Ekklesia/docs/development/testing.md

Testing

Ekklesia separates fast code-level evidence from live infrastructure evidence.

Unit and repository verification

pnpm test
pnpm verify

These commands exercise package tests, repository boundaries, type checking, offline Core and module migration discovery, identity schema drift detection, formatting, linting, and builds. Migration-runner unit tests use a narrow database/session port to prove deterministic planning, approval, transaction orchestration, cleanup, and sanitized errors. They do not claim to prove PostgreSQL behavior. Identity unit tests cover branded opaque-ID contracts, principal shapes, configuration validation (secret rules, canonical-origin-only base URL and trusted origins, and the full loopback/unspecified classifier production rejects — .localhost, all of 127.0.0.0/8, 0.0.0.0, ::1 compressed and expanded, ::, and IPv4-mapped forms), bounded/validated connection and operation (query_timeout/statement_timeout) database timeouts, the Core identity runtime (explicit construction, exact schema mapping, only email/password and rate limiting enabled, an explicit advanced.disableCSRFCheck/disableOriginCheck: false pair, an explicit error-only sanitized logger that strips Better Auth's own message and arguments regardless of what they contain, pinned 8–128 password-length bounds, sanitized failures, exact whole-name session-cookie detection that never fires on a prefixed, suffixed, or differently-cased lookalike, and that a broken session store during principal resolution is never silently reported as anonymous while an unrelated cookie performs no identity database work at all), the PostgreSQL transport policy (core/tests/database-transport.test.ts: mode validation including every loose or aliased spelling, production's plaintext refusal with no loopback exception, connection-string TLS-parameter rejection, and that no configuration path yields rejectUnauthorized: false or leaks CA material into an error), and Core migration discovery (deterministic ordering, the reserved core identity, and collision rejection) — all without a live database. apps/api's own unit tests cover its pure Hono app factory (route shapes, module route mounting/enforcement/CORS/response sanitization, CORS allow/deny/preflight, deterministic 404 and sanitized error responses), its runtime configuration, its idempotent shutdown lifecycle (concurrent callers share one server-close-then-module-runtime-stop-then-pool-close promise; a module-stop failure never prevents pool cleanup; a startup failure after the pool was already created still closes it, stopping the module runtime first if it had already started; SIGINT/SIGTERM wiring is exercised through an injectable exit function rather than the real process.exit), and that importing any of its modules — including the executable server module without calling its exported main — never reads the environment, connects to PostgreSQL, imports a module entrypoint, or starts a server. Real, working (non-throwing) module fixtures prove production module-runtime startup/shutdown ordering and cleanup against a mocked pool/Better Auth/server — see Authorization.

Phase 7B pass 1 adds browser-authentication evidence (ADR 0034). API tests pin the two exact browser-boundary exceptions. Any body-bearing successful 2xx from POST /api/auth/sign-in/email preserves Better Auth's Set-Cookie and status but returns fixed {"ok":true}, with no token/user body. /api/auth/get-session, /api/auth/list-sessions, /api/auth/update-session, and /api/auth/change-password each receive the identical sanitized 404 without invoking the identity handler; failures and every other supported auth route retain their native semantics. The live browser transport additionally proves the host-only cookie and that no token, session, user, or email appears in the sign-in body.

Shell protocol and React tests cover principal-first anonymous/authenticated/failure states; the labelled email/password form and keyboard submission; autocomplete, pending state, and same-tick duplicate suppression; password clearing; fixed credential-versus-service errors; no credential or raw exception logging; fresh principal and navigation resolution after login; sign-out followed by principal resolution; and retry without a full-page reload.

The same shell suites exercise session transitions under repeated timing: active module cleanup and realtime teardown on sign-out/verified expiry; stale attempts after unmount; platform bootstrap 401 mapping directly to unauthenticated; and deduplicated principal revalidation for both module 401 and 403. They prove an authenticated recheck preserves the module's original response, session, and mounted UI; an anonymous recheck unmounts once and shows login; a revalidation failure is a sanitized host failure; and a response from an already-unmounted frame cannot overwrite a newer transition. Return/callback input is not part of the browser contract, and tests pin that external, protocol-relative, backslash-prefixed, and scheme-bearing values cannot become navigation. Modularity checks keep authentication out of the module UI SDK/context and Dashboard login code out of the separate display bundle.

pnpm verify is intentionally Docker-free. It never plans or applies migrations against a live database.

Live PostgreSQL migration evidence

Docker and the Docker Compose plugin are required for the live suite:

pnpm test:migrations:integration

The command first builds the workspace through the repository's Turbo build graph, then starts the repository's pinned PostgreSQL image in a unique Compose project, waits for bounded health readiness, runs the production PostgreSQL adapter and migration services, and removes the container, network, and volume. A build failure occurs before Docker provisioning. Each scenario also receives a safely named disposable database. Direct host runs use a Docker-assigned random port bound only to 127.0.0.1; a containerized Gitea job joins the isolated Compose network and uses the PostgreSQL service name.

The live suite proves fresh read-only planning, initial and incremental application, genuine no-op behavior, exact applied checksums and history, edited or corrupt history rejection, whole-batch DDL rollback, Drizzle breakpoint execution, real advisory-lock concurrency and timeouts, lock release after failure, exact destructive approvals, and public-error sanitization. It does not make migrations automatic during Core startup.

A dedicated set of scenarios exercises the committed Core identity migration against real PostgreSQL: read-only planning on a fresh database, applying it and confirming exactly the four expected ekklesia_core_* tables and their indexes exist, a genuine no-op on a second apply, Core executing before a fixture module migration in one batch, previously applied module-only history staying byte-for-byte unchanged once Core is introduced, and a later module migration's failure rolling back cleanly without damaging the already-committed Core identity migration or leaking the advisory lock.

A second dedicated suite (apps/api/tests/api-http.integration.ts, included in the same pnpm test:migrations:integration run) starts the real apps/api HTTP server against a freshly migrated disposable database and exercises it with real HTTP requests: public sign-up (with no automatic session, per autoSignIn: false), explicit sign-in producing a host-only HttpOnly/SameSite=Lax/secure session cookie and fixed {"ok":true} body containing no token, /api/v1/principal for both the authenticated and anonymous shapes, sign-out actually invalidating the session, a malformed or missing cookie resolving as anonymous, duplicate sign-up creating no second identity row, two distinct accounts producing identical principal shapes with no elevation and no owner/role/permission data at the database level, no plaintext password ever stored, the exact 8–128 password-length boundary (7 rejected, 8 accepted, 128 accepted, 129 rejected), CORS never returning an allow-origin header for an untrusted origin, Better Auth's handler being unreachable outside /api/auth/*, a database outage during principal resolution producing a sanitized 503 rather than a false anonymous 200, starting the API against an unmigrated database creating no tables at all, and a clean listener/pool shutdown.

The browser-authentication chain in that live suite also creates a development identity, invokes Core's real single-use platform-owner bootstrap for that existing id, signs in through the real HTTP transport, presents the returned cookie to /api/v1/principal, and reaches protected Dashboard navigation with the owner authority. It proves composition across identity, authorization, and the Dashboard resource; because production signup is prohibited, it deliberately does not claim to prove a production first-user enrollment path.

A dedicated group within that suite proves session-store readiness against the real database rather than a mock, since that is the only place a broken store looks exactly like an anonymous request: a request carrying no cookies and one carrying only an unrelated cookie both resolve as anonymous while issuing no identity database work at all; a request presenting the configured session cookie against a dropped session table, a dropped column, and a revoked SELECT privilege each produce a sanitized 503 PRINCIPAL_RESOLUTION_FAILED with no relation name, column name, SQL, or driver text in the body or the logs; and a genuinely expired or unknown session against a healthy store still resolves as anonymous.

The same suite also proves Better Auth's own origin/CSRF protection is production-equivalent under the test runner, not just in a plain Node script: a no-cookie first authentication attempt succeeds from a trusted browser origin and is rejected (403, no permissive CORS header, no leaked detail) from an untrusted one; a cookie-bearing state-changing request (sign-out) is rejected the same way from an untrusted origin, and the presented session remains valid afterward, proving the rejected request never reached sign-out logic. See ADR 0012 for why this previously could not be reproduced reliably under Vitest (Better Auth's own default silently skips these checks whenever isTest() is true, unless advanced.disableOriginCheck is set explicitly) and how the fix removes that gap entirely. A further scenario forces a real identity database failure and asserts, against actual console.error/warn/log output, that Better Auth's own internal error logging — configured explicitly in core/src/identity/runtime.ts — never emits a raw exception, SQL, connection detail, or the rejected request's origin.

A dedicated group in the API suite also drives Pass 3's explicit-grant invariants through a real authorized grants-management session: a malformed non-UUID revoke returning a sanitized 400 with no database work, a canonical unknown UUID returning 404, a scoped create body returning 400 with nothing written, normalized-impossible/non-canonical/past expirations returning 400, an equivalent creation returning a sanitized 409 that names no existing grant, a bounded paginated list traversed across three pages with relationship grants absent from every one, and one HTTP revocation clearing every unrevoked legacy equivalent with a per-row audit entry.

A third dedicated suite (apps/api/tests/module-database.integration.ts, included in the same pnpm test:migrations:integration run) proves context.services.database against a real, installed test module (core/tests/fixtures/module-database-integration/db-facade/ — never the repository's live modules/ directory, and invisible to normal discovery/Doctor/modules:validate unless a test explicitly points discovery at it): the module's own table does not exist until its migration is explicitly applied, and constructing the pool or facade never creates it; start, stop, and an authorized HTTP handler all receive the identical ModuleRuntimeServices object, proven by reference equality; a real authorized mutation persists and a later read observes it, gated by the same permission/trusted-Origin enforcement every other module route uses; a real committed transaction persists and a real rollback (a deliberate unique-constraint violation) leaves no partial mutation; a module deliberately catching a real PostgreSQL failure returns its own response untouched, while an uncaught one sanitizes to the existing MODULE_HANDLER_FAILED boundary with no leaked SQL, error text, or path; and a stop hook's write completes and is independently verifiable, through the harness's own administrative connection — never the API's own pool — before that pool closes. See ADR 0016.

A fourth dedicated suite (core/tests/database-tls.integration.ts, included in the same pnpm test:migrations:integration run) proves the production PostgreSQL transport policy against a real TLS-enabled PostgreSQL rather than against the shape of an options object: a trusted certificate whose identity matches the host connects and is confirmed encrypted through pg_stat_ssl; a certificate that does not chain to the configured CA is refused (UNABLE_TO_VERIFY_LEAF_SIGNATURE); the same server reached through an address its certificate does not cover is refused (ERR_TLS_CERT_ALTNAME_INVALID); the identical connection without the policy's pinned checkServerIdentity succeeds, which is the concrete gap the policy closes; and verify-full cannot be satisfied by a plaintext-only endpoint. Every CA, certificate, and private key is generated at test runtime into a temporary directory and deleted afterwards — no private key material is ever committed to this repository — and the disposable container is removed in afterAll even when an assertion fails. See ADR 0018.

A fifth dedicated suite (core/tests/module-lifecycle-store.integration.ts, included in the same pnpm test:migrations:integration run) proves the module enablement store against real PostgreSQL: that migration 0004 really creates ekklesia_core_module_state with exactly the four columns and the one declared CHECK constraint the schema states — and, load-bearingly, that enabled has no column default, so the platform's default for a newly discovered module is genuinely the absence of a row rather than a materialized decision nobody made; that a decision survives a completely new connection; that the upsert is idempotent on module_id and bumps updated_at in place rather than inserting a second row; that a malformed identifier is rejected before it reaches PostgreSQL and the empty-string case is refused by the database too; and that a record for a module which is no longer deployed is retained and simply ignored by resolution, so redeploying restores the decision.

The API suite additionally drives the whole lifecycle end to end against a real database and real fixture modules: a disabled module's compiled entrypoint is never imported (proven by its register/start/stop markers staying empty), its routes return the platform's ordinary 404, its UI artifact and inventory and navigation entries are all absent, and its neighbours run normally; a module blocked by a disabled required dependency is explained over HTTP with a structured reason naming that dependency and no filesystem path anywhere in the payload; a change made through PUT /api/v1/module-lifecycle/<id>/enabled reports restartRequired: true while the module demonstrably keeps serving, and is applied only by the next start; and a disabled module's migrations are still discovered, because deployed schema ownership is not runtime enabled state. See ADR 0028.

A sixth dedicated suite (core/tests/module-configuration-store.integration.ts, included in the same pnpm test:migrations:integration run) proves the module configuration store against real PostgreSQL: that migration 0005 really creates ekklesia_core_module_configuration with the composite (module_id, key) primary key, the declared CHECK constraints, and no column that could hold secret material; that a replace is transactional, whole-document, and scoped to one module so another module's rows are untouched; that stored values survive a completely new connection; that an oversized value is refused by the application and by the database; that a store error never carries the offending value; and — the security-relevant one — that a module id or key is only ever data in a parameterized column, so an injection attempt is a validation error and the table still exists afterwards.

The API suite additionally proves the configuration boundary end to end: a secret's value never appears in any response (including when a row for that key is planted directly in storage), a declared-but-undeclared key, a wrong JSON type, a constraint violation, and a secret key each fail with a distinct sanitized code and store nothing, a reserved property name such as __proto__ is rejected without touching Object.prototype, and a module blocked by its own configuration is absent from the mountable UI inventory exactly like a disabled one. See ADR 0029.

A seventh dedicated suite (core/tests/module-storage.integration.ts, included in the same pnpm test:migrations:integration run) proves the object-storage contract against a real S3-compatible server — the SeaweedFS service the repository's Compose file already pins, brought up alongside PostgreSQL by the same harness, so this adds no new CI infrastructure.

It exists because mocking would have hidden a real divergence: probing the pinned SeaweedFS 4.28 image directly showed that CreateBucket on an existing bucket reports BucketAlreadyExists, not AWS's BucketAlreadyOwnedByYou. A provider treating only the AWS spelling as success would have failed on every second boot against this repository's own development storage, and no unit test would have caught it. The suite pins that idempotency, plus a real round trip with content type and unquoted ETag preserved, streaming a body without buffering, a Web stream going straight into a Response, prefix listing and real pagination, a missing object mapping to a structured not-found error carrying no endpoint or XML, two modules declaring the same logical bucket staying genuinely isolated on one server, the longest bucket name the mapping can produce being accepted, a real connection failure mapping to STORAGE_PROVIDER_UNAVAILABLE, and a bucket surviving the deletion of every object in it.

An eighth suite (core/tests/module-delivery.integration.ts, same harness) proves direct browser delivery against that same live server, because every claim it makes is a property of the agreement between the signer and the provider rather than of code in this repository (ADR 0031). It pins: a delivery URL returning the object with its content type intact to an ordinary credential-free fetch; the same object being refused without the URL's authorization, so delivery can never be quietly "fixed" by making a bucket public; tampered signatures, swapped keys, and swapped buckets all rejected; expiry actually enforced; a Range request answered directly with 206 and a correct Content-Range; repeated calls inside a window returning a byte-identical URL so browser caches and URL-keyed DOM elements hold; Cache-Control never outliving the URL; an optional download filename; and two modules' identical logical bucket names resolving to different objects.

Its most load-bearing tests concern the browser delivery origin. They sign for a public origin and replay the request against the real backend through node:http with an explicit Host header — which fetch cannot do, since Host is a forbidden header name there and it would silently send the wrong one and prove nothing. Together they establish that a signature survives a reverse proxy that preserves Host (even across TLS termination, because the scheme is not signed), that it fails when a proxy rewrites Host, and that a URL signed for the internal endpoint cannot be rewritten to the public one afterwards — which is the entire reason the delivery origin is a signing input rather than a post-processing step.

An Announcements-shaped fixture in the same file models the real slideshow: it lists its own slides, filters images, builds a delivery URL for each, and fetches every one directly. It is written to demonstrate what it cannot say — its whole world is a ModuleStorage and two names from its own manifest — and it asserts that the module's durable state holds keys while only the URL carries host detail. It also proves that forty URLs cost no provider round trips.

Unit suites (core/tests/module-storage.test.ts, core/tests/module-storage-lifecycle.test.ts) use an injected in-memory provider and never touch the network. They prove the parts the protocol cannot: binding determinism and collision-safety, scoping and cross-module isolation, key validation, lazy provisioning and retry-after-failure, configuration validation and credential redaction, and the lifecycle integration including dependency propagation. They also own the delivery bounds, which need no server at all: every out-of-range lifetime refused rather than clamped, every download filename that could inject a response header refused, and the module-facing surface pinned so a new accessor cannot appear unnoticed. See ADR 0030 and ADR 0031.

Phase 7A pass 5 adds realtime evidence at every boundary (ADR 0032). Contract and Core unit suites prove channel grammar, module/permission scoping, staged registration, mandatory initial snapshots, atomic snapshot-plus-subscribe, sequence ordering, payload/container limits, latest-state coalescing, subscriber limits, exact shutdown behavior, per-module publisher deactivation, sanitized errors, and that modules declaring no realtime continue unchanged. Browser SDK/shell suites prove that the public contract has no server dependency or transport vocabulary, descriptors and local channels are strictly contained, credentials are host-owned, reconnect requires a fresh snapshot, terminal EventSource failure cleans up listeners, and explicit unmount is idempotent.

apps/api/tests/module-realtime-routes.test.ts exercises permission-before-allocation, the snapshot/update wire envelope, lack of replay ids, stream headers, comment keepalives, forced five-minute reauthorization through a short test-only override, shutdown closure, sanitized subscriber failures, and the explicit HEAD refusal that prevents an infinite discarded body from retaining a subscriber. apps/api/tests/module-realtime-live.test.ts binds an actual Node HTTP server on an OS-assigned loopback port and drives an Announcements-shaped module through real discovery, registration, startup, two live SSE clients, fanout, disconnect cleanup, reconnect with a fresh latest snapshot, and shutdown. Every bind, fetch, stream read, cancellation, cleanup poll, and listener close is bounded; the test also pins an opaque signed delivery URL byte-for-byte. It uses no PostgreSQL or object-store double because neither belongs in realtime transport.

The live PostgreSQL API suite adds lifecycle composition evidence: an enabled module's realtime route is live while disabled and dependency-blocked modules have no route, just like their HTTP/UI surfaces. Its real SSE fetch/read/cancel path is bounded and migrations remain independent of runtime enablement. During realtime development, run the focused live Node suite repeatedly as well as pnpm verify and pnpm test:migrations:integration; consecutive runs are specifically intended to catch retained timers, listeners, subscribers, or loopback servers.

Phase 7A pass 6 adds display evidence across the same boundaries (ADR 0033). Contract tests prove the manifest v6 hard cutover, optional role combinations, module-owned non-public display permission, and collision rejection. Discovery tests prove inert compiled-artifact validation, containment, symlink collision detection, deterministic diagnostics, and that modules without a display remain valid. Builder tests prove the exact src/ui/index.tsx / src/display/index.tsx sources, deterministic role order, preflight/compile atomicity, no runtime-entrypoint overwrite, and identical browser/CSS boundaries.

Core/API tests prove independently frozen primary and display publication records, exact anonymous artifact routing, permission-gated stable page and descriptor routes, eligible/running lifecycle filtering, sanitized failures, and startup-time frozen host assets. Surface tests prove principal- first descriptor bootstrap, the role-neutral host context, full-viewport mounting without Dashboard chrome or static module imports, late-import cleanup, manual retry, realtime reuse, and unmount. The canonical template builds and mounts both artifacts through parseModuleUiExport; contract fixtures also retain no-display and backend-only cases. These are code-only tests: display hosting adds no PostgreSQL, object-store, or Docker requirement beyond the existing suites.

A further suite (apps/api/tests/initial-owner.integration.ts, same harness) covers the initial platform-owner bootstrap (ADR 0035). It deliberately does not stop at asserting rows: a fresh installation is bootstrapped and the resulting account is then signed in through the exact browser-facing login transport, its principal read from GET /api/v1/principal, and used to open a platform-owner-protected resource that stays closed to an anonymous caller. Anything less would prove a user was written, not that an operator can reach their dashboard.

Every scenario runs with publicSignupEnabled: false — the production posture — so nothing can accidentally depend on the signup route a real deployment does not have, and one scenario asserts that a freshly bootstrapped deployment still refuses POST /api/auth/sign-up/email. The suite also pins the guard (a second run creates no owner, changes no password, and leaves no stray identity), the 8–128 password bound and email validation coming from the identity system rather than a CLI-specific policy, adoption of an existing identity without touching its password, refusal of a credential-less row without consuming the single-use sentinel, a partial failure that stays recoverable by rerunning, and five racing invocations yielding exactly one owner, one sentinel, and one user row.

The command itself is exercised as a real child process — success and failure exit codes, the password arriving on standard input and never in argv, --password refused, a missing DATABASE_URL, and an unmigrated database refused rather than silently migrated — with assertions that neither stdout nor stderr contains the password or the connection string.

A further suite (apps/api/tests/device-pairing.integration.ts, same harness) proves display device pairing end to end against real PostgreSQL and the real HTTP transport (ADR 0036). None of it is mocked: a screen with no identity is refused the descriptor but served the page it needs, asks to be paired, and receives a code; a real platform owner — created by the Pass 1B bootstrap and signed in through the real login transport — looks the code up and approves it by name; the screen collects its credential from a real Set-Cookie and then opens the real display descriptor and a real realtime stream; and the owner revokes it, after which both fail while the device still holds its cookie.

The suite also pins the properties that make the design safe rather than merely working: that the raw credential appears nowhere in the database or in any response body, that the permission is taken from the display declaration and not from a request that tried to name core.devices.manage, that someone who knows only the code (an onlooker reading a television) is issued nothing, that anonymous callers, a paired device, and a cross-origin administrator are all refused approval, that one code cannot be approved twice or claimed into two credentials under a race, that re-pairing produces a fresh identity while the old credential stays dead, and that a browser holding both a human session and a device cookie remains a person.

Device authority itself is proven without a database in core/tests/authorization-device.test.ts, which is mostly a list of things a device must not acquire: authority from being a principal, from an authenticated baseline, from platform ownership, or from a user grant. Its ports are wired to answer "yes" to every human question, so any accidental fall-through into a human path would flip those assertions.

Outbound email is proven against a real SMTP server rather than a mock (core/tests/mail-smtp.integration.test.ts, ADR 0037). smtp-server — the reference implementation from the same project as the client — runs in-process, so this needs no container and adds no CI infrastructure. No test sends real email, and no live credential exists in this repository or in CI.

It pins the things only a server can settle: a delivered message whose envelope sender and From: header are the host's configured address rather than anything a caller chose; both bodies and a reply-to arriving intact; authentication succeeding and its refusal mapping to a sanitized MAIL_REJECTED that echoes neither the password nor the server's response text; a mailbox refused at RCPT TO mapping to MAIL_RECIPIENT_INVALID rather than a generic failure; a refused connection mapping to MAIL_PROVIDER_UNAVAILABLE; a server that accepts a connection and never greets hitting its deadline instead of hanging; a starttls transport refusing to continue when the upgrade is not offered; verify() connecting without sending anything; and one send producing exactly one message.

core/tests/mail-contract.test.ts covers the boundary with no network at all, and is mostly negative — the contract's job is to make an impossible message impossible to construct. It also pins that a validated message has no from key, which is the structural reason no feature can spoof a sender. core/tests/mail-boundaries.test.ts asserts the properties no runtime test can: the SMTP dependency exists in Core alone, no browser-facing package imports it, module SDKs have no way to send, no HTTP route references the mail module, Core's mail code names no provider hostname and never mentions disabling certificate verification, and .env.example documents the variables with no password value.

Which kind of timing proof lives where

Pass 4 introduced several deadlines, and their tests deliberately differ in kind:

• Unit deadline tests (core/tests/module-runtime.test.ts) drive module stop and rollback hooks through a short, test-only hookTimeoutMs override with manually controlled never-settling promises — no real waiting, no sleeps, and an explicit unhandledRejection check proving a hook that rejects after its deadline stays handled.
• Real raw-HTTP force-close (apps/api/tests/api-http.integration.ts) binds the real listener, opens a real socket, writes a deliberately incomplete request, and proves shutdown force-closes it at a short drainTimeoutMsOverride and still reaches pool closure.
• Real TCP connection timeout (core/tests/postgres-migration.integration.ts) binds a loopback blackhole listener that accepts the socket and never completes PostgreSQL startup, proving the configured connection deadline; the listener and every accepted socket close in finally.
• Real PostgreSQL operation timeout and SQL batching (same file) use a bounded server-side delay only because the deadline itself is what is under test, and disposable databases to prove that marker bytes inside literals, dollar-quoted bodies, comments, and quoted identifiers execute intact, and that a failure rolls the whole batch back.

Every one of these uses a short test-only override; the production defaults (10 s per module hook, 30 s HTTP drain, 10 s migration connection/query/statement) are asserted separately as constants and are never waited out.

CI runs the same command in the dedicated migration-smoke job. After this change is merged, add the following pull-request status context to main branch protection:

CI / migration-smoke (pull_request)

Safe troubleshooting

Confirm the prerequisites without printing configuration:

docker info
docker compose version

Then rerun pnpm test:migrations:integration. The harness emits fixed failure messages and does not dump PostgreSQL logs, SQL, container inspection output, or the generated database URL. Do not enable query logging, print the test environment, run docker compose config without redaction, or paste Docker inspection output into an issue. If the Docker daemon itself stopped during cleanup, restore it before rerunning so the harness can create a fresh uniquely named project.