Maturity & Versions
The authoritative "what's real" page. Every capability on this site carries one of five badges; this is the legend and the full truth table.
Last verified against published npm packages: 2026-07-14.
Badge legend
| Badge | Meaning |
|---|---|
| Shipped | Published to npm and supported. Documented commands work as described. |
| Preview | Published but intentionally limited (e.g. dry-run only). Boundaries stated on its page. |
| Pro | Shipped, requires a Pro entitlement. Fails closed without one. |
| Gated | Planned/underway but not shipped. Do not expect it in any installable version. |
| Unpublished | Implemented in source but not yet in the published npm package. Arrives with the next publish. |
Published versions
| Package | Version | CLI |
|---|---|---|
sdtk-kit (umbrella) | 1.53.0 | sdtk |
sdtk-spec-kit | 0.9.2 | sdtk-spec |
sdtk-design-kit | 0.14.1 | sdtk-design |
sdtk-code-kit | 0.4.4 | sdtk-code |
sdtk-ops-kit | 0.2.10 | sdtk-ops |
sdtk-wiki-kit | 0.18.1 | sdtk-wiki |
sdtk-agent-kit | 0.5.0 | sdtk-agent |
sdtk-agent-hermes-adapter | 0.3.7 | (adapter, loaded by sdtk-agent) |
sdtk-brain-kit | 0.2.2 | sdtk-brain (standalone — not in the umbrella) |
Check any package yourself: npm view <package> version.
Truth table
| Capability | Status |
|---|---|
sdtk umbrella: init / runtime / update / doctor / activate | Shipped |
sdtk umbrella: login / logout (named account slots) | Shipped |
sdtk umbrella: account (status / link / guide) | Shipped |
sdtk umbrella: usage (token meter) | Shipped |
sdtk umbrella: statusline (opt-in, Claude + Codex) | Shipped |
| SDTK-SPEC: init, generate, analyze, config (apply-profile/check), atlas, runtime, entitlement | Shipped |
SDTK-SPEC: project ingest/audit/refresh (Project Intelligence) | Pro |
| SDTK-DESIGN: full command surface incl. Preview Studio | Shipped |
| SDTK-CODE: start/plan/build/verify/ship + status/doctor/resume | Shipped |
| SDTK-CODE: guardrails, readiness | Shipped |
SDTK-CODE: sleep plan / sleep report (dry-run preflight) | Shipped |
SDTK-CODE: Sleep Execute — unattended execution (attended sleep authorize → headless sleep execute; doc-class, hard ceilings, BYO API key) | Pro |
| SDTK-CODE: agent-team (execution bounded to docs/governance Markdown) | Shipped |
| SDTK-OPS: init/update/runtime + ops-* skill journeys | Shipped |
| SDTK-WIKI: query/search/atlas/kanban/maintain (lint-only) | Shipped |
SDTK-WIKI: ask — cited retrieval | Shipped |
SDTK-WIKI: ask — synthesized grounded answer | Pro |
SDTK-BRAIN: ingest/compile/discover/enrich second-brain pipeline (standalone sdtk-brain-kit) | Shipped |
| SDTK-AGENT: init, workflow validate, run start/continue/status/report, gates | Shipped |
SDTK-AGENT: run cancel, task retry | Shipped — published in sdtk-agent-kit@0.4.0 |
| Hermes Kanban adapter (dry-run plan compiler) | Preview |
| Hermes live dispatch (cards/bots/messaging) | Gated — validation-only; source is published in sdtk-agent-hermes-adapter@0.3.7 but live stays gated pending real-box evidence |
| Workflow packs | Gated — no pack format/runner ships today |
Standing truth statements
These hold everywhere on this site:
- Local-first, precisely. Repo-owned artifacts (specs, ledgers, graphs, reports) stay in your repository. External model/API/runtime calls happen only when you explicitly invoke a runtime (a Claude Code/Codex session, or
sdtk-wiki askon Pro — the free Ask tier is pure local retrieval and calls nothing). - Unattended execution is opt-in and bounded. The free product runs nothing while you are away —
sleep plan|reportare preflight only. Sleep Execute (Pro) is the sole unattended path: it runs only after an attendedsleep authorize, stays doc-class, and is capped by hard ceilings (10 iterations / $10 / 24h) with a per-iteration readiness re-check, pinned scope, and worktree isolation. - No live Hermes yet. The published adapter (
sdtk-agent-hermes-adapter@0.3.7) contains live-mode source, but live dispatch is gated behind a fail-closed preflight (config +live_ack+ credential + gateway/board checks) and remains validation-only — not public-ready — until real-box evidence and independent review land. Treat Hermes as dry-run/mock for normal use; never claim live cards/bots/messaging work. - Evidence trust is same-machine in v0.x — no remote-worker authenticity claims.
- Docs track the installable surface. If a capability is in source but not published, it's badged Unpublished here and not documented as usable.