Stack Maturity Matrix

P: SSOT for how mature each shipped stack template actually is — the honest answer to "is this stack production-ready?", separating breadth (how many stacks exist) from depth (how many are validated). Closes the claim-vs-maturity gap (strategic-audit 2026-06, report §S3). R: Before claiming stack coverage in marketing/docs, before promoting a stack, or when deciding whether to add a new stack vs deepen an existing one (report §S4 depth-over-breadth). S: You only need the count of stacks — that is ls src/templates minus the infra dirs (_base, _presets, meta). N: docs/engineering/doc-system-overhaul-roadmap.md, docs/playbooks/template-authoring.md, docs/architecture/adr/0010-consumer-distribution-version-gate.md

Nav: Governance Index | Docs Index

Why this exists

Coding OS ships many stack templates. Stack count is the wrong headline metric: a stack that nobody has validated end-to-end is a liability, not a feature — an agent will trust its scaffold and conventions, and if they are wrong the loop bakes drift into a real consumer project. The strategic audit found the repo claimed broad coverage while only a handful of stacks had any automated validation. This matrix is the truthful breakdown, and a test (tests/test_stack_maturity.py) re-derives it from ground truth so the doc cannot silently drift.

Maturity is earned, not declared. A stack does not get to call itself "stable" in its own stack.yaml; it earns the tier by what objectively exists in the repo. That is why the tier is derived from three signals rather than a hand-set field (which can lie and rots): the audit's "add a maturity: field" prescription was deliberately rejected in favor of derivation.

Tier definitions (the derivation rule)

Tier	Earned by (objective signal)	What it means for a consumer
Stable	A golden fixture exists at `tests/golden/claude_<stack>` — i.e. the full render is captured and parity-tested in CI	Scaffold + adapter render are validated; safe to build on.
Beta	A full framework overlay (`stack.yaml` with dimensions + skills + scaffold) but no golden fixture	Usable, but the render is unverified — review the scaffold before trusting it.
Stub	Named `*-plain` — a language-only skeleton (no framework opinion, minimal overlay)	A starting point only; expect to author conventions yourself.

Infra directories under src/templates/ are not stacks and are excluded from this matrix: _base (shared base overlay), _presets, and meta (this repo's own dogfood stack).

The matrix (26 stacks)

Derived 2026-06-19; the test is the SSOT — if this table and the test disagree, the test wins and this table is stale.

Stable — 4 (golden-validated)

django · nextjs · node-express · vue-nuxt

Beta — 16 (full overlay, not yet golden-validated)

angular · aspnet-core · astro · fastapi · flutter · go · go-fiber · laravel · nestjs · python · rails · react-native · rust-axum · spring-boot · svelte-sveltekit · wordpress

Stub — 6 (`*-plain` language skeletons)

csharp-plain · go-plain · java-plain · ruby-plain · rust-plain · typescript-plain

Positioning consequences (read before claiming coverage)

Headline honestly. "27 stacks" is true but misleading. The honest line is "4 validated stacks, 16 in beta, 6 language skeletons." Marketing copy and the Hub should reflect tier, not raw count.
Depth over breadth (report §S4). Do not add a new stack while 16 betas lack golden validation. The cheapest, highest-trust improvement is promoting a beta to stable by adding its golden fixture — not minting stack #27.
The promotion path is mechanical. Add tests/golden/claude_<stack> (and the codex twin where applicable) via make golden-capture SECTION=<id>, confirm tests/test_golden_parity.py is green, and the stack auto-promotes to Stable in this matrix on the next derivation — no edit to this doc's logic required.

Drift guard

tests/test_stack_maturity.py re-derives the three sets from the filesystem and asserts this document lists exactly those stacks per tier. Adding, removing, renaming, or golden-validating a stack without updating this table fails that test. This is the same anti-drift contract that the stack-count lint enforces for the "N stacks" literals in AGENTS.md (audit group A / TASK-461).