Skip to content

Quickstart

Zero to your first governed workflow in a few minutes. This page covers both supported runtimes — Claude Code and Codex.

Prerequisites

  • Node.js ≥ 18.13 (node --version)
  • One AI runtime installed: Claude Code or Codex CLI
  • PowerShell (Windows PowerShell 5.1+ or pwsh) — required by the spec/ops/code runtime assets
  • Python 3.8+ only if you plan to use sdtk-spec atlas (the wiki atlas, sdtk-wiki atlas build, is native Node and needs no Python)

1. Install the suite

bash
npm install -g sdtk-kit

This puts seven commands on your PATH: the unified sdtk plus sdtk-spec, sdtk-design, sdtk-code, sdtk-ops, sdtk-wiki, and sdtk-agent.

bash
sdtk --version

2. Initialize your project

From your project root, run one command for your runtime:

bash
sdtk init --runtime claude
bash
sdtk init --runtime codex

sdtk init chains each toolkit's setup in order: spec → ops → code → design (runtime-aware, installing skills), then wiki → agent (plain init — these two are CLI-driven and install no skills).

  • Claude Code defaults to project scope: skills land in .claude/skills/ inside your repo.
  • Codex defaults to user scope: skills land in $CODEX_HOME/skills/ (default ~/.codex/skills/).

See Installation for scopes, standalone kits, and update/uninstall.

3. Run your first governed workflow

Option A — start from an idea (spec pipeline)

Open your runtime in the project and invoke the orchestrator:

text
/orchestrator
I want to build a lightweight CRM for solo consultants to track leads.
text
Use the orchestrator skill.
I want to build a lightweight CRM for solo consultants to track leads.

The orchestrator classifies your request, runs discovery, and — once the requirement is ready — walks the PM → BA → ARCH → DEV → QA phase gates, writing real spec artifacts into docs/. See Idea to Ship.

Option B — start from an existing codebase (memory layer)

bash
sdtk-wiki init
sdtk-wiki ingest
sdtk-wiki compile --mode safe
sdtk-wiki atlas build
sdtk-wiki atlas open

This builds a local knowledge graph from your repo and opens the Atlas viewer in the browser. See Existing Codebase.

Option C — a durable multi-step agent run

bash
sdtk-agent init
sdtk-agent workflow validate --file ./workflow.json
sdtk-agent run start --workflow ./workflow.json --runtime-map ./runtime-map.json \
  --feature-key MY_FEATURE --goal "Describe the goal"
sdtk-agent run continue --run-id <run_id>

The run lives entirely in a file ledger under .sdtk/agent-runtime/runs/<run_id>/ — pause it, close the terminal, and continue tomorrow. See Durable Agent Runs.

4. Verify your setup

bash
sdtk-spec runtime status
sdtk-code doctor
sdtk-wiki --version

If anything looks off, head to Troubleshooting → Installation.

Where to go next

SDTK — governed, auditable, resumable AI-assisted engineering.