrator. Read the Brownfield journey when taking over an existing system.

3. Run the first workflow

Story already prepared

In the agent selector, choose skraft-orchestrator, then give it the refined story. This is the sole entrypoint for the engineering pipeline. It loads its persistent state and resumes from the last validated point.

Backlog still raw

First choose backlog-discoverer in the agent selector. Once triage is complete, choose backlog-planner to refine the selected issue. The refined story then becomes the input to skraft-orchestrator.

4. Understand what will run

The core journey has two distinct areas:

Optional standalone product preparation
- DISCOVER triages and prioritizes issues
- DISCUSS turns an issue into a verifiable story
Engineering pipeline driven by skraft-orchestrator
- RESEARCH reduces uncertainty when the work justifies it
- DESIGN makes and records architecture decisions
- DISTILL produces executable scenarios and the plan
- DELIVER implements with Outside-In TDD and gathers evidence

RESEARCH may be skipped when routing concludes that a dedicated investigation would add no value. DISCOVER and DISCUSS are never dispatched by skraft-orchestrator: you choose them before it when your input is not yet a ready story.