FAQ

What is SDTK in one sentence?

Six local-first CLI toolkits around one governance kernel that turn AI-assisted software work into governed, auditable, resumable engineering — on Claude Code or Codex.

Is SDTK free?

The core suite is free and MIT-licensed. A one-time Pro purchase unlocks Project Intelligence (sdtk-spec project …) and grounded Ask (sdtk-wiki ask) on a machine. See Installation → Pro activation.

Does my code leave my machine?

SDTK's CLIs run locally and make no model calls of their own; repo-owned artifacts — specs, ledgers, graphs, reports — stay in your repository. External model/API/runtime calls happen when you invoke an AI runtime: a Claude Code or Codex session, or sdtk-wiki ask. SDTK is local-first, not air-gapped — it governs those calls rather than hiding them.

Which AI runtimes are supported?

Claude Code and the Codex CLI. Same skills, same artifacts; sdtk init --runtime <claude|codex> handles either. See Runtimes & Skills.

Do I need all six toolkits?

No. sdtk-kit installs everything, but each toolkit is a standalone npm package. Common single-kit entry points: sdtk-wiki-kit for a knowledge graph over an existing codebase, sdtk-agent-kit for durable agent runs.

Do I need an AI runtime at all?

For the CLIs themselves, no — scaffolding, graphs, ledgers, validation, and viewers all run without a model. The skills (orchestrator, role phases, design/ops journeys) need a runtime session to execute.

Can SDTK run agents unattended overnight?

Not today. sdtk-code sleep plan|report are dry-run preflight tools; unattended execution is not shipped. What is shipped is durability: an sdtk-agent run pauses at gates and resumes whenever you return — see Durable Agent Runs.

Does the Hermes adapter create real Kanban cards?

No. The currently published npm adapter compiles dry-run plans only and refuses mode:"live" fail-closed. Live dispatch remains gated/unpublished until a coordinated publish with real-box evidence and independent review — until then, treat Hermes as dry-run/mock only. See Hermes Adapter.

The docs mention a command my install doesn't have — why?

Check Maturity & Versions. Every documented command tracks the installable npm surface; if a verb is in source but not yet published, it's badged Unpublished there and lands with a later publish. (For example, sdtk-agent run cancel and task retry were Unpublished before sdtk-agent-kit@0.4.0 and are now Shipped.)

Windows support?

Yes. Node CLIs run natively; spec/ops/code runtime assets use PowerShell (5.1+ or pwsh); the agent runtime has hardened Windows dispatch (PATH×PATHEXT resolution, strict .cmd/.bat argument policy).

How is SDTK-AGENT different from just scripting the CLIs?

The kernel gives you what a script can't: a durable run ledger that survives interruption, a pure state machine anyone can resume, attributable human gates, adapter allowlists that fail closed, and an auditable event log + report. See The Governance Kernel.

Where do I report bugs or ask questions?

github.com/codexsdtk/sdtk-toolkit. Attach sdtk --version, sdtk-code doctor, and sdtk-spec runtime status output when relevant — see Troubleshooting.