Ekklesia/docs/development/api.md

# API

Phase 4B adds `apps/api`: a minimal Hono-based HTTP transport that mounts Better Auth and exposes
one Ekklesia-owned route to resolve the current principal. It does not add a Dashboard, Website, or
Mobile Surface, and it does not add roles, permissions, or a platform-owner concept — see
[ADR 0011](../architecture/decisions/0011-identity-http-transport-and-public-enrollment.md) for the
full architectural decision and [Identity](identity.md) for the identity foundation this builds on.
A follow-up corrective slice hardened this transport (sanitized Better Auth logging,
production-equivalent origin/CSRF protection under the test runner, bounded identity database
timeouts, idempotent shutdown, and pinned password-length boundaries) without changing any of the
above; see [ADR 0012](../architecture/decisions/0012-identity-origin-csrf-protection-under-test.md).
Phase 4C adds Core's authorization foundation and, on top of this transport, a platform-owner
bootstrap command plus an enforced explicit-grant HTTP API — see [Authorization](authorization.md).
Phase 5D pass 1 hardened this same public perimeter against an independent security audit: a
server-owned client-address resolution boundary for the rate limiter, public sign-up disabled by
default, an explicit request-body-size policy for every public route (including a module route's
own declared contract), and finite Node header/request timeouts — see
[ADR 0017](../architecture/decisions/0017-public-request-perimeter-hardening.md). Pass 3 made the
explicit-grant API's own request contracts exact: canonical UUID grant ids, one canonical
future-only expiration format, unscoped-only creation, a `409` duplicate conflict, and bounded
keyset listing — see
[Authorization](authorization.md#explicit-grant-http-api) and
[ADR 0019](../architecture/decisions/0019-durable-authorization-history-and-grant-management.md).

## What this is, and is not

After this slice it is possible to start the API, sign in, resolve a request to an Ekklesia
principal, and sign out; public account creation is also possible once explicitly enabled (see
[Public registration policy](#public-registration-policy)). It is **not** possible to verify an
email address, recover a forgotten password, use any provider besides email/password, reach a
Dashboard, or hold any role or permission — none of that exists yet. **Public registration is not
ready for a real public launch**: there is no email delivery service, so an unverified email is
trusted as entered and a forgotten password has no recovery path, and no durable, distributed
abuse control exists yet even with signup enabled.

## Starting the API

```bash
pnpm migrations:apply   # the database must already be migrated — the API never does this itself
pnpm api:start
```

`pnpm api:start` follows the repository's build-first convention (`turbo run build`) so it works
from a clean checkout, then runs the compiled `apps/api/dist/server.js`. It reads configuration
from the environment (optionally via an uncommitted `.env`, loaded with
`--env-file-if-exists=.env`) and fails with a concise, sanitized error and a nonzero exit code if
required configuration is missing or invalid — it never applies migrations, creates the identity
tables, or repairs schema, and it fails safely (without mutating the database) if the required
schema is absent.

### Required configuration

No real values are shown here — configure per environment and never commit them:

| Variable | Requirement |
| --- | --- |
| `DATABASE_URL` | An absolute `postgres://` or `postgresql://` URL to an already-migrated database. It must **not** carry a TLS parameter (`sslmode`, `ssl`, `sslrootcert`, …) — see [Database transport (TLS)](#database-transport-tls). |
| `EKKLESIA_DATABASE_TLS_MODE` | Optional. Exactly `"verify-full"` or `"disable"`. Defaults to `verify-full` in production and `disable` in development/test. Production refuses `disable`. |
| `EKKLESIA_DATABASE_CA` | Optional. PEM text of a private certificate authority to trust in addition to the system store, for a database using an internal CA. |
| `EKKLESIA_API_HOST` | Optional. Listen address. Defaults to loopback (`127.0.0.1`). An explicit non-loopback address is permitted for future deployment. |
| `EKKLESIA_API_PORT` | Optional. Listen port, `0`–`65535` (`0` requests an OS-assigned port). Defaults to `3000`. |
| `EKKLESIA_IDENTITY_SECRET` | At least 32 characters, high entropy (for example `openssl rand -base64 32`). Same requirement as Phase 4A. |
| `EKKLESIA_IDENTITY_BASE_URL` | Exactly one canonical `http`/`https` origin — scheme, host, and non-default port only, with no path, trailing slash, query, fragment, credentials, or explicit default port. `https` is required in production; loopback and unspecified hosts are rejected in production. |
| `EKKLESIA_IDENTITY_TRUSTED_ORIGINS` | Comma-separated list of canonical origins, validated identically to the base URL (no wildcards). Used for both Better Auth's own CSRF/origin checks and the API's CORS policy — there is only one trusted-origin list. |
| `EKKLESIA_PUBLIC_SIGNUP_ENABLED` | Optional, **development/test only**. Exactly `"true"` enables public sign-up when `NODE_ENV=development`; defaults to disabled, and a production configuration rejects `"true"` outright. Omitted, misspelled, or any other value never enables it. See [Public registration policy](#public-registration-policy). |
| `EKKLESIA_CLIENT_ADDRESS_MODE` | Optional. `"direct"` (default when unset) or `"trusted-proxy"`. See [Client-address resolution](#client-address-resolution). |
| `EKKLESIA_CLIENT_ADDRESS_HEADER` | Required only when `EKKLESIA_CLIENT_ADDRESS_MODE=trusted-proxy`. The one forwarded-header name the reverse proxy is trusted to set (for example `x-forwarded-for`). |
| `EKKLESIA_CLIENT_ADDRESS_TRUSTED_PROXIES` | Required only when `EKKLESIA_CLIENT_ADDRESS_MODE=trusted-proxy`. Comma-separated list of the proxy's own IPv4/IPv6 addresses or CIDR ranges — never a broad private range that also covers real clients. |
| `NODE_ENV` | `development` explicitly opts into relaxed (loopback/`http`) origin validation for local development. Anything else, including unset, validates as production. |

## Routes

| Route | Owner | Notes |
| --- | --- | --- |
| `POST\|GET /api/auth/*` | Better Auth (mounted directly) | Better Auth's own request/response shapes pass through unmodified — see [Better Auth's email/password docs](https://www.better-auth.com/docs/authentication/email-password) for the exact endpoints (`/sign-up/email`, `/sign-in/email`, `/sign-out`, `/get-session`, ...). |
| `GET /api/v1/principal` | Ekklesia | Returns the minimal `EkklesiaPrincipal` contract as JSON. `Cache-Control: no-store`. |
| `GET\|POST /api/v1/grants*` | Ekklesia | Explicit authorization grant management, gated by `core.authorization.grants.manage`. Listing is bounded keyset pagination; creation is unscoped-only and conflicts (`409`) on an equivalent effective grant; revocation clears every unrevoked equivalent. See [Authorization](authorization.md#explicit-grant-http-api). |
| `/api/v1/modules/<module-id>/*` | Installed modules | Every route an installed module registered during its `register` hook, gated by that route's own declared permission. See [Module HTTP routes](#module-http-routes). |
| `GET /api/v1/module-realtime/<module-id>/<channel>` | Installed modules | Snapshot-first Server-Sent Events for one registered module-local channel, gated by that channel's declared permission. See [Module realtime](#module-realtime). |
| `GET /api/v1/module-ui` | Ekklesia | The browser host's module UI inventory: which module UIs may be mounted, where to fetch each, where each module's HTTP and realtime roots live, and the viewer's granted permissions for it. Requires `core.modules.ui.read` (baseline `authenticated`). See [Module UI inventory](#module-ui-inventory). |
| `GET /api/v1/dashboard/navigation` | Ekklesia | The resolved Dashboard navigation: the same mountable modules, grouped, labelled, and ordered as this installation has configured them. Requires `core.modules.ui.read`. See [Dashboard navigation](#dashboard-navigation). |
| `PUT\|DELETE /api/v1/dashboard/navigation/{groups,placements}/*` | Ekklesia | Organize the Dashboard: create/rename/order/delete a group, and place a module. Requires `core.dashboard.navigation.manage` (baseline `restricted`) and a trusted `Origin`. See [Dashboard navigation](#dashboard-navigation). |
| `GET /api/v1/module-ui/<module-id>/<version>` | Ekklesia | Publishes an installed module's validated `entrypoints.ui` artifact at a browser-loadable URL. No authentication; only the declared artifact, never another file. See [Module UI asset publication](#module-ui-asset-publication). |
| `GET /api/v1/module-lifecycle` | Ekklesia | Which installed modules are enabled, what this process actually did with each, and whether a restart is pending. Requires `core.modules.lifecycle.manage` (baseline `restricted`). See [Module lifecycle](#module-lifecycle). |
| `PUT /api/v1/module-lifecycle/<module-id>/enabled` | Ekklesia | Enable or disable one installed module. Takes effect at the next API start. Requires `core.modules.lifecycle.manage` and a trusted `Origin`. See [Module lifecycle](#module-lifecycle). |
| `GET /api/v1/module-configuration/<module-id>` | Ekklesia | One module's declared configuration schema, its stored values, and whether a restart is pending. Never returns a secret value. Requires `core.modules.configuration.manage` (baseline `restricted`). See [Module configuration](#module-configuration). |
| `PUT /api/v1/module-configuration/<module-id>` | Ekklesia | Replace one module's stored configuration. Takes effect at the next API start. Requires `core.modules.configuration.manage` and a trusted `Origin`. See [Module configuration](#module-configuration). |

Better Auth's handler is unreachable outside `/api/auth/*`; every other unmatched path returns a
deterministic `404 {"code":"NOT_FOUND"}`.

### `GET /api/v1/principal` response shapes

Anonymous (no session, or an expired/revoked/malformed session):

```json
{ "kind": "anonymous" }
```

Authenticated:

```json
{ "kind": "authenticated", "userId": "2aKp4jPeV1QbXwyz1OH6IcAUfl2n1fc1" }
```

Infrastructure failure (the identity session store was not usable while resolving the principal —
never silently reported as anonymous):

```json
{ "code": "PRINCIPAL_RESOLUTION_FAILED" }
```
returned with HTTP `503`.

A request carrying no cookies, or only unrelated cookies, never reaches that path and never performs
identity database work at all: only a request presenting the *configured session cookie* that Better
Auth resolved to `null` is ambiguous enough to check. When it is checked, a missing session table, a
dropped column, a revoked `SELECT` privilege, an unreachable database, and a query timeout all
produce this same sanitized `503` — carrying no relation name, column name, SQL, driver message,
host, or stack. See [Identity § How a principal is resolved](identity.md#how-a-principal-is-resolved).

This is deliberately not the same thing as Better Auth's own `GET /api/auth/get-session` response,
which returns Better Auth's full `{session, user}` shape (session id, expiry, IP address, user
agent, email, name, image, email-verification status, and more). `/api/v1/principal` is the
Ekklesia-owned contract every other Core/Surface consumer should use; it never contains email,
name, image, session id or expiry, session token, account/provider details, or any role,
permission, membership, or volunteer data.

## Public registration policy

**Public sign-up is disabled by default** (`EKKLESIA_PUBLIC_SIGNUP_ENABLED` unset or anything other
than exactly `"true"`). This is a strict, typed gate (`IdentityConfig.publicSignupEnabled`,
`core/src/identity/config.ts`) wired directly to Better Auth's own `emailAndPassword.disableSignUp`
flag (`core/src/identity/runtime.ts`) — there is no second, platform-owned signup gate, because the
pinned release's own mechanism already has every required property: every `POST
/api/auth/sign-up/email` call is rejected with a fixed, sanitized `400
{"code":"EMAIL_PASSWORD_SIGN_UP_DISABLED"}` and `Cache-Control: no-store` *before* Better Auth ever
looks up the email, so a disabled deployment never reveals whether an address is already
registered. Sign-in, session resolution, and sign-out are entirely unaffected by this gate.

**Production cannot enable public signup at all right now.** A production configuration that sets
the flag to `true` is rejected outright with the stable
`PUBLIC_SIGNUP_UNSUPPORTED_IN_PRODUCTION` configuration error, and `pnpm doctor` reports that state
as a blocking `NOT READY` failure rather than a warning. This is not a preference: the durable
signup-abuse prerequisites a public registration endpoint needs — verified email or an invitation
requirement, and durable shared/edge throttling rather than this release's per-process in-memory
limiter — **do not exist in this repository yet**, so there is no production configuration in which
the flag could be honored safely. The rejection happens in `createIdentityConfig`, which
`apps/api`'s executable entry point calls *before* `startApi`, so such a deployment never opens a
PostgreSQL pool, constructs Better Auth, starts the module runtime, or binds a listener.

Development and test composition may still enable it explicitly, through the same typed path
(`EKKLESIA_PUBLIC_SIGNUP_ENABLED=true` with `NODE_ENV=development`, or
`{ environment: "development" }` programmatically) — that is how this repository's own integration
suite creates accounts over HTTP. When the prerequisites above are actually implemented and proven,
`validatePublicSignupEnabled` (`core/src/identity/config.ts`) is the single place that changes. See
[ADR 0017](../architecture/decisions/0017-public-request-perimeter-hardening.md).

Once enabled, the rest of this section's behavior is unchanged from ADR 0011. A
registered account is an **authentication identity** — proof that someone can present this email
and password — not a church-membership record; it does not imply attendance or involvement, and
nothing about it is recorded beyond Better Auth's own required `user`/`account` rows. Every
account, first or five-hundredth, receives identical baseline authenticated status: no Dashboard
access, no module-management access, no staff/volunteer-management access, and no
media/event/form/roster/people-management permission — Core has no authorization schema in this
slice to grant any of that, and no code path in this repository inspects how many accounts already
exist or treats an empty database as special. Baseline authenticated status is what will eventually
let a signed-in person use ownership-scoped personal features (watch history, continue watching,
saved content, preferences) — none of those features are implemented in this PR.

Explicit **platform-owner bootstrap** is now available as a local operator command — see
[Authorization](authorization.md#platform-owner-bootstrap) — and grants an *existing* identity
platform-owner authority; signup order is never part of that decision. Future
**relationship-derived access** (for example, a person added to a volunteer roster automatically
gaining narrow, reversible read access to that roster) belongs to a later roster slice, not this
one.

`emailAndPassword.autoSignIn` is set to `false`. Signing up creates the account but does **not**
automatically create a session — no `Set-Cookie` is sent, and the sign-up response's `token` field
is `null`. The caller must call `/api/auth/sign-in/email` explicitly afterward. This is deliberate:
per the pinned Better Auth release, `autoSignIn: false` activates its built-in email-enumeration
protection, so a duplicate sign-up returns the same `200` shape (with a fresh, unrelated random id)
whether or not the email was already registered, rather than revealing which. No custom
cryptography replaces Better Auth's own password hashing. Password length is pinned explicitly to
8–128 characters (`minPasswordLength`/`maxPasswordLength` in `core/src/identity/runtime.ts`) rather
than left on Better Auth's fallback default — the pinned release's default happens to be the same
8–128 range, but an implicit default can change silently on a future Better Auth upgrade, while an
explicit value cannot. `apps/api/tests/api-http.integration.ts` proves the exact boundary against a
real server: 7 characters rejected, 8 accepted, 128 accepted, 129 rejected.

## Origin, CORS, and cookies

The API reuses the same `EKKLESIA_IDENTITY_TRUSTED_ORIGINS` list for both CORS and Better Auth's
own origin/CSRF checks — there is no second, competing configuration. CORS reflects a request's
`Origin` header back only when it exactly matches a configured trusted origin (never a wildcard,
since `Access-Control-Allow-Origin: *` cannot be combined with `Access-Control-Allow-Credentials:
true`); a disallowed origin never receives the header at all, though the request itself may still
be processed — CORS is a browser-enforced, response-header mechanism, not a server-side gate. A
request with no `Origin` header at all (native or server-to-server clients) is unaffected by CORS.
Note this is narrower than it sounds for `/api/auth/sign-in/email` and `/api/auth/sign-up/email`
specifically: any client built on the Fetch API — every browser, and also Node's own `fetch()` —
automatically sends a `Sec-Fetch-Mode: cors` header on every request, and Better Auth's first-login
CSRF protection treats that as "this came from a Fetch client" and requires a trusted `Origin` (or
`Referer`) on those two routes regardless of whether a cookie is present. A genuinely non-Fetch
caller (curl, or most backend HTTP client libraries, which send none of the `Sec-Fetch-*` headers)
is unaffected and needs no `Origin` header at all — see the dedicated raw-socket test in
`apps/api/tests/api-http.integration.ts` for the exact distinction proven against a real server.

Better Auth's own CSRF/origin protections are explicitly kept on:
`core/src/identity/runtime.ts` sets `advanced: { disableCSRFCheck: false, disableOriginCheck:
false }`. This is deliberately explicit rather than left unset — the pinned release's own default
is environment-sensitive (`skipOriginCheck` falls back to `true`, silently disabling both origin
and CSRF checking, whenever `isTest()` is true, which Vitest's default `NODE_ENV=test` triggers),
so an unset config would mean origin/CSRF protection is silently off under the test runner and on
in production. Setting both flags explicitly removes that environment-sensitive branch entirely,
so `pnpm test:migrations:integration` now automatically proves the exact production behavior: any
non-`GET`/`HEAD`/`OPTIONS` request carrying a session cookie must also carry a trusted `Origin` (or
`Referer`), or Better Auth rejects it with `403` (and the presented session remains valid
afterward); a no-cookie first authentication attempt from an untrusted origin is rejected the same
way; a trusted origin is accepted in both cases; and a denied origin never receives a permissive
CORS header. See [ADR 0012](../architecture/decisions/0012-identity-origin-csrf-protection-under-test.md)
for the root-cause investigation and the full account of what changed.

Cookies keep Better Auth's defaults: `HttpOnly` always, `SameSite=Lax` always, `Secure` when the
configured base URL is `https://`. Cross-subdomain cookies are not enabled and no cookie `Domain`
attribute is set, so every cookie is host-only. See [Client-address resolution](#client-address-resolution)
for how (and whether) a reverse proxy's forwarded headers are trusted — a separate concern from
CORS/CSRF above.

## Logging

Two independent, complementary layers, both configured explicitly in `core/src/identity/runtime.ts`
rather than left on any environment-sensitive default:

1. **`logger`.** Better Auth's own default (`level: "warn"`, plain `console.*` output) is replaced.
   The pinned release's log call sites are not a stable, fully-audited contract — several pass a
   raw internal exception, and at least one (the origin-check middleware) interpolates a
   user-controlled `Origin` header directly into the message text. Rather than sanitize each call
   site's message and arguments individually, Core's logger configuration drops the message and
   every argument entirely and emits only a fixed, generic string
   (`"[identity] Better Auth reported an error."`) via `console.error`, and only for error-level
   events — nothing is ever emitted for warn/info/debug.
2. **`onAPIError: { throw: true }`.** Discovered while testing (1) above: the pinned release's own
   HTTP transport dependency, `better-call`'s router, has an *unconditional*
   `console.error("# SERVER_ERROR: ", error)` fallback for any thrown value it does not recognize
   as one of Better Auth's own `APIError`s — this fires before Better Auth's `logger` option is
   ever consulted, so no `logger` configuration can reach it. `onAPIError.throw: true` closes that
   gap: Better Auth's own deliberate `APIError`s (wrong password, invalid origin, disabled sign-up,
   password out of bounds, ...) still resolve to their normal, correct HTTP response completely
   unaffected, but a genuinely unexpected failure (for example a database outage mid-request) now
   propagates out of `auth.handler()` instead of being caught and printed raw internally by
   `better-call`, reaching `handleAuthRequest`'s own try/catch and becoming a sanitized
   `IdentityRuntimeError` like any other unexpected failure — and nothing is logged at all.

Both apply in every environment, including under Vitest: neither is a `NODE_ENV=test` special case.
A focused unit test (`core/tests/identity-runtime.test.ts`) simulates a hostile log call (a raw
exception carrying a password, a host, a port, a filesystem path, and a rejected origin) and asserts
none of it reaches `console.error`, and separately asserts `onAPIError` is configured; a live
integration test forces a real identity-database failure through Better Auth's own sign-in handler
and asserts nothing at all reaches `console.error`/`warn`/`log` in the real process.

## Client-address resolution

Better Auth's rate limiter (below) needs a trustworthy per-client identity, and the pinned
release's own client-address resolution only ever reads request headers — it cannot see the real
TCP socket peer. `apps/api` resolves and sanitizes that identity once, at the Node/API transport
boundary (`apps/api/src/client-address.ts`, wrapping the `fetch` callback passed to
`@hono/node-server`'s `serve()` in `apps/api/src/server.ts`), *before* Better Auth or any other
route ever sees the request. The resolution policy itself (`core/src/identity/client-address.ts`)
is framework-agnostic and lives in Core; only the Node-specific glue that reads
`env.incoming.socket.remoteAddress` lives in `apps/api`.

- **Direct mode** (the default, `EKKLESIA_CLIENT_ADDRESS_MODE` unset or `"direct"`). Trusts only
  the actual immediate socket peer. No `X-Forwarded-For`, `Forwarded`, `CF-Connecting-IP`,
  `X-Real-IP`, or any other request header is ever consulted for identity.
- **Trusted-proxy mode** (`EKKLESIA_CLIENT_ADDRESS_MODE=trusted-proxy`, with
  `EKKLESIA_CLIENT_ADDRESS_HEADER` and `EKKLESIA_CLIENT_ADDRESS_TRUSTED_PROXIES` both required).
  A forwarded header's content is trusted **only** when the actual immediate socket peer matches
  one of the configured trusted-proxy addresses/CIDR ranges; otherwise the peer address itself is
  used, exactly as in direct mode. A comma-separated forwarded chain is walked **right to left**:
  trailing hops matching a trusted proxy are skipped, and the first hop from the right that does
  not match is the resolved client address. A missing, empty, malformed, or entirely-trusted-hops
  chain from an otherwise-trusted peer fails closed with a sanitized
  `503 {"code":"CLIENT_ADDRESS_UNRESOLVED"}` — Better Auth, Core, and every module handler are all
  unreachable in that case. IPv4, IPv6, and IPv4-mapped IPv6 (`::ffff:a.b.c.d`, the form a
  dual-stack socket or a real proxy actually emits) all normalize consistently before comparison.

Whichever address is resolved is written into one fixed internal header
(`TRUSTED_CLIENT_ADDRESS_HEADER = "x-ekklesia-client-address"`, `core/src/identity/runtime.ts`),
**overwriting** — never appending to — any externally supplied copy of that exact header name, so
a caller can never choose its own "resolved" identity by sending it directly. Better Auth is
configured to trust only this one header (`advanced.ipAddress.ipAddressHeaders`), never its own
default (`x-forwarded-for`) and never `trustedProxies` (that pinned-release option only inspects
header *content*, never who actually connected — see [ADR 0017](../architecture/decisions/0017-public-request-perimeter-hardening.md)
for why that is insufficient on its own).

**A reverse proxy in front of this API must be configured to *overwrite*, not append to, the
configured forwarded header**, and the origin must not be reachable except through that proxy —
neither is something application-level validation can enforce; both are the deployment's own
network/firewall responsibility. `pnpm doctor`'s `identity.client-address` check distinguishes
three states: safe direct mode, an explicitly configured (and structurally valid) trusted-proxy
mode, and an invalid/contradictory configuration (which fails Doctor, not merely warns).

## Rate limiting

Better Auth's built-in rate limiting is always enabled, including by the executable entry point —
it is never disabled. Its built-in per-path rules (a strict default of 3 requests per 10 seconds on
sign-in/sign-up-prefixed paths) are preserved unmodified in production composition; only the
integration test suite injects a narrow, explicit override so it is not itself rate-limited by its
own many requests. Storage remains the default in-memory backend, which is **per-process**: no
protection across multiple API instances or restarts. Client identity for this limiter is now
always the resolved, server-owned address described above — never a caller-controlled header — so
rotating a spoofed `X-Forwarded-For` on every request no longer lets a caller choose a fresh
bucket, and unrelated real peers are never collapsed into one shared bucket merely because no
header was present. This closes the rate-limit-identity gap; it does not add distributed storage —
a genuinely multi-instance deployment still needs a shared or edge limiter, reserved for pass 5.

## Request body limits

Every public body-consuming route has an explicit, platform-enforced byte ceiling
(`apps/api/src/body-limits.ts`, built on Hono's pinned `hono/body-limit` middleware — verified
directly against its source to reject both an oversized declared `Content-Length` and an oversized
declared-length-free streamed body without ever buffering the whole payload first):

| Surface | Limit |
| --- | --- |
| `/api/auth/*` (every Better Auth route) | 8 KiB |
| `POST /api/v1/grants`, `POST /api/v1/grants/:grantId/revoke` | 16 KiB |
| A module route (`http.route(...)`'s own `maxBodyBytes`) | Up to 1 MiB, the platform ceiling (`MODULE_ROUTE_MAX_BODY_BYTES_CEILING`, `@ekklesia/contracts`) |

An oversized request always receives the identical stable, sanitized contract —
`413 {"code":"REQUEST_BODY_TOO_LARGE"}` with `Cache-Control: no-store` — and never invokes Better
Auth's business logic, a Core mutation, permission evaluation (where avoidable), a module handler,
or module database behavior. For a state-changing route the body limit is enforced after
`requireTrustedOrigin` but before permission evaluation, so a rejected cross-origin mutation's body
is never read for size at all, while an oversized same-origin request still fails before any
authorization work runs. See [Module HTTP routes](#module-http-routes) for the module-route body
contract itself, and [Module development](../module-development/README.md#http-routes) for the
module-authoring side of it. Large media transfer is explicitly out of scope for ordinary module
HTTP bodies — that belongs to a future dedicated storage design, never to widening this ceiling.

## Node request and header timeouts

`apps/api/src/server.ts` configures the underlying Node HTTP server (via `@hono/node-server`'s
`serverOptions`, passed straight through to `http.createServer`) with finite
`headersTimeout` (20s), `requestTimeout` (30s), and `keepAliveTimeout` (5s, Node's own default,
made explicit). Node's own documented behavior responds `408` and closes the connection without
ever invoking a request listener when either timeout is exceeded, so a timed-out connection cannot
invoke a handler or receive a second response by construction.

Two non-obvious details, both found only by reproducing them directly against the pinned Node 24
runtime through a real socket:

1. **Detection latency.** Node's `http.Server` only *checks* for a timeout violation when its
   periodic `connectionsCheckingInterval` fires, and that option's own default is 30 seconds —
   leaving it unset would silently let a violating connection sit for up to a further 30 seconds
   past its configured timeout. `connectionsCheckingIntervalMs` is therefore also set explicitly
   (5s).
2. **The observable event is `'clientError'`, not `'timeout'`.** Node's general server `'timeout'`
   event belongs to `server.setTimeout()`'s idle-socket timer, which this server never arms; it
   does **not** fire for `headersTimeout`/`requestTimeout`. Those surface as `'clientError'` with
   `err.code === "ERR_HTTP_REQUEST_TIMEOUT"`. `apps/api/src/server.ts` therefore observes
   `'clientError'`, and — because registering any `'clientError'` listener *suppresses* Node's own
   default response entirely (`_http_server.js` only falls back when the emit returns `false`) —
   that handler reproduces Node's default byte-for-byte rather than merely watching: `408 Request
   Timeout` for a timeout, `431 Request Header Fields Too Large` for a header overflow, and `400
   Bad Request` otherwise, each followed by destroying the socket, and each written only when the
   socket is still writable and no response has already begun (so a handler that already responded
   before an unread body timed out never receives a second, corrupting status line).

Exactly one fixed, sanitized operational line is logged per transport-level client failure — one
message for a timeout, one for any other malformed/oversized request. Neither ever contains the
request path, query, body, cookies, authorization value, client-address header, forwarding chain,
raw peer address, stack, or the caught error's own message (a parse failure's message is derived
from attacker-controlled input). This is proven at the real transport boundary by raw-socket
scenarios in `apps/api/tests/api-http.integration.ts` — a real slow-header connection, a real
slow-body connection, a malformed request line, and an oversized header block — each asserting the
exact response bytes, exactly one status line, socket closure, no handler invocation, and exactly
one sanitized log. See
[ADR 0017](../architecture/decisions/0017-public-request-perimeter-hardening.md) for the full
account.

## Module discovery, permission composition, and module runtime startup

`startApi()` composes production in one fixed order:

1. Discover every installed module's manifest exactly once and compose the one immutable platform
   permission registry (Core's own declarations plus every module's) — `prepareStartupModuleInventory`,
   `apps/api/src/platform-permissions.ts`.
2. Construct the one shared PostgreSQL pool.
3. Construct the module database facade and the frozen `ModuleRuntimeServices` object from that same
   pool (see [Module database access](#module-database-access)).
4. Construct the Core identity runtime, authorization store, and evaluator.
5. Construct the host object-storage provider from validated configuration, if this installation has
   one — building the client opens no connection, so a storage outage can never stop the platform
   from booting (ADR 0030). Then read the stored module enablement configuration and the stored module
   configuration, and resolve, over that exact same discovered inventory, which modules are
   **eligible** to run and what each eligible module's frozen configuration snapshot is (`resolveModuleEligibility`,
   `@ekklesia/core/modules` — see [Module lifecycle](#module-lifecycle) and
   [Module configuration](#module-configuration)). The two reads are sequential rather than raced so
   a broken deployment reports the same error code every time.
6. Start the module runtime from the eligible subset (`startDiscoveredModuleRuntime`) — modules are
   never rediscovered for this step — passing it the services object every module's `start`/`stop`
   hook receives.
7. Create the Hono application, mounting the started module runtime's HTTP routes and realtime
   channels, publishing UI for that same eligible set, and passing the identical scoped runtime
   objects to module HTTP handlers.
8. Bind the HTTP server.

Discovery and permission composition happen before any pool, Better Auth instance, module
entrypoint import, or HTTP server exists, so a discovery or composition failure (an invalid
manifest, a module claiming the reserved `core.*` permission namespace, a duplicate permission
declaration, ...) fails startup outright, before there is anything to close; startup never falls
back to a Core-only registry. Core infrastructure exists before any executable module lifecycle
code runs, and every module route is published only after the *entire* module runtime has
registered and started successfully, so no request can ever reach a module before every module has
started. See [Authorization](authorization.md#api-startup-discovery-composition-and-enforcement)
for the permission-composition detail, and [Module development](module-development/README.md) for
the manifest shape that produces these declarations.

Unlike discovery/composition/Doctor/`modules:validate` (which only ever read manifests), a
successful production startup does import each **eligible** module's compiled server entrypoint and
run its `register`/`start` hooks — this is the one place in the platform that deliberately executes
module code, and only after every other startup step already listed has succeeded. A module that is
disabled, or blocked by a disabled required dependency, is never imported at all.

Step 5 makes startup depend on a reachable, migrated database: if the enablement configuration
cannot be read, `startApi` **refuses to start** with Core's sanitized `MODULE_LIFECYCLE_READ_FAILED`
(and `MODULE_CONFIGURATION_READ_FAILED` if the enablement read succeeded and the configuration read
did not) rather than starting every discovered module. The platform cannot know what an administrator
decided, and starting a module they disabled is worse than not starting. Nothing is created or
repaired on that path.

## Module lifecycle

```text
GET /api/v1/module-lifecycle
PUT /api/v1/module-lifecycle/<module-id>/enabled     { "enabled": boolean }
```

The platform's operational view of its installed modules
([ADR 0028](../architecture/decisions/0028-module-operational-lifecycle.md)): which are enabled, what
this process actually did with each of them, and whether the two currently disagree.

It is a **sibling** of `/api/v1/modules/`, never a child, for the same reason `/api/v1/module-ui` is
(ADR 0025): `enabled` is a legal module route segment, so a module named `announcements` could
otherwise register `PUT /enabled` and collide with the platform at exactly
`/api/v1/modules/announcements/enabled`.

```json
{
  "modules": [
    {
      "id": "storage",
      "version": "1.0.0",
      "configured": { "enabled": false, "explicit": true },
      "runtime": { "state": "disabled" },
      "restartRequired": false
    },
    {
      "id": "announcements",
      "version": "1.0.0",
      "configured": { "enabled": true, "explicit": false },
      "runtime": {
        "state": "blocked",
        "reason": "required_dependency_disabled",
        "dependency": { "kind": "module", "moduleId": "storage" }
      },
      "restartRequired": false
    }
  ]
}
```

### Two axes, deliberately not one status word

`configured` is the decision that survives restarts. `explicit: false` means nobody has decided and
the platform default (**enabled**) applies. `runtime` is what *this process* did:

| state      | meaning                                                                       |
| ---------- | ----------------------------------------------------------------------------- |
| `running`  | the module was eligible and its `start` hook completed in this process         |
| `disabled` | configuration says it must not run, so it was never imported                   |
| `blocked`  | it should run, but a **required** dependency cannot — see `reason`/`dependency` |

`running` says the `start` hook completed. It says **nothing** about whether every external resource
that module uses is reachable right now; Ekklesia has no runtime health-probe contract, and this
state must not be read as one. There is deliberately no `failed` state: module startup is
all-or-nothing, so a module that fails to start aborts API startup and no live process could ever
report it.

`reason` is one of `required_dependency_disabled` (the named dependency was turned off) or
`required_dependency_blocked` (the named dependency is itself blocked). Following the chain always
terminates at a `..._disabled` entry. An **optional** dependency never blocks anything.

### Changing enablement requires a restart, and the response says so

There is no hot module installation or removal, so `PUT` writes a decision and changes nothing about
the running process:

```json
{ "id": "announcements", "version": "1.0.0",
  "configured": { "enabled": false, "explicit": true },
  "runtime": { "state": "running" },
  "restartRequired": true }
```

Reporting `runtime: disabled` there would be a lie about a process still serving that module's
routes. `restartRequired` compares what the current configuration *would* start against what the
runtime reports it *did* start, so it is symmetric and transitive: disabling a module also marks its
dependents, because they really will stop on the next boot.

A module whose enablement changed but whose process has not restarted stays fully exposed — routes,
artifact, inventory, and navigation. It really is still mounted; hiding it now would break a viewer
mid-session for a change that has not taken effect.

### What a disabled module does not have

Never imported ⇒ no `register`, `start`, or `stop` hook runs; no HTTP routes are mounted (an
unmounted path gets the platform's ordinary `404`, with no per-request enablement check anywhere);
no capabilities are registered; `GET /api/v1/module-ui/<id>/<version>` returns the same `404` an
uninstalled module does; and it is absent from both `GET /api/v1/module-ui` and
`GET /api/v1/dashboard/navigation`.

Its **migrations are still discovered and applied**. Deployed schema ownership is not runtime
enabled state: a disabled module's tables must already exist when it is enabled, or enabling it would
require a schema change in the middle of a restart. Its **permissions also stay registered**, so
grants survive an enable/disable cycle and can be issued ahead of enabling a module; nothing enforces
them, because it mounts no routes.

### Authority

Both verbs require `core.modules.lifecycle.manage` (baseline `restricted`). The read is gated too,
because this resource is the only place a disabled module, a blocked module, and the
required-dependency edges between them are visible at all — `core.modules.ui.read` deliberately
covers only what is mountable. Mutations additionally require an exact trusted `Origin` and a bounded
body (`CORE_MUTATION_MAX_BODY_BYTES`) before authorization work begins.

The body is closed: any key beyond `enabled` invalidates the whole request rather than being dropped.
A `PUT` naming a module this process did not discover is `404 {"code":"MODULE_NOT_INSTALLED"}` —
enablement is operational configuration, not an unbounded key-value store. Existing records for
modules that are no longer deployed are retained and ignored, so redeploying restores the decision.

No response carries a filesystem path, package name, manifest path, entrypoint, caught error, or
stack.

### Recovering from a module that breaks startup

Startup is all-or-nothing, so a module whose `start` hook throws takes the API down — including this
API. The out-of-band path is a Core CLI, in the same shape as `pnpm authorization:bootstrap-owner`
(explicit, never HTTP-exposed):

```bash
pnpm modules:disable -- --module announcements
pnpm modules:enable  -- --module announcements
```

It reads `DATABASE_URL`, runs the same manifest-only discovery the API does (so it refuses a module
this deployment cannot see), writes the decision, and starts nothing. Restart the API to apply it.

## Module configuration

```text
GET /api/v1/module-configuration/<module-id>
PUT /api/v1/module-configuration/<module-id>     { "values": { "slideSeconds": 12 } }
```

What one installed module declares it needs configured, what an operator has stored, and whether the
running process is still using it
([ADR 0029](../architecture/decisions/0029-module-configuration-and-secrets.md)).

A **sibling** of `/api/v1/modules/`, never a child, for the same reason `/api/v1/module-ui` and
`/api/v1/module-lifecycle` are: `config` and `configuration` are legal module route segments, so a
module could otherwise collide with the platform at `/api/v1/modules/<id>/config`.

```json
{
  "moduleId": "announcements",
  "version": "1.0.0",
  "keys": [
    { "key": "slideSeconds", "type": "integer", "required": false, "default": 7,
      "description": "How long each slide is shown.", "configured": false },
    { "key": "bucket", "type": "string", "required": true, "value": "slides", "configured": true },
    { "key": "storageCredential", "type": "secret", "required": true, "configured": true }
  ],
  "unknownKeys": [],
  "restartRequired": false
}
```

### A secret's value is absent, not redacted

Note what the `storageCredential` entry above does **not** have: a `value` field. Not `null`, not
`"***"` — absent. Ekklesia never persists secret material, so the store this route reads from
contains none, and there is no code path that could serve one. `configured` answers the only question
an operator needs from here: does the deployment supply it?

### The declared types

| type | notes |
| --- | --- |
| `string` | optional `minLength` / `maxLength` |
| `integer` | optional `minimum` / `maximum`; submitted as a JSON number, never a string |
| `boolean` | submitted as a JSON boolean |
| `enum` | closed `values` list |
| `url` | absolute `http:`/`https:` only |
| `secret` | deployment-supplied; never stored, never returned, never defaulted |

No nesting, no arrays, no regular expressions. This is a validation vocabulary, not a form-builder
DSL; a module needing structured data owns a table for it.

### Writing is a whole-document replace, with closed validation

A key the module does not declare is `400 CONFIGURATION_KEY_UNDECLARED`. A wrong JSON type or a value
failing the module's own declared constraints is `400 CONFIGURATION_VALUE_INVALID`. A `secret` key is
`400 CONFIGURATION_SECRET_NOT_WRITABLE` — there is no supported way to submit secret material to this
platform. Nothing is coerced: `{"slideSeconds": "7"}` is a `400`, not a silent `7`.

Constraint checking runs through **the same resolver startup uses**, over the exact document about to
be stored, so the API and startup can never disagree about whether a value is valid.

An *incomplete* document is accepted deliberately — an operator may save partial work, and the module
is reported `blocked · configuration_missing` until they finish. An *invalid* one never is.

A key omitted from the submission is deleted. Stored keys the module's current manifest no longer
declares are **retained and ignored**, and listed as `unknownKeys`, so a downgrade or a rename does
not destroy an operator's value.

### Changes take effect at the next start

Modules receive a configuration snapshot resolved once at startup, so a `PUT` changes nothing about
the running process and the response reports `restartRequired` rather than pretending otherwise.
Nothing is hot-reloaded. A setting that genuinely must change live is the module's own domain state,
stored in the module's own tables and exposed through its own route — not platform configuration.

### Missing or invalid configuration blocks the module

An enabled module whose required configuration cannot be satisfied is never imported. It appears in
`GET /api/v1/module-lifecycle` as `blocked` with a structured reason naming the offending **key**:

```json
{ "runtime": { "state": "blocked", "reason": "configuration_missing", "configurationKey": "bucket" } }
```

`configuration_missing` (set a value), `configuration_invalid` (correct the value), and
`secret_missing` (set it in the deployment environment) are separate reasons because each has a
different fix. A blocked module blocks its required dependents exactly like a disabled one does.

### Secrets come from the deployment

A module declares a `secret` key; the host maps it to a deployment-supplied value. In an Ekklesia
deployment that mapping is an environment variable:

```text
EKKLESIA_MODULE_SECRET__<MODULE_ID>__<KEY>
EKKLESIA_MODULE_SECRET__ANNOUNCEMENTS__STORAGE_CREDENTIAL
```

The module never sees that name — it calls `config.getSecret("storageCredential")` — which is what
lets a future standalone host supply the same declared secret from somewhere else entirely.

### Authority

Both verbs require `core.modules.configuration.manage` (baseline `restricted`), deliberately separate
from `core.modules.lifecycle.manage`: deciding whether a module runs and deciding how it behaves are
different authorities. Mutations additionally require an exact trusted `Origin` and a bounded body.

No response carries a filesystem path, package name, manifest path, caught error, stack, or secret
value.

## Module object storage

Ekklesia can provide modules with logical object storage backed by any S3-compatible service
([ADR 0030](../architecture/decisions/0030-host-provided-module-object-storage.md)). There is **no
HTTP API for it**: storage is a runtime capability modules use, not a resource the platform exposes,
and no route returns a provider endpoint or credential.

### Host configuration

| variable | meaning |
| --- | --- |
| `S3_ENDPOINT` | Absolute `http`/`https` URL. **Absent means this installation has no object storage.** |
| `S3_ACCESS_KEY`, `S3_SECRET_KEY` | Required once an endpoint is set |
| `S3_REGION` | Defaults to `us-east-1` |
| `S3_FORCE_PATH_STYLE` | Defaults to `true`; self-hosted services need it |
| `S3_BUCKET_PREFIX` | Defaults to `ekklesia`; namespaces this installation on a shared service |
| `S3_DELIVERY_ENDPOINT` | The origin a **browser** reaches storage on. Defaults to `S3_ENDPOINT`; set it only when that is not browser-reachable |

Production requires `https`, exactly as the database transport requires verified TLS. An endpoint
embedding credentials is rejected rather than stripped. These values reach Core's provider and
nothing else — never a module, module UI, browser bundle, diagnostic, or API response.

The repository's development Compose stack already runs SeaweedFS for this, and `.env.example`
already carries matching values.

`S3_DELIVERY_ENDPOINT` exists because an S3-compatible signature covers the request's `Host`, so a
delivery URL cannot be signed for one origin and rewritten to another — the browser-facing origin has
to be known when the URL is signed. Leave it unset whenever `S3_ENDPOINT` is itself reachable from a
browser, which is true of the development stack (Compose publishes storage on loopback) and of any
deployment where Core and browsers use the same name. Set it when Core reaches storage privately —
`http://object-storage:8333` — and browsers reach it publicly. The scheme is *not* signed, so a
reverse proxy may terminate TLS and forward plaintext, **provided it passes the original `Host`
header through unchanged**; a proxy that rewrites `Host` to its backend makes every delivery URL fail
with `403`.

### Logical buckets, and what modules get

A module declares logical buckets in its manifest (`storage.buckets`) and receives a scoped facade at
`context.storage` on its runtime and HTTP handler contexts. It names a logical bucket; the host maps
it to a physical one deterministically and never exposes that name.

An installation with **no** provider is a complete deployment: modules declaring no bucket run
normally. A module that *does* declare one is reported `blocked · storage_unavailable` on
`GET /api/v1/module-lifecycle`, naming the logical bucket:

```json
{ "runtime": { "state": "blocked", "reason": "storage_unavailable",
               "storageBucketId": "announcement-assets" } }
```

Its required dependents block behind it through the ordinary dependency mechanism.

### Binding availability is not provider health

Eligibility asks whether storage is **configured**, never whether the service answered. A configured
provider that is temporarily unreachable is an operational error the module catches and retries —
`running` still means the `start` hook completed, and no lifecycle state changes because a storage
call timed out. Buckets are created lazily on first use rather than at startup, so a storage outage
never prevents the platform from booting, and nothing ever deletes a bucket.

### Serving objects to browsers

A module can hand a browser a temporary URL and let it fetch the object from storage directly
([ADR 0031](../architecture/decisions/0031-direct-object-delivery.md)):

```text
browser → GET /api/v1/modules/<id>/slides → module route → { key, url, expiresAt }
browser → GET <url> ─────────────────────────────────────→ storage service
```

The module authorizes the request in its own route, asks its logical bucket for a URL, and returns
it. Ekklesia is not in the data path, so byte ranges, caching, and revalidation are the storage
service's job — a `Range` request against a delivery URL is answered directly with `206`, which is
what makes media seeking work without a range proxy in Core.

```ts
const assets = context.storage.bucket("announcement-assets");
const { url, expiresAt } = await assets.createReadUrl("slides/welcome.png");
```

A delivery URL is a **bearer capability**: whoever holds it can read that one object until it
expires. Deciding who may have one is the module's authorization decision, made before it asks;
the storage contract only bounds how far the URL reaches. Treat a complete URL as a secret — never
log one (see [Security](security.md#object-storage-is-scoped-by-binding-and-modules-never-hold-credentials)).

Streaming through the module's own route with `context.storage.get(...)` remains available and is
still the right choice when the response must be computed, transformed, or authorized per byte.
There are no signed **write** URLs: uploads go through a module route, where the module can validate
what it is accepting.

## Module HTTP routes

An installed module's routes (declared via `http.route(...)` in its `register` hook — see
[Module development](module-development/README.md#http-routes) for the module-authoring contract)
are mounted under `/api/v1/modules/<module-id>`, in the module runtime's published order, by
`apps/api/src/module-http-routes.ts`. Mounting never invokes a handler.

- **Enforcement.** Every route runs through the identical `requirePermission` middleware Core's own
  routes use, with the route's own declared permission id as the only permission ever consulted —
  its trusted baseline always comes from the composed platform registry, never from the route or
  the request. A `public`, `authenticated`, and `restricted` module permission are enforced through
  the exact same path; a public-baseline route still resolves and receives an authorized principal.
- **Trusted origin.** `POST`, `PUT`, `PATCH`, and `DELETE` module routes additionally require an
  exact trusted `Origin` (see [Origin, CORS, and cookies](#origin-cors-and-cookies)) before
  authorization ever runs, exactly like the existing grants-mutation routes. `GET` does not require
  one. `OPTIONS` remains entirely platform-owned: a module cannot register it, and a CORS preflight
  never reaches authorization or a handler. `GET` inherits Hono's own implicit `HEAD` handling
  (same permission check, response body stripped); a module cannot separately author `HEAD`.
- **Body limit.** A route's own `maxBodyBytes` (see [Module development](module-development/README.md#http-routes)
  and [Request body limits](#request-body-limits)) is enforced immediately after the trusted-origin
  check and before permission evaluation — `0` means no body is accepted at all. An oversized
  request receives the platform's stable `413 {"code":"REQUEST_BODY_TOO_LARGE"}` and never reaches
  authorization, the handler, or the module database.
- **Handler input.** A handler receives the raw Fetch `Request` (so `Request.url`, headers, and body
  all behave normally, and the full query string remains available) plus a frozen context of exactly
  `moduleId`, the authorized `EkklesiaPrincipal`, frozen Hono-decoded path parameters (an empty
  frozen object for a route with none), `services`, and the module's scoped configuration, storage,
  and realtime publisher. It never receives a Hono `Context`, authorization/grant stores, jobs, or
  another module's scoped objects. Core does not serialize configuration or secrets into responses.
- **Handler failure.** A handler that throws, rejects, or returns anything other than a `Response`
  becomes a fixed `500 {"code":"MODULE_HANDLER_FAILED"}` with `Cache-Control: no-store` — never the
  caught value, a stack, the route's path or permission id, or the module id. A `Response` a
  handler deliberately returns — including a 4xx/5xx one — passes through unchanged.
- **CORS and caching.** The platform remains the sole owner of CORS for module routes exactly like
  every other `/api/v1/*` route (see [Origin, CORS, and cookies](#origin-cors-and-cookies), now
  additionally advertising `PUT`/`PATCH`/`DELETE`); any CORS-control header a handler sets on its
  own response is stripped before the response leaves the process. Every module response is forced
  to `Cache-Control: no-store` and a `Vary` that always includes `Origin` and `Cookie` (a handler's
  own legitimate `Vary` values are preserved alongside those) — deliberately conservative until a
  dedicated cache-policy contract exists, since even a public-baseline handler receives an
  authorized principal and may personalize its response.
- **Static snapshot.** Routes are part of the same static startup snapshot as the permission
  registry: installing, removing, or editing a module's routes takes effect only on the next API
  restart. Removing a module makes its former routes return the ordinary `404 {"code":"NOT_FOUND"}`.

## Module realtime

An eligible module may declare a module-local state channel during `register`. Ekklesia mounts one
exact route for each successfully started declaration:

```text
GET /api/v1/module-realtime/<module-id>/<channel>
```

The path is host-private. Module UI receives the module's server-produced realtime base in its mount
descriptor and calls `context.realtime.subscribe("state", handlers)`; it never builds this URL,
chooses a module id, supplies a permission, or sees an EventSource. Browser commands remain ordinary
module HTTP requests, retaining their body, origin, authorization, and response semantics.

The response is a Server-Sent Events stream. It begins with exactly one authoritative `snapshot`
event and then carries `update` events:

```text
event: snapshot
data: {"sequence":1,"payload":{"mode":"on","currentSlide":0}}

event: update
data: {"sequence":2,"payload":{"mode":"on","currentSlide":1}}
```

Core atomically captures the latest state while installing the subscriber, so there is no separate
GET-then-subscribe gap. Sequence is monotonically increasing per channel for this process only; there
is no `id` field, `Last-Event-ID` replay, durable history, or exactly-once promise. A reconnect is a
new subscription and begins with the latest snapshot. Slow clients retain only the newest pending
update, so sequence gaps are expected and intermediate states may be coalesced.

- **Authentication and authorization.** The SSE GET uses the existing session cookie and the same
  principal resolver, permission registry, and evaluator as module HTTP. The channel's declaration
  fixes the permission before any request exists. Permission evaluation completes before a Core
  subscriber is allocated. There is no query-string token or second realtime auth system.
- **Reauthorization.** A stream has a host-owned maximum age of five minutes. The host closes it at
  that boundary; native EventSource reconnects and the ordinary cookie and permission checks run
  again. Grant revocation, grant expiry, and session changes therefore do not leave an accepted
  stream authorized forever.
- **Bounds.** Core accepts at most 16 channels per module, 100 subscribers per channel, and 64 KiB
  of serialized JSON per publication. Publishing never awaits a browser. One latest snapshot and
  one pending update per subscriber are the only state retained.
- **Transport behavior.** Responses use `text/event-stream; charset=utf-8`, `Cache-Control:
  no-store`, `Vary: Origin, Cookie`, `X-Accel-Buffering: no`, and a 15-second comment keepalive.
  Keepalives consume no module sequence and never reach module UI. A reverse proxy must forward
  streaming bodies incrementally and permit that idle interval. No proxy product is part of the
  module contract.
- **Methods and errors.** Hono normally lets `HEAD` reuse GET handlers; this route explicitly returns
  `405` before allocating a stream because discarding an infinite body could leak a subscriber.
  Subscriber exhaustion is the fixed `503 {"code":"REALTIME_SUBSCRIBER_LIMIT"}`; other stopped or
  unavailable runtime states are `503 {"code":"REALTIME_UNAVAILABLE"}`. Raw errors, paths, URLs,
  payloads, permissions, secrets, and stacks are never reflected.
- **Lifecycle.** Disabled and blocked modules have no route because their entrypoint never runs and
  their channel is never mounted. Shutdown first refuses and closes realtime subscriptions so they
  cannot hold HTTP drain open. Ordinary in-flight module HTTP handlers may finish and publish during
  that drain. Core then stops modules in reverse dependency order, deactivating each module's scoped
  publisher immediately before its own stop hook; a captured publisher fails predictably afterward.

Realtime carries opaque module state. If a module includes a direct object-delivery URL, the exact
serialized string passes through unchanged: Core neither re-signs it per subscriber nor extends its
expiry. Image, video, and audio bytes remain object-storage traffic, never realtime payloads. See
[ADR 0032](../architecture/decisions/0032-native-module-realtime-transport.md).

## Dashboard navigation

```text
GET    /api/v1/dashboard/navigation
PUT    /api/v1/dashboard/navigation/groups/<group-id>        { "label": "Media", "order": 10 }
DELETE /api/v1/dashboard/navigation/groups/<group-id>
PUT    /api/v1/dashboard/navigation/placements/<module-id>   { "groupId": "media"|null, "order": 10 }
```

How this installation presents its mountable modules
([ADR 0027](../architecture/decisions/0027-dashboard-navigation-and-flat-modules.md)).

```json
{
  "groups": [
    {
      "id": "media",
      "label": "Media",
      "order": 10,
      "modules": [
        {
          "name": "Livestream",
          "module": {
            "id": "livestream",
            "version": "1.0.0",
            "ui": { "url": "/api/v1/module-ui/livestream/1.0.0" },
            "api": { "baseUrl": "/api/v1/modules/livestream" },
            "realtime": { "baseUrl": "/api/v1/module-realtime/livestream" },
            "permissions": ["livestream.streams.read"]
          }
        }
      ]
    }
  ],
  "ungrouped": []
}
```

### Navigation versus inventory

They are deliberately separate, and this route is layered **on top of** the inventory rather than
merged into it:

- `GET /api/v1/module-ui` answers *what may be mounted*, host-neutrally. A standalone host consumes
  it and never learns Ekklesia has folders. It carries no display name, group, or order.
- `GET /api/v1/dashboard/navigation` answers *how this Dashboard presents those modules*, and
  **embeds** each generic descriptor untouched under `module`.

Both compose that descriptor through one shared function, so they cannot disagree about what exists
or about any URL.

### Grouping is presentation, never module structure

A group is a label on a rendering bucket. It is **not a module**: it has no manifest, version,
artifact, API, permission, or lifecycle, and nothing depends on it. Placing Livestream and Video in
one folder creates no relationship between them, and moving a module changes a single presentation
row — never its artifact, API root, permissions, dependencies, or startup order. Real technical
dependencies remain the module dependency contract's job and are the only relationship modules have.

### Resolution rules

- A module with **no placement is ungrouped and visible** — installing it is enough, with no
  initialization write and no manual database edit.
- A stored placement for a module that is **not installed renders nothing and is retained**, so
  uninstall-and-reinstall preserves an arrangement.
- **Empty groups are kept** (hiding one would look like deletion), and a placement naming a missing
  group renders as ungrouped rather than disappearing.
- Deleting a group **ungroups its modules** through the schema's `on delete set null`; a dangling
  reference is impossible.
- Ordering is total: groups by `(order, id)`, modules by `(configured, order, id)` — placed modules
  in the administrator's order, then never-configured ones by id, with the identifier always the
  final tie-break, so colliding order values are still deterministic.

### Authority

Reading requires `core.modules.ui.read` — the same permission the inventory requires, because this
response contains it and serves the same audience. Organizing requires
`core.dashboard.navigation.manage` (baseline `restricted`), a **platform** permission and never a
module's, plus an exact trusted `Origin` and a bounded body. Holding it grants no module authority:
every module route stays enforced against its own declared permission, and no arrangement of folders
changes an authorization decision.

Group ids are opaque, stable, lowercase kebab-case keys, never displayed and never re-derived from a
label — so renaming a group cannot re-organize the Dashboard. Bodies are closed: any key beyond the
declared ones makes the whole request invalid rather than being silently dropped. Configuration is
platform-global; Ekklesia has no organization or tenant model to scope it to yet.

Responses are per-principal, so `Cache-Control: no-store` and `Vary: Origin, Cookie`.

## Module UI inventory

```text
GET /api/v1/module-ui
```

The one query a browser host makes before it can mount anything
([ADR 0026](../architecture/decisions/0026-minimal-ekklesia-module-ui-host.md)). It is the collection
root of the same resource the artifact route below serves: one prefix, two representations.

```json
{
  "modules": [
    {
      "id": "example",
      "version": "0.1.0",
      "ui": { "url": "/api/v1/module-ui/example/0.1.0" },
      "api": { "baseUrl": "/api/v1/modules/example" },
      "realtime": { "baseUrl": "/api/v1/module-realtime/example" },
      "permissions": ["example.records.read"]
    }
  ]
}
```

- **Six fields, each because a `ModuleUiHostContext` needs it.** No filesystem path, package root,
  package name, server/schema/migrations entrypoint, capability, dependency, or storage bucket. And
  no `name`, `icon`, `description`, `category`, `folder`, or `order`: those are navigation metadata,
  and [ADR 0021](../architecture/decisions/0021-modules-as-hostable-products.md) keeps Dashboard
  grouping an administrator's presentation configuration rather than a module declaration.
- **All three URLs are server-produced.** `ui.url` comes from `moduleUiArtifactUrl`, `api.baseUrl`
  from `moduleApiBaseUrl`, and `realtime.baseUrl` from `moduleRealtimeBaseUrl`, so each URL shape is
  constructed in exactly one place in the repository. A browser never rebuilds any platform module
  prefix, which lets one change without touching host or module code.
- **The module set is the publication registry**, not a rescan: modules that survived discovery and
  declare `entrypoints.ui`, captured after the module runtime started. A module with no UI is absent.
  Order is the registry's own deterministic discovery order.
- **Authenticated, unlike the artifact route.** Requires `core.modules.ui.read`, baseline
  `authenticated` — any signed-in user, and nobody else. The artifact bytes are anonymous because
  fetching them confers no authority, but that route is deliberately not an enumeration oracle, and
  this endpoint *is* that enumeration. `authenticated` rather than `restricted` because a shell that
  rendered nothing until an operator granted every user a permission would be unusable.
- **`permissions` is the subset of that module's own declared permissions the caller holds**,
  evaluated server-side by Core's one authorization evaluator — so baselines, the platform-owner
  bypass, grant expiry, and deny-by-default behave exactly as on a real request. Granted ids only,
  scoped to the module: no module learns the viewer's authorization state anywhere else in the
  platform. **These are presentation hints, never authority**; every module route stays enforced by
  `requirePermission` regardless of what a browser was told.
- **Failures.** Anonymous or unpermitted callers get `403 {"code":"AUTHORIZATION_DENIED"}` from the
  shared middleware. An authorization-evaluation outage is `503
  {"code":"AUTHORIZATION_EVALUATION_FAILED"}` rather than an empty hint set, because a hint that
  silently read "denied" during an outage would blank the UI and look like a permissions problem.
- **Per-principal**, so `Cache-Control: no-store` and `Vary: Origin, Cookie`, exactly like
  `/api/v1/principal`. `GET` only; no other method is routed.

## Module UI asset publication

A module may declare `entrypoints.ui`, a compiled browser ESM artifact
([ADR 0021](../architecture/decisions/0021-modules-as-hostable-products.md),
[ADR 0023](../architecture/decisions/0023-module-ui-build-pipeline.md)). Core's discovery resolves it
to an absolute path on the server's filesystem; a browser needs a URL. This route is the bridge
([ADR 0025](../architecture/decisions/0025-trusted-module-ui-asset-publication.md)).

```text
GET /api/v1/module-ui/<module-id>/<version>
```

- **Identity, not paths.** `:moduleId` and `:version` are lookup keys handed to a publication
  registry Core builds from the discovered, started module inventory. Only an exact match on **both**
  causes the already-canonical artifact path discovery validated to be read. No request value is ever
  joined, resolved, or concatenated into a filesystem path, and the route declares no wildcard
  segment — so traversal, encoded separators, sibling bundle chunks, package-root files, and source
  maps are unreachable by construction. **An installed module directory is never a static file
  root**, and only the single artifact its manifest declares is ever served.
- **A sibling of `/api/v1/modules/`, never a child.** A module may legitimately register the route
  path `/ui/:version`, so nesting publication inside the module prefix would let a module and the
  platform compete for one URL with routing order deciding the winner.
- **Publishable means discovered and started.** A module directory on disk is not enough: the module
  must be in the validated set the API accepted at startup *and* declare `entrypoints.ui`. There is
  no installer, enable/disable state, or trust store to consult, and none was invented here.
- **No authentication, deliberately.** Fetching the bundle confers no authority — every module API
  stays behind `requirePermission`, and the host, not the artifact, supplies the principal and
  permissions at mount time. Requiring a cookie would also break cross-origin dynamic `import()`,
  which sends none. The cost, stated plainly: a caller who already knows a module id **and** its exact
  installed version can download that bundle. Unknown module, no declared UI, and wrong version all
  return one byte-identical `404 {"code":"MODULE_UI_NOT_PUBLISHED"}`, so the route is not an
  enumeration oracle.
- **Response.** `Content-Type: text/javascript; charset=utf-8` with `X-Content-Type-Options:
  nosniff`, and the artifact byte-for-byte — never transformed, re-bundled, minified, rewritten, or
  executed. A module UI whose top level throws is served like any other; the server never imports it.
- **Caching.** `Cache-Control: no-cache` plus a strong SHA-256 `ETag` over the bytes actually served,
  with RFC 9110 `If-None-Match` handling (wildcard, weak comparison, `304`). Not `immutable`: without
  an installer nothing stops a version being rebuilt with different bytes, and an unrecoverable stale
  cache is worse than one conditional request. A version bump changes the URL anyway.
- **Artifact changed after discovery.** Discovery validated the artifact at startup; it may vanish
  later. The route opens exactly the previously resolved path once, with no fallback and no
  re-resolution, and a read failure returns a sanitized `503
  {"code":"MODULE_UI_ARTIFACT_UNAVAILABLE"}` carrying no path, `errno`, or caught error.

`moduleUiArtifactUrl({ moduleId, version })` (`apps/api/src/module-ui-routes.ts`) is the one
construction of this URL. It accepts identity only and percent-encodes both components. The result is
root-relative, so a same-origin host uses it unchanged and a split-origin host resolves it against
the same API base URL it already needs for `ModuleUiApiClient`.

**Nothing mounts a module UI yet.** A host can now obtain a browser URL that
`@ekklesia/module-ui-host` accepts, which was previously impossible — but Ekklesia has no Dashboard
shell, no navigation, and no code that calls this route outside its own tests.

## Module database access

`context.services.database` gives a module's `start`/`stop` lifecycle hooks and every authorized
HTTP handler a typed Drizzle query surface over its own tables — never during `register`, which
stays declarative and database-free. See
[Module development](module-development/README.md#database-access) for the module-authoring
contract; this section covers only how `apps/api` composes and wires it.

- **`ModuleDatabase`.** A restricted, typed view over `NodePgDatabase`
  (`packages/contracts/src/module-database.ts`), exposing exactly `select`, `selectDistinct`,
  `selectDistinctOn`, `insert`, `update`, `delete`, `execute`, and `transaction`, each preserving
  Drizzle's own table-aware argument and result inference. It exposes no `$client`, no internal
  session/dialect state, no schema-bound relational `query` builder, and no connection string,
  pool, or raw `pg` client — not merely hidden by the TypeScript type, but physically absent from
  the constructed object at runtime (`core/src/modules/database.ts`'s `createModuleDatabaseView`
  builds a frozen object with each operation explicitly bound, rather than handing out Drizzle's
  own instance).
- **One pool, one facade, one services object.** `apps/api/src/database.ts`'s `createApiDatabase`
  wraps the one shared PostgreSQL pool in three independent typed Drizzle views — identity,
  authorization, and now `module` (a `ModuleDatabase`). `startApi` builds one frozen
  `ModuleRuntimeServices` object (currently just `{ database }`) from that pool and passes the
  *same* object to every module's `start` and `stop` hook and to every module HTTP handler for the
  lifetime of one running API — never a fresh object per hook or per request, and never a second
  pool or facade.
- **No database access without authorization.** A handler's context — and therefore
  `services.database` — is unreachable until `requirePermission` (and, for a state-changing
  method, `requireTrustedOrigin`) has already succeeded; a denied request, a rejected-Origin
  request, and an `OPTIONS` preflight never invoke a module handler at all, so they never touch a
  module's database.
- **Errors.** A handler-thrown database error is sanitized through the exact same
  `MODULE_HANDLER_FAILED` boundary described above — there is no separate, database-specific public
  error contract. A module that deliberately catches its own database failure may still return any
  valid `Response` of its own choosing.
- **Migrations stay explicit.** Constructing the facade performs no query and no migration.
  Module schema and migration changes still follow the existing generation, validation, policy, and
  deployment flow (see [Module development](module-development/README.md#database-access)); nothing
  in module discovery, registration, start, or an HTTP request infers a schema or creates a table
  automatically. See [ADR 0016](../architecture/decisions/0016-module-database-access.md) for the
  complete design and its rejected alternatives.

## Database connection limits

`apps/api/src/server.ts` configures the one shared PostgreSQL pool (`createApiDatabase`,
`apps/api/src/database.ts`, built on `createValidatedPostgresPool` from
`core/src/database/pool.ts`) with both a connection timeout and an operation timeout (10 seconds
each), the same bounded, validated pattern `core/src/migrations/postgres-database.ts` already
uses. `connectionTimeoutMillis` alone only bounds *establishing* a connection; it does nothing once
a query is already running on one. The operation timeout additionally sets `pg`'s `query_timeout`
and `statement_timeout`, so a hung or unexpectedly slow query is aborted rather than holding a pool
connection (and its caller) indefinitely. An invalid timeout (non-integer, zero, negative, or
larger than PostgreSQL's own maximum) is rejected as configuration error before any pool is
constructed, sanitized the same way an invalid connection string already is. The identity,
authorization, and module domains all read from this one pool through independent typed Drizzle
views — see [Authorization](authorization.md#shared-database-composition) for why there is exactly
one pool, not two.

## Database transport (TLS)

`core/src/database/transport.ts` is the one typed policy deciding how *every* Ekklesia executable
reaches PostgreSQL. It has exactly two modes:

| Mode | Meaning | Default in |
| --- | --- | --- |
| `verify-full` | TLS required, certificate chain verified, **and** the server's identity verified against the configured host. | production |
| `disable` | No TLS. Refused outright in production. | development/test |

There is deliberately no encryption-only mode (`require`), no chain-only mode (`verify-ca`), and no
`rejectUnauthorized: false` escape hatch: encryption without identity verification is exactly what
an interception attack defeats. Every loose or aliased spelling — `VERIFY-FULL`, `" verify-full"`,
`require`, `prefer`, `no-verify`, `true`, `1`, empty — is rejected with `DATABASE_TLS_MODE_INVALID`
in *both* environments, so a typo can never quietly become a weaker mode.

**Production has no plaintext exception — not even for a loopback database.** `disable` in
production fails with `DATABASE_TLS_REQUIRED_IN_PRODUCTION` whether `DATABASE_URL` points at a
remote host, `127.0.0.1`, `localhost`, or `[::1]`. A connection string cannot prove the database is
actually process-local, so no "same host, therefore safe" carve-out exists.

Verification is certificate *and* hostname, not encryption alone. The policy sets
`rejectUnauthorized: true` and an explicit `checkServerIdentity` pinned to the host parsed from
`DATABASE_URL`. That second part closes a real gap rather than restating a default: the pinned `pg`
driver sets TLS `servername` only for non-IP hosts, and Node performs **no** identity check when it
is unset — so `rejectUnauthorized: true` alone accepts a certificate from any host the trusted CA
ever signed when you connect by IP address. Both halves are proven against a real TLS-enabled
PostgreSQL in `core/tests/database-tls.integration.ts`.

**`DATABASE_URL` may not carry its own TLS parameters.** `sslmode`, `ssl`, `sslcert`, `sslkey`,
`sslrootcert`, `sslpassword`, `sslnegotiation`, and `uselibpqcompat` are all rejected
(case-insensitively) with `DATABASE_TLS_CONNECTION_STRING_CONFLICT`. This is not stylistic: the
pinned driver merges parsed connection-string values *over* explicitly passed options, so a
`?sslmode=disable` in the URL would silently replace the policy's verified `ssl` object. TLS is
decided in exactly one place, and it is never the URL.

For a database behind an internal CA, supply the CA's PEM text in `EKKLESIA_DATABASE_CA`; it is
added to the trusted set for verification and never appears in any error, log, or `pnpm doctor`
output. Loading it from a file path or a secret store is out of scope here, as are certificate
rotation and client-certificate authentication.

The same policy is resolved identically by every executable that opens a connection — `apps/api`,
`pnpm migrations:plan`, `pnpm migrations:apply`, the platform-owner bootstrap CLI, and
`pnpm doctor`'s own connection — through one shared environment reader
(`core/src/cli/database-transport-environment.ts`), so `process.env` stays out of Core's reusable
constructors. `apps/api` validates it in `createApiRuntimeConfig`, *before* any pool, Better Auth
instance, module runtime, or listener exists. `pnpm doctor` reports it as `database.transport` in
its PostgreSQL section: a verified production policy passes, and an unsafe or invalid one is a
blocking `NOT READY` failure — never a warning — naming only the effective mode, never the URL,
host, port, credentials, or CA material.

TLS protects the *connection*, not the exposure: it does not make an internet-reachable database
safe, and the network boundary remains the deployment's own responsibility. See
[ADR 0018](../architecture/decisions/0018-identity-session-resolution-and-database-transport.md).

## Graceful shutdown

`startApi()` (the composition function `apps/api/src/server.ts` exports, used by both the
executable entry point and tests) returns `{app, server, shutdown}`. `shutdown()` memoizes one
shared promise the first time it is called: every caller, concurrent or sequential, awaits the
exact same operation, and that operation runs exactly once no matter how many times `shutdown()`
is called. In order: it refuses new realtime subscribers and closes every existing stream; then
stops the HTTP listener from accepting new connections and lets finite in-flight requests finish;
then stops the module runtime in reverse dependency order; then closes storage and the one shared
PostgreSQL pool (identity, authorization, and the module database facade all read from it — see
[Authorization](authorization.md#shared-database-composition)). Unlike a startup-failure cleanup
path (below), a module-stop failure here is not swallowed — it is a genuine result a caller should
observe — but pool cleanup always still runs first, so a stop failure never leaves the pool open,
and no module code can run against it afterward.

Closing streams and disabling publishers are deliberately separate. Streams are long-lived HTTP
responses and would otherwise force every healthy shutdown to its drain deadline. Publishers remain
active during the finite HTTP drain so an already-running module command can commit and publish its
result without turning a successful request into a false `500`. As reverse-order module stop reaches
each module, Core disables that module's publisher immediately before its stop hook; dependencies
remain live while their dependents stop, and a stale publisher fails predictably once its own module
begins stopping. A module's `stop` hook may still use `services.database` and storage here, since
those providers have not yet closed.

The executable entry point's `SIGINT`/`SIGTERM`
wiring is extracted into `installShutdownSignalHandlers(running, exit)` — the same function production
composition uses, but independently testable with a fake `running`/`exit` so a test can trigger "a
signal arrived" without sending a real process signal or invoking the real `process.exit`.

A startup failure after the database pool was already created always closes that pool before the
failure propagates: module-services/identity/authorization construction failing (no module runtime
was started yet — the frozen `ModuleRuntimeServices` object is built in the same step, from the
same pool); the server failing to bind its listen address; or an app-construction failure, both of
which also stop the module runtime first, since it may have already started successfully. A module
load/register/start failure uses Core's own rollback — the module whose `start` hook failed is
stopped first, through its own `stop` hook, then already-started modules in reverse order, each
under the same bounded per-hook deadline — before the failure propagates, and then closes the pool — there is no separate
runtime handle to stop in that case, since `startDiscoveredModuleRuntime` never returned one.

## A real signup → sign-in → principal → sign-out transcript

Captured from this PR's own integration test against a real, migrated, disposable PostgreSQL
database, configured as **development** with public sign-up explicitly enabled — public sign-up is
disabled by default and cannot be enabled in production at all (see
[Public registration policy](#public-registration-policy)). Session tokens, cookie values, and
generated ids below are sanitized/truncated examples, not real credentials.

```text
POST /api/auth/sign-up/email
{"email":"[email protected]","name":"Alice","password":"<redacted>"}
→ 200
{"token":null,"user":{"id":"2aKp4jPeV1QbXwyz1OH6IcAUfl2n1fc1","email":"[email protected]", ...}}

POST /api/auth/sign-in/email
{"email":"[email protected]","password":"<redacted>"}
→ 200
Set-Cookie: better-auth.session_token=<redacted>; Max-Age=604800; Path=/; HttpOnly; SameSite=Lax; Secure
{"redirect":false,"token":"<redacted>","user":{"id":"2aKp4jPeV1QbXwyz1OH6IcAUfl2n1fc1", ...}}

GET /api/v1/principal
Cookie: better-auth.session_token=<redacted>
→ 200
Cache-Control: no-store
{"kind":"authenticated","userId":"2aKp4jPeV1QbXwyz1OH6IcAUfl2n1fc1"}

POST /api/auth/sign-out
Cookie: better-auth.session_token=<redacted>
Origin: https://dashboard.example.test
→ 200
Set-Cookie: better-auth.session_token=; Max-Age=0; Path=/; HttpOnly; SameSite=Lax
{"success":true}

GET /api/v1/principal
Cookie: better-auth.session_token=<redacted, now revoked>
→ 200
{"kind":"anonymous"}
```

## What's next

See the repository root [README](../../README.md#current-status) and
[platform architecture](../architecture/platform-architecture.md#current-implementation-status)
for what has landed across every phase and what remains before Surface work begins. This document
tracks `apps/api`'s own HTTP, module-runtime, and database-composition contracts as they evolve.