Files
Ekklesia/docs/architecture/repository-boundaries.md
T
ekklesiaandClaude Opus 5 07835b7b56 feat: give module UI real stylesheets, and every artifact its own root
ADR 0023 refused stylesheet imports for a good reason: esbuild would emit a
`.css` sidecar beside the artifact, nothing would declare it, no host would
fetch it, and it would work in a local test and silently do nothing in a real
host. Inline style objects were the whole story. They cannot express
`:hover`, `:focus-visible`, `:disabled`, `@media`, or `@keyframes` — so an
accessible focus treatment was not expressible, and the shared elements added
in the previous commit are built from real CSS for exactly that reason.

Compile the stylesheet into the artifact instead of beside it. A module
imports plain `.css`; the canonical builder resolves local `@import`s through
a nested bundle and turns each file into its compiled text, which the author
declares as `defineModuleUi({ styles })`. One declared file still contains
everything: nothing extra to publish, nothing to fetch, and no way for CSS to
become an asset nobody serves. The build additionally fails if a CSS output
ever appears beside an artifact, so a future loader change cannot bring the
sidecar back. Two rules are enforced with their own codes: a module stylesheet
may reference `--ekklesia-*` and never declare one, and every `@import` must
resolve locally or the artifact would fetch styles at run time. Plain `.css`
only; other dialects are a toolchain decision needing their own ADR.

Then make the isolation real. The generic loader mounts every browser artifact
— primary, display, and later a contributed renderer — inside an open Shadow
DOM root it creates, installs the artifact's stylesheet there, and hands the
artifact an element inside it. Shadow DOM is the one mechanism a browser
actually enforces; a class prefix or a "scope your selectors" instruction is a
convention one mistake defeats, and the failure mode is a module silently
restyling the Dashboard. It applies unconditionally, including to artifacts
with no stylesheet, because a conditional boundary is one a module can fall
out of by editing a source file. Both wrappers are `display: contents` through
base rules installed before the artifact's own, so layout is unchanged and a
full-viewport display can still override `:host`. Unmount removes the whole
root in a `finally`, so a cleanup that throws cannot leave styles behind, and
remounting cannot accumulate a second copy.

This is a correctness boundary for styles and is not described as a sandbox:
the root is `open` precisely because module code is trusted, same-realm code
that could reach `document` anyway.

An artifact's DOM is no longer a direct child of the host's element, so the
mount ABI is versioned explicitly. `defineModuleUi` stamps
`MODULE_UI_ABI_VERSION` so it can never drift from the SDK the artifact
compiled against, and a version-1 artifact fails with its own code — "rebuild
your module", not a typo hunt. No dual acceptance, matching every manifest
cutover here. Migration is one `pnpm build`.

`pnpm boundaries` allows `.css` only under `src/ui/` and `src/display/`: a
stylesheet anywhere else in a module is the undeclared sidecar again, just
relocated.

Co-Authored-By: Claude Opus 5 <[email protected]>
2026-08-27 07:21:14 -05:00

150 lines
12 KiB
Markdown

# Repository boundaries
These rules map platform ownership to filesystem and package boundaries. The executable check is
`pnpm boundaries`; CI must reject violations.
| Area | Owns | May depend on |
| --- | --- | --- |
| `apps/*` | Thin deployable composition roots | Public Core APIs, generated bootstrap, or a surface as required by the app |
| `core` | Feature-agnostic platform implementations | Public contracts and external infrastructure libraries |
| `modules/*` | Domain capability, data, and the module's own UI | Module SDK, module UI SDK, the approved design packages, public contracts, React, and external backend libraries |
| `surfaces/*` | The application shell, UI, renderers, navigation, and interaction | Surface SDK, API/auth clients, the module UI SDK and host loader, design packages, and a browser framework |
| `builders/*` | Declarative surface composition tooling | Builder/surface SDKs and public contracts |
| `packages/*` | Explicitly scoped SDKs, contracts, clients, design, testing, or tooling | Only dependencies appropriate to the package's declared level |
| `templates/module` | A small example backend module | The same dependencies allowed for any module |
## Mandatory rules
- Core never imports a feature module, surface, Builder implementation, or app.
- A module never imports Core implementations, another module, a surface, Builder code, or an app.
- A module may own UI ([ADR 0021](decisions/0021-modules-as-hostable-products.md)) and authors it in
React and TypeScript ([ADR 0022](decisions/0022-module-ui-authoring-and-host-contract.md)). A
module may therefore contain `.tsx` source and depend on `react`, `react-dom`, and
`@ekklesia/module-ui-sdk`. Since
[ADR 0040](decisions/0040-module-stylesheets-and-artifact-isolation.md) it may also contain plain
`.css`, but **only** inside a browser-artifact source root the canonical builder compiles
(`src/ui/`, `src/display/`); a stylesheet anywhere else in a module is a file no build reads and no
host loads. It still cannot contain `.jsx`, other stylesheet dialects, single-file components,
Astro/Marko/MDX/HTML views, or SVG presentation assets, and cannot depend on any other UI framework
(including `react-router-dom`, `react-native`, and `@vitejs/plugin-react`) or on any shared role
beyond the two approved design packages named below. A module needs no bundler plugin: `@ekklesia/module-ui-build` — a `tooling` package,
a role modules could already depend on — owns the entire module UI build
([ADR 0023](decisions/0023-module-ui-build-pipeline.md)). An unsupported framework aliased behind an allowed name is resolved and rejected by its
real target.
- Module UI source is subject to the identical layer rules as module backend source: it may not
import Core, an app, a surface, a builder, or another module. It also may not import
`@ekklesia/module-ui-host`: that package is host infrastructure, and a module bundling the loader
would blur the author/host split ([ADR 0024](decisions/0024-browser-safe-ui-sdk-and-host-loader.md)).
- A module may consume exactly two shared **design** roles: `design-tokens` (`@ekklesia/design-tokens`)
and `ui` (`@ekklesia/ui`), added by
[ADR 0039](decisions/0039-shared-design-tokens-and-elements.md). This is a named allowlist, not a
general "browser-safe shared packages are fine" relaxation: `api-client`, `auth-client`,
`surface-sdk`, and `module-ui-host` remain rejected inside a module. Both approved packages are
browser-safe by construction — `design-tokens` has no dependencies at all and `ui` depends only on
it — so a module UI file that imports them still cannot typecheck `process.env`, proven by a real
compile under the exact browser project a module uses. Both are also pinned at the bottom of the
presentation stack by `design-packages-depend-downward-only`: neither may reach an implementation
layer, an API/auth client, or the backend contracts barrel, because a design package that could
would put all of it inside every module artifact that renders a button.
- A module **references** registered global semantic tokens and may derive private,
differently-prefixed aliases from them inside its own artifact. It may never define an
`--ekklesia-*` custom property: that would invent a global theme variable outside the registry,
which a future theme editor could not enumerate and another module could unknowingly depend on.
- `@ekklesia/module-ui-sdk` has no dependencies at all, and `@ekklesia/module-ui-host` depends only
on it. Both are browser-safe by construction: neither can pull a backend type graph — and with it
Node's global declarations — into browser code. Core still owns the containing
application shell and all platform policy, and module UI reaches the host only through
`@ekklesia/module-ui-sdk`.
- A module stylesheet is compiled **into** the one declared JavaScript artifact as text and installed
by the host at mount; there is no sidecar file, no second publication route, and no manifest field
for one. Every browser artifact is mounted inside its own Shadow DOM root the generic loader
creates, so one module's styles cannot reach the Dashboard, another module, or another renderer,
and the host's cannot reach in. That root is a correctness boundary for styles, never a sandbox:
module code remains trusted, same-realm code. The browser mount ABI is versioned
(`MODULE_UI_ABI_VERSION`) and a mismatched artifact fails closed rather than mounting under guessed
semantics.
- Manifest v6 permits independently optional primary (`entrypoints.ui`) and full-viewport display
(`entrypoints.display`) artifacts. Modules own both feature experiences; surfaces own their
containers, routes, navigation/chrome, credentials, error states, and publication wiring. The
fixed authoring sources are `src/ui/index.tsx` and `src/display/index.tsx`; both use the same
role-neutral UI SDK ABI and neither may import a surface or host loader.
- Module UI **asset publication** runs Core → `apps/api`, never the other way
([ADR 0025](decisions/0025-trusted-module-ui-asset-publication.md)). Core owns the policy
(`createModuleUiPublicationRegistry`, built only from already-validated `DiscoveredModule`
records); `apps/api` owns the HTTP route and the one canonical URL construction. Neither
`@ekklesia/module-ui-host` nor `@ekklesia/module-ui-sdk` learns anything about how Ekklesia
publishes artifacts — the loader takes a URL and stays host-neutral — and no module may reach
publication internals. This introduced no new package and no boundary relaxation: `apps/*` could
already depend on Core and on shared packages.
- A publication helper accepts module **identity** only. Neither Core's registry nor the API's URL
builder takes a filename, relative path, or any other artifact selector, so no caller can use one
to reach a second file inside a module package.
- A surface never imports Core or a feature module. It uses public API/client contracts, and reaches
the server over HTTP.
- A surface may depend only on browser-safe shared roles (`api-client`, `auth-client`,
`design-tokens`, `module-ui-host`, `module-ui-sdk`, `surface-sdk`, `ui`) plus dev-time `testing`
and `tooling`. **`@ekklesia/contracts` is deliberately excluded**, despite being a "public
contract" package: its barrel pulls Drizzle's types, and with them Node's global declarations, into
any program that imports it — measured in
[ADR 0024](decisions/0024-browser-safe-ui-sdk-and-host-loader.md), which removed exactly that
import from `@ekklesia/module-ui-sdk` to close the browser boundary. `module-sdk` re-exports the
same barrel and is excluded for the same reason. A surface declares its own types for the wire and
validates what arrives ([ADR 0026](decisions/0026-minimal-ekklesia-module-ui-host.md)).
- Dashboard navigation configuration is **Core-owned platform state**, never module-owned: Core
stores groups and placements and resolves the view, `apps/api` exposes it, and `surfaces/shell`
renders it. A group is presentation configuration and never module structure — it has no manifest,
version, artifact, or lifecycle, and nothing in discovery, dependency resolution, or module startup
may ever read it
([ADR 0027](decisions/0027-dashboard-navigation-and-flat-modules.md)). A module's manifest declares
its own display name and nothing about where it is presented.
- The browser host (`surfaces/shell`) statically imports no module. Every installed module UI it can
mount arrives at run time as a URL from the module UI inventory, is fetched by the generic loader,
and is never bundled into the shell. Its own build (esbuild, owned by that package) is a separate
concern from `@ekklesia/module-ui-build`, which compiles a *module's* declared artifact.
- Dashboard login, logout, and session-transition UX are surface-owned
([ADR 0034](decisions/0034-host-owned-browser-authentication.md)). They use the API over HTTP and
must not add credentials, cookies, tokens, or auth methods to `ModuleUiHostContext`; modules keep
receiving only the host-neutral principal and permission hints already in the UI SDK.
- The display browser application is a separate surface entrypoint under `surfaces/shell`; it also
statically imports no module and mounts only the exact display descriptor/artifact supplied by the
API. It contains no Dashboard chrome and does not move display rendering into the surface—the
module still owns that renderer.
- Separate surfaces do not import each other's internals.
- Builder code composes contracts and surface renderer identifiers; it does not import module or
surface internals.
- Contract and SDK packages cannot depend upward on implementations.
- Workspace package consumers use declared `exports`, not private subpaths.
- Every internal dependency uses a `workspace:` range and every imported workspace package is
declared by the consumer.
- Workspace aliases are forbidden because they obscure package ownership in the dependency graph.
- Workspace package names are unique and shared packages declare a recognized architectural role.
- Circular source and workspace-package dependencies are rejected.
- Cross-module data access uses services, events, capabilities, or opaque identifiers rather than
direct table imports and foreign keys. Modules may not import another module's Drizzle tables,
columns, relations, schema entrypoint, or migration internals.
The design packages are covered the same way: one fixture proves a module may depend on both, one
proves every other shared role stays rejected, and one proves neither design package may depend
upward.
The boundary test suite includes intentionally invalid fixture graphs. This proves that the guard
rejects every currently machine-checkable repository rule even before production packages fill
every directory. The UI guard rejects component, view, template, stylesheet, shared UI-role, and
common UI-framework dependencies inside modules. This deny list is defense in depth; package
review remains required when adopting a new frontend technology. ADR 0022 narrowed the guard
precisely rather than removing it: React was un-denied, every other framework stayed denied, and the
layer rules were untouched.
The Phase 3 schema validator complements source boundaries by checking table ownership, foreign-key
targets, and physical table namespaces in compiled Drizzle metadata. Offline migration validation
checks each module's packaged Drizzle inventory and policy contract, while deliberately making no
claim that opaque SQL ownership can be proven without PostgreSQL-aware analysis.
## Generated code
Generated artifacts are reproducible outputs with a named generator and validation step. A
generated module bootstrap under the root `generated/` boundary may import discovered module entry
points, and apps may import that generated bootstrap. Handwritten app and Core code still cannot
import feature modules directly. Generated files are never an excuse for a hand-maintained central
module registry.