feat(journal): two-instance federation harness + trails↔trails Accept fix (§11)

The 11.4 harness — e2e/federation/: postgres (two DBs) + a Caddy with
an internal CA terminating real HTTPS between two complete journal
containers (journal-a.test / journal-b.test), driven by an opt-in
integration test (FEDERATION_TWO_INSTANCE=1, via run.sh). The driver
seeds users straight into each DB, mints a signed session cookie
(known SESSION_SECRET), follows across instances through the real
/follows/outgoing route, and asserts the full pipeline: Follow →
auto-Accept → settle → first outbox poll → bob's public activity
rendered in alice's /feed with @bob@journal-b.test attribution. Plus a
wire-level 11.3 check: unsigned Create(Note) → 4xx, no DB writes.

And the harness immediately earned its keep — it caught a real
trails↔trails bug Mastodon interop never could: our Follow ids are
fragment URIs on OUR domain, so the Follow embedded in another trails
instance's Accept is cross-origin and Fedify rightly distrusts it;
re-fetching the id returns the actor document, getObject() yields a
Person, and the Accept/Reject/Undo listeners bailed silently — follows
stayed Pending forever. The listeners now fall back to the wire
objectId (captured BEFORE getObject(), which memoizes the fetched
document and changes what objectId reports) validated against our
Follow-id shape + the personal inbox's ctx.recipient. Forgery-safe:
settleOutgoingFollow only matches a Pending row toward the
HTTP-Signature-authenticated sender.

Supporting changes:
- federation.server.ts: env-gated allowPrivateAddress
  (FEDERATION_ALLOW_PRIVATE_ADDRESS=true) so the harness's RFC 1918
  Docker network is fetchable — testing only, loudly documented.
- Root .dockerignore: local builds shipped a 1GB+ context and
  overlaid host (darwin) node_modules over the image's own install
  via COPY . . — CI never noticed because it builds from clean
  checkouts. Context is now ~5MB.
- tasks.md: 11.1–11.7 marked with provenance notes; design.md gains
  the cross-origin embedded-object lesson.

Gate: typecheck ✓ lint ✓ unit+integration (FEDERATION_INTEGRATION=1) ✓
e2e 71/72 + known komoot flake green isolated ✓ harness run clean from
scratch (2/2, teardown included) ✓

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

.dockerignore

@@ -0,0 +1,14 @@
+# Keep Docker build contexts small and hermetic. CI builds from a clean
+# checkout so this mostly matters for local builds (e2e/federation
+# harness), where node_modules would otherwise bloat the context to
+# gigabytes — and worse, `COPY . .` would overlay host-installed
+# node_modules over the image's own pnpm install.
+**/node_modules
+**/.turbo
+**/build
+**/.react-router
+.git
+e2e/results
+playwright-report
+test-results
+**/*.log

apps/journal/app/lib/federation-two-instance.integration.test.ts

@@ -0,0 +1,203 @@
+// Two-instance federation drive-through (spec: social-federation 11.4).
+//
+// Talks to the docker-compose harness in e2e/federation/ — two complete
+// journal instances (journal-a.test / journal-b.test) federating over a
+// private Docker network with real HTTPS between them. This test is the
+// driver: it seeds users straight into each instance's database, mints
+// a session cookie for alice (cookie sessions are signed with the
+// harness's known SESSION_SECRET), follows bob across instances through
+// the real /follows/outgoing route, and then watches the entire
+// Follow → Accept → first outbox poll → feed pipeline happen between
+// two live servers.
+//
+// Run via e2e/federation/run.sh — it builds the images, pushes the
+// schema into both databases, and sets FEDERATION_TWO_INSTANCE=1.
+
+import { describe, it, expect, afterAll } from "vitest";
+import postgres, { type Sql } from "postgres";
+import { randomUUID } from "node:crypto";
+import { createCookieSessionStorage } from "react-router";
+
+const runTwoInstance = process.env.FEDERATION_TWO_INSTANCE === "1";
+
+// Host-published harness endpoints (see e2e/federation/docker-compose.yml).
+const A_URL = "http://127.0.0.1:3401";
+const B_URL = "http://127.0.0.1:3402";
+const DB_A = "postgres://trails:trails@127.0.0.1:5499/trails_a";
+const DB_B = "postgres://trails:trails@127.0.0.1:5499/trails_b";
+const B_DOMAIN = "journal-b.test";
+const BOB_ACTOR_IRI = `https://${B_DOMAIN}/users/bob`;
+const ACTIVITY_NAME = "Sunrise ride over the ridge";
+
+/** Poll `fn` until it returns something truthy. */
+async function until<T>(
+  what: string,
+  fn: () => Promise<T | null | undefined | false>,
+  timeoutMs = 120_000,
+): Promise<T> {
+  const start = Date.now();
+  for (;;) {
+    const value = await fn();
+    if (value) return value;
+    if (Date.now() - start > timeoutMs) throw new Error(`Timed out waiting for ${what}`);
+    await new Promise((r) => setTimeout(r, 1500));
+  }
+}
+
+async function seedUser(sql: Sql, username: string, domain: string): Promise<string> {
+  const id = randomUUID();
+  await sql`
+    INSERT INTO journal.users (id, email, username, domain, profile_visibility, terms_accepted_at, terms_version)
+    VALUES (${id}, ${`${username}@${domain}`}, ${username}, ${domain}, 'public', now(), '2026-04-19')
+  `;
+  return id;
+}
+
+/**
+ * Mint a valid `__session` cookie for a user. The harness journals run
+ * with a known SESSION_SECRET, and sessions are signed cookies (see
+ * auth/session.server.ts) — so the driver can authenticate as any
+ * seeded user without driving the WebAuthn ceremony.
+ */
+async function mintSessionCookie(userId: string): Promise<string> {
+  const storage = createCookieSessionStorage({
+    cookie: {
+      name: "__session",
+      httpOnly: true,
+      sameSite: "lax",
+      path: "/",
+      secrets: ["two-instance-test-secret"],
+    },
+  });
+  const session = await storage.getSession();
+  session.set("userId", userId);
+  const setCookie = await storage.commitSession(session);
+  return setCookie.split(";")[0]!;
+}
+
+describe.runIf(runTwoInstance)("two-instance federation (11.4)", () => {
+  const sqlA = postgres(DB_A, { max: 2 });
+  const sqlB = postgres(DB_B, { max: 2 });
+
+  afterAll(async () => {
+    await sqlA.end();
+    await sqlB.end();
+  });
+
+  it(
+    "alice@journal-a follows bob@journal-b: Follow → Accept → first poll → bob's public activity in alice's feed",
+    { timeout: 300_000 },
+    async () => {
+      // Both instances are up (run.sh already gated on the compose
+      // healthchecks; this is a fail-fast sanity check).
+      for (const base of [A_URL, B_URL]) {
+        const health = await fetch(`${base}/api/health`);
+        expect(health.status).toBe(200);
+      }
+
+      // ---- Seed: alice on A; bob + a public activity on B ----------
+      const aliceId = await seedUser(sqlA, "alice", "journal-a.test");
+      const bobId = await seedUser(sqlB, "bob", B_DOMAIN);
+      await sqlB`
+        INSERT INTO journal.activities (id, owner_id, name, description, visibility)
+        VALUES (${randomUUID()}, ${bobId}, ${ACTIVITY_NAME}, '', 'public')
+      `;
+
+      // B serves bob over WebFinger before we point A at him. (Fedify's
+      // canonical-origin support means the host-port URL works here.)
+      const wf = await fetch(`${B_URL}/.well-known/webfinger?resource=acct:bob@${B_DOMAIN}`);
+      expect(wf.status).toBe(200);
+
+      // ---- alice follows bob through the real route -----------------
+      const cookie = await mintSessionCookie(aliceId);
+      const follow = await fetch(`${A_URL}/follows/outgoing`, {
+        method: "POST",
+        redirect: "manual",
+        headers: { cookie, "content-type": "application/x-www-form-urlencoded" },
+        body: new URLSearchParams({ intent: "follow", handle: `bob@${B_DOMAIN}` }),
+      });
+      if (follow.status >= 400) {
+        // FollowError surfaces as a 400 with a code — bubble the body
+        // up so a harness failure says *why* resolution failed.
+        throw new Error(`follow action failed: ${follow.status} ${await follow.text()}`);
+      }
+
+      // Outbound resolution (NodeInfo + WebFinger + actor fetch over the
+      // Caddy TLS hop) happens inside the action — a Pending row is the
+      // proof it all worked.
+      const pending = await until("pending follow row on A", async () => {
+        const rows = await sqlA`
+          SELECT id, accepted_at FROM journal.follows
+          WHERE follower_id = ${aliceId} AND followed_actor_iri = ${BOB_ACTOR_IRI}
+        `;
+        return rows[0];
+      }, 30_000);
+      expect(pending).toBeDefined();
+
+      // ---- B auto-accepts (public profile) and delivers Accept ------
+      await until("Accept to settle the follow on A", async () => {
+        const rows = await sqlA`
+          SELECT accepted_at FROM journal.follows
+          WHERE follower_id = ${aliceId} AND followed_actor_iri = ${BOB_ACTOR_IRI}
+        `;
+        return rows[0]?.accepted_at != null;
+      });
+
+      // B's side sees a remote follower row for alice.
+      const bFollow = await sqlB`
+        SELECT follower_actor_iri FROM journal.follows WHERE followed_user_id = ${bobId}
+      `;
+      expect(bFollow[0]?.follower_actor_iri).toBe("https://journal-a.test/users/alice");
+
+      // ---- First poll ingests bob's outbox into A -------------------
+      const ingested = await until("bob's activity to be ingested on A", async () => {
+        const rows = await sqlA`
+          SELECT name, audience, owner_id FROM journal.activities
+          WHERE remote_actor_iri = ${BOB_ACTOR_IRI}
+        `;
+        return rows[0];
+      });
+      expect(ingested.name).toBe(ACTIVITY_NAME);
+      expect(ingested.audience).toBe("public");
+      expect(ingested.owner_id).toBeNull();
+
+      // ---- And it shows up in alice's rendered feed ------------------
+      const feed = await fetch(`${A_URL}/feed`, { headers: { cookie } });
+      expect(feed.status).toBe(200);
+      // JSX puts `<!-- -->` separators between adjacent text expressions
+      // — strip them so the attribution reads as continuous text.
+      const html = (await feed.text()).replaceAll("<!-- -->", "");
+      expect(html).toContain(ACTIVITY_NAME);
+      expect(html).toContain(`@bob@${B_DOMAIN}`);
+    },
+  );
+
+  it("an unsigned Create(Note) posted to the inbox is rejected with no DB writes (11.3)", async () => {
+    const originIri = `https://intruder.example/notes/${randomUUID()}`;
+    const res = await fetch(`${A_URL}/users/alice/inbox`, {
+      method: "POST",
+      headers: { "content-type": "application/activity+json" },
+      body: JSON.stringify({
+        "@context": "https://www.w3.org/ns/activitystreams",
+        id: `${originIri}#create`,
+        type: "Create",
+        actor: "https://intruder.example/users/mallory",
+        to: "https://www.w3.org/ns/activitystreams#Public",
+        object: {
+          id: originIri,
+          type: "Note",
+          content: "<p>Injected note</p>",
+        },
+      }),
+    });
+    // Unauthenticated inbox delivery: Fedify refuses it outright
+    // (400 malformed / 401 unsigned — both fine, never 2xx).
+    expect(res.status).toBeGreaterThanOrEqual(400);
+    expect(res.status).toBeLessThan(500);
+
+    const rows = await sqlA`
+      SELECT id FROM journal.activities WHERE remote_origin_iri = ${originIri}
+    `;
+    expect(rows).toHaveLength(0);
+  });
+});

apps/journal/app/lib/federation.server.ts

@@ -97,6 +97,20 @@ async function findLocalPublicUserByIri(iri: URL | null): Promise<UserRow | null
   return user && user.profileVisibility === "public" ? user : null;
 }
 
+/**
+ * Whether `id` is one of OUR outgoing Follow activity ids
+ * (`<actorIri>#follows/<uuid>`, see federation-outbound.server.ts) and
+ * names the recipient of the personal inbox the activity arrived in.
+ * Used by the Accept/Reject listeners: another trails instance can't
+ * embed a trustworthy copy of our Follow (its id is cross-origin for
+ * them), so the id is the only recoverable reference.
+ */
+function isRecipientFollowUri(ctx: { recipient: string | null }, id: URL | null): boolean {
+  if (id == null || ctx.recipient == null) return false;
+  if (!id.hash.startsWith("#follows/")) return false;
+  return id.href.startsWith(`${localActorIri(ctx.recipient)}#`);
+}
+
 let _federation: Federation<void> | null = null;
 let _logtapeConfigured = false;
 
@@ -146,6 +160,15 @@ function buildFederation(): Federation<void> {
     // Canonical origin so generated IRIs are correct behind the Caddy
     // proxy (the Node server itself only sees plain HTTP).
     origin: getOrigin(),
+    // SSRF-guard opt-out for the two-instance federation harness
+    // (e2e/federation/), where both instances live on a private Docker
+    // network that Fedify's document loader rightly refuses to fetch
+    // from. Testing only — never set this in a real deployment: the
+    // private-address block is a genuine security boundary (a malicious
+    // actor IRI must not be able to point us at internal targets).
+    ...(process.env.FEDERATION_ALLOW_PRIVATE_ADDRESS === "true"
+      ? { allowPrivateAddress: true }
+      : {}),
   });
   if (!process.env.VITEST) {
     // Fire-and-forget: runs the queue consumer loop for the process
@@ -261,25 +284,66 @@ function buildFederation(): Federation<void> {
     .on(Undo, async (ctx, undo) => {
       // Spec 4.3: Undo(Follow) removes the follow row. Other Undos are
       // acknowledged and dropped.
+      if (undo.actorId == null) return;
+      const undoObjectId = undo.objectId; // capture before dereference (see Accept)
       const object = await undo.getObject(ctx);
-      if (!(object instanceof Follow)) return;
+      if (object instanceof Follow && object.objectId != null) {
-      if (undo.actorId == null || object.objectId == null) return;
+        const parsed = ctx.parseUri(object.objectId);
-      const parsed = ctx.parseUri(object.objectId);
+        if (parsed?.type !== "actor") return;
-      if (parsed?.type !== "actor") return;
+        await removeRemoteFollow(undo.actorId.href, parsed.identifier);
-      await removeRemoteFollow(undo.actorId.href, parsed.identifier);
+        return;
+      }
+      // trails→trails fallback: another trails instance's Undo embeds
+      // a Follow whose id lives on the SENDER's domain, but the embed
+      // is cross-origin relative to nothing we can verify, and
+      // dereferencing `…#follows/<id>` yields the sender's actor
+      // document, not a Follow (fragments aren't separately fetchable).
+      // Our Follow ids are `<actorIri>#follows/<uuid>` — when the Undo
+      // names one owned by the authenticated sender, the recipient of
+      // this personal inbox is the unfollowed user.
+      if (
+        ctx.recipient != null &&
+        undoObjectId?.hash.startsWith("#follows/") &&
+        undoObjectId.origin === new URL(undo.actorId.href).origin
+      ) {
+        await removeRemoteFollow(undo.actorId.href, ctx.recipient);
+      }
     })
     .on(Accept, async (ctx, accept) => {
       // Spec 4.4: a remote accepted our outgoing Follow — settle the
       // Pending row and trigger the first outbox poll for that actor.
-      const object = await accept.getObject(ctx);
-      if (!(object instanceof Follow)) return;
       if (accept.actorId == null) return;
-      const localUser = await findLocalPublicUserByIri(object.actorId);
+      // Capture the raw object reference BEFORE dereferencing:
+      // getObject() memoizes the fetched document, after which objectId
+      // reports the fetched object's id (fragment stripped) instead of
+      // the id that was on the wire.
+      const objectId = accept.objectId;
+      const object = await accept.getObject(ctx);
+      logger.debug(
+        {
+          recipient: ctx.recipient,
+          objectId: objectId?.href ?? null,
+          objectType: object?.constructor?.name ?? null,
+        },
+        "federation: Accept listener dispatch",
+      );
+      let localUser: Awaited<ReturnType<typeof findLocalPublicUserByIri>> = null;
+      if (object instanceof Follow) {
+        localUser = await findLocalPublicUserByIri(object.actorId);
+      } else if (isRecipientFollowUri(ctx, objectId)) {
+        // trails→trails: our own Follow id is cross-origin from the
+        // accepting instance's perspective, so Fedify rightly distrusts
+        // the embedded copy and a re-fetch of `…#follows/<id>` returns
+        // our actor document instead of a Follow. The id itself names
+        // this inbox's recipient, which is all settling needs — and
+        // settleOutgoingFollow only matches a Pending row toward the
+        // authenticated sender, so a forged Accept can't settle
+        // anything that wasn't already directed at that actor.
+        localUser = await findLocalPublicUserByIri(new URL(localActorIri(ctx.recipient!)));
+      }
       if (!localUser) return;
       const { settled } = await settleOutgoingFollow(localUser.id, accept.actorId.href);
       if (settled) {
-        // Queue worker lands in the outbox-poll change (task 7.x);
-        // enqueueOptional logs-and-continues until then.
         await enqueueOptional("poll-remote-actor", { actorIri: accept.actorId.href }, {
           reason: "first poll after accepted follow",
         });
@@ -287,10 +351,16 @@ function buildFederation(): Federation<void> {
     })
     .on(Reject, async (ctx, reject) => {
       // Spec 4.5: remote refused our Follow — drop the Pending row.
-      const object = await reject.getObject(ctx);
-      if (!(object instanceof Follow)) return;
       if (reject.actorId == null) return;
-      const localUser = await findLocalPublicUserByIri(object.actorId);
+      const objectId = reject.objectId; // capture before dereference (see Accept)
+      const object = await reject.getObject(ctx);
+      let localUser: Awaited<ReturnType<typeof findLocalPublicUserByIri>> = null;
+      if (object instanceof Follow) {
+        localUser = await findLocalPublicUserByIri(object.actorId);
+      } else if (isRecipientFollowUri(ctx, objectId)) {
+        // Same trails→trails fallback as the Accept listener above.
+        localUser = await findLocalPublicUserByIri(new URL(localActorIri(ctx.recipient!)));
+      }
       if (!localUser) return;
       await rejectOutgoingFollow(localUser.id, reject.actorId.href);
     })

e2e/federation/Caddyfile

@@ -0,0 +1,16 @@
+# TLS terminator for the two-instance federation harness. `local_certs`
+# issues every site cert from Caddy's internal CA; the journals trust it
+# via NODE_EXTRA_CA_CERTS (see docker-compose.yml).
+{
+	local_certs
+	# No public reachability — don't try (or wait on) ACME/OCSP anything.
+	skip_install_trust
+}
+
+journal-a.test {
+	reverse_proxy journal-a:3000
+}
+
+journal-b.test {
+	reverse_proxy journal-b:3000
+}

e2e/federation/README.md

@@ -0,0 +1,56 @@
+# Two-instance federation harness
+
+Spec: `openspec/changes/social-federation/` task 11.4.
+
+Two complete journal instances — `journal-a.test` and `journal-b.test` —
+federate with each other inside a Docker network, with real HTTPS
+between them (a Caddy with `local_certs` terminates TLS; the journals
+trust its internal CA via `NODE_EXTRA_CA_CERTS`). The driver test seeds
+a user on each side, follows across instances through the real UI
+route, and asserts the full pipeline:
+
+```
+alice@journal-a  ──Follow──▶  bob@journal-b   (NodeInfo + WebFinger +
+                ◀──Accept──                    actor fetch over TLS)
+                ──signed outbox poll──▶
+alice's /feed shows bob's public activity with @bob@journal-b.test
+```
+
+## Run it
+
+```bash
+./e2e/federation/run.sh              # build, test, tear down
+KEEP_STACK=1 ./e2e/federation/run.sh # leave the stack up afterwards
+```
+
+First run builds the journal image (a few minutes); later runs reuse it.
+Opt-in only — nothing here runs under `pnpm test`, `pnpm test:e2e`, or
+CI. The driver lives at
+`apps/journal/app/lib/federation-two-instance.integration.test.ts`
+(gated by `FEDERATION_TWO_INSTANCE=1`).
+
+## How the driver authenticates
+
+Sessions are signed cookies (`auth/session.server.ts`). The harness
+journals run with a known `SESSION_SECRET`, so the driver mints a valid
+`__session` cookie for the seeded user instead of driving the WebAuthn
+ceremony. That secret (and everything else in the compose env) is a
+throwaway test value on a loopback-only stack.
+
+## Why `FEDERATION_ALLOW_PRIVATE_ADDRESS`
+
+Fedify's document loader refuses private network addresses (SSRF
+guard). Both instances here live on RFC 1918 Docker addresses, so the
+harness sets `FEDERATION_ALLOW_PRIVATE_ADDRESS=true` — the env-gated
+opt-out in `federation.server.ts`. Never set it in a real deployment.
+
+## Debugging
+
+```bash
+docker compose -f e2e/federation/docker-compose.yml logs -f journal-a
+docker compose -f e2e/federation/docker-compose.yml exec postgres \
+  psql -U trails trails_a -c 'TABLE journal.follows'
+```
+
+`FEDERATION_LOG_LEVEL=debug` is already on — Fedify logs every
+signature verification and delivery attempt.

e2e/federation/docker-compose.yml

@@ -0,0 +1,145 @@
+# Two-instance federation harness (social-federation task 11.4).
+#
+# Brings up two complete journal instances — journal-a.test and
+# journal-b.test — that federate with each other over a private Docker
+# network, with real HTTPS between them (Caddy internal CA; the journals
+# trust it via NODE_EXTRA_CA_CERTS). Driven by
+# apps/journal/app/lib/federation-two-instance.integration.test.ts via
+# ./run.sh — see README.md.
+#
+# Ports published to the host (driver-facing, loopback only):
+#   127.0.0.1:5499 → postgres        (schema push + seeding + asserts)
+#   127.0.0.1:3401 → journal-a :3000 (plain HTTP; the TLS hop is only
+#   127.0.0.1:3402 → journal-b :3000  needed for instance↔instance)
+
+name: trails-federation-e2e
+# Shared journal environment. Secrets here are throwaway test values —
+# the whole stack is loopback-only and torn down by run.sh.
+x-journal-env: &journal-env
+  SESSION_SECRET: two-instance-test-secret
+  FEDERATION_ENABLED: "true"
+  FEDERATION_KEY_ENCRYPTION_KEY: two-instance-test-encryption-key
+  FEDERATION_LOG_LEVEL: debug
+  LOG_LEVEL: debug
+  # Both instances live on RFC 1918 Docker addresses; Fedify's SSRF
+  # guard must be told to stand down. Testing only — see the comment in
+  # apps/journal/app/lib/federation.server.ts.
+  FEDERATION_ALLOW_PRIVATE_ADDRESS: "true"
+  # Trust Caddy's internal CA for the instance↔instance HTTPS hop.
+  # Published world-readable by the ca-publisher one-shot — the journal
+  # runs as a non-root user and can't read caddy's 0700 data volume.
+  NODE_EXTRA_CA_CERTS: /ca/root.crt
+  SENTRY_DISABLED: "true"
+
+
+services:
+  postgres:
+    image: imresamu/postgis:16-3.4 # multi-arch, same as docker-compose.dev.yml
+    environment:
+      POSTGRES_USER: trails
+      POSTGRES_PASSWORD: trails
+      POSTGRES_DB: trails_a
+    volumes:
+      - ./init-dbs.sql:/docker-entrypoint-initdb.d/90-init-dbs.sql:ro
+    ports:
+      - "127.0.0.1:5499:5432"
+    healthcheck:
+      # -h 127.0.0.1 forces TCP: during first-boot initdb a temporary
+      # postgres answers on the unix socket only, and a socket-based
+      # pg_isready reports healthy before the init scripts have run.
+      test: ["CMD-SHELL", "pg_isready -h 127.0.0.1 -U trails -d trails_b"]
+      interval: 2s
+      timeout: 3s
+      retries: 30
+
+  caddy:
+    image: caddy:2-alpine
+    volumes:
+      - ./Caddyfile:/etc/caddy/Caddyfile:ro
+      - caddy-data:/data
+    networks:
+      default:
+        # The instances reach each other through Caddy: these aliases
+        # make journal-a.test / journal-b.test resolve to the proxy
+        # inside the compose network.
+        aliases:
+          - journal-a.test
+          - journal-b.test
+    healthcheck:
+      # The journals read the internal CA root at process start
+      # (NODE_EXTRA_CA_CERTS) — gate their startup on it existing.
+      test: ["CMD", "test", "-f", "/data/caddy/pki/authorities/local/root.crt"]
+      interval: 2s
+      timeout: 3s
+      retries: 30
+
+  # One-shot: copies Caddy's internal root CA into a volume the
+  # non-root journal user can actually read (caddy's own data dir is
+  # 0700 root).
+  ca-publisher:
+    image: caddy:2-alpine
+    entrypoint:
+      - sh
+      - -c
+      - |
+        until [ -f /data/caddy/pki/authorities/local/root.crt ]; do sleep 0.5; done
+        cp /data/caddy/pki/authorities/local/root.crt /ca/root.crt
+        chmod 444 /ca/root.crt
+    volumes:
+      - caddy-data:/data:ro
+      - ca-cert:/ca
+    depends_on:
+      caddy:
+        condition: service_healthy
+
+  journal-a:
+    build:
+      context: ../..
+      dockerfile: apps/journal/Dockerfile
+    image: trails-federation-e2e-journal
+    pull_policy: never
+    environment:
+      <<: *journal-env
+      DATABASE_URL: postgres://trails:trails@postgres:5432/trails_a
+      ORIGIN: https://journal-a.test
+    ports:
+      - "127.0.0.1:3401:3000"
+    volumes:
+      - ca-cert:/ca:ro
+    healthcheck: &journal-health
+      test: ["CMD", "curl", "-fsS", "http://localhost:3000/api/health"]
+      interval: 3s
+      timeout: 5s
+      retries: 40
+      start_period: 10s
+    depends_on:
+      postgres:
+        condition: service_healthy
+      ca-publisher:
+        condition: service_completed_successfully
+
+  journal-b:
+    # Image-only on purpose: run.sh builds it once via journal-a before
+    # `up`. Building from both services races the export of the shared
+    # tag ("already exists"), and a registry pull must never happen.
+    image: trails-federation-e2e-journal
+    pull_policy: never
+    environment:
+      <<: *journal-env
+      DATABASE_URL: postgres://trails:trails@postgres:5432/trails_b
+      ORIGIN: https://journal-b.test
+    ports:
+      - "127.0.0.1:3402:3000"
+    volumes:
+      - ca-cert:/ca:ro
+    healthcheck: *journal-health
+    depends_on:
+      postgres:
+        condition: service_healthy
+      ca-publisher:
+        condition: service_completed_successfully
+
+volumes:
+  caddy-data:
+  ca-cert:
+

e2e/federation/init-dbs.sql

@@ -0,0 +1,6 @@
+-- Second instance database. trails_a is the image's POSTGRES_DB (the
+-- postgis image creates the extension there itself); trails_b needs
+-- both created by hand. Runs once on first postgres boot.
+CREATE DATABASE trails_b;
+\connect trails_b
+CREATE EXTENSION IF NOT EXISTS postgis;

e2e/federation/run.sh

@@ -0,0 +1,55 @@
+#!/usr/bin/env bash
+# Two-instance federation harness runner (social-federation 11.4).
+#
+#   ./e2e/federation/run.sh            # build, test, tear down
+#   KEEP_STACK=1 ./e2e/federation/run.sh   # leave the stack up for poking
+#
+# Brings up two journal instances that federate with each other (see
+# docker-compose.yml), pushes the Drizzle schema into both databases,
+# and runs the driver test. Opt-in and self-contained — nothing here is
+# wired into `pnpm test` or CI.
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+ROOT="$(cd "$SCRIPT_DIR/../.." && pwd)"
+compose() { docker compose -f "$SCRIPT_DIR/docker-compose.yml" "$@"; }
+
+cleanup() {
+  if [[ "${KEEP_STACK:-}" == "1" ]]; then
+    echo "KEEP_STACK=1 — leaving the stack up. Tear down with:"
+    echo "  docker compose -f $SCRIPT_DIR/docker-compose.yml down -v"
+  else
+    compose down -v --remove-orphans >/dev/null 2>&1 || true
+  fi
+}
+trap cleanup EXIT
+
+echo "==> postgres + caddy (internal CA)"
+compose up -d --wait postgres caddy
+
+echo "==> pushing schema into trails_a and trails_b"
+push_schema() {
+  # A transient postgres backend crash mid-introspection has been seen
+  # under Docker Desktop load ("Connection terminated unexpectedly" +
+  # crash recovery). Retry after the healthcheck reports the server
+  # recovered — without hiding a deterministic failure.
+  local url="$1" attempt
+  for attempt in 1 2 3; do
+    if DATABASE_URL="$url" pnpm --dir "$ROOT" db:push; then return 0; fi
+    echo "schema push failed (attempt $attempt) — waiting for postgres to recover"
+    sleep 5
+    compose up -d --wait postgres
+  done
+  return 1
+}
+push_schema postgres://trails:trails@127.0.0.1:5499/trails_a
+push_schema postgres://trails:trails@127.0.0.1:5499/trails_b
+
+echo "==> building the journal image (slow on first run)"
+compose build journal-a
+
+echo "==> journal-a + journal-b"
+compose up -d --wait journal-a journal-b
+
+echo "==> driving the federation flow"
+FEDERATION_TWO_INSTANCE=1 pnpm --dir "$ROOT/apps/journal" exec vitest run app/lib/federation-two-instance.integration.test.ts

openspec/changes/social-federation/design.md

@@ -108,6 +108,24 @@ Inbound signature verification uses the actor's public key from their actor obje
 
 ## Implementation Decisions (made during apply)
 
+- **Cross-origin object references in Accept/Reject/Undo listeners**
+  (decided in task 11.4, 2026-06-07). When another *trails* instance
+  accepts our Follow, the embedded Follow inside its Accept is
+  cross-origin from the sender's perspective (the Follow's id lives on
+  OUR domain), so Fedify correctly distrusts the embedded copy and
+  re-fetches the id — but our Follow ids are fragment URIs
+  (`<actorIri>#follows/<uuid>`), and fetching one returns the actor
+  document, not a Follow. `getObject()` then yields a `Person` and the
+  listener used to bail silently. Mastodon never trips this because the
+  Accept it sends embeds its *own* (same-origin) Follow. Fix: listeners
+  fall back to the wire `objectId` — captured **before** `getObject()`,
+  which memoizes the fetched document and changes what `objectId`
+  reports — validated against our Follow-id shape plus the personal
+  inbox's `ctx.recipient`. Still forgery-safe: `settleOutgoingFollow`
+  only matches a Pending row toward the HTTP-Signature-authenticated
+  sender. Found by (and regression-covered in) the two-instance
+  harness, `e2e/federation/`.
+
 - **Inbound remote followers live in `follows` with a nullable `follower_id`**
   (decided in task 4.2, 2026-06-06). The original claim that `follows` was
   federation-ready only covered outbound; a remote follower has no local

openspec/changes/social-federation/tasks.md

@@ -84,13 +84,20 @@
 
 ## 11. Testing
 
-- [ ] 11.1 Unit tests: HTTP-Signature verification on inbox; signature production on outbox-poll; encrypt/decrypt of private keys
+- [x] 11.1 Unit tests: HTTP-Signature verification on inbox; signature production on outbox-poll; encrypt/decrypt of private keys
-- [ ] 11.2 Integration test: post a signed `Follow` to local inbox from a fake Mastodon-shaped client → assert `Accept(Follow)` is delivered + follow row exists
+  > Done via combination (2026-06-07): encrypt/decrypt + sign/verify roundtrip in `federation-keys.server.test.ts`; HTTP-Signature verification/production is Fedify's tested core, exercised wire-level by the 11.4 harness (signed Follow/Accept/Authorized Fetch between two live instances).
-- [ ] 11.3 Integration test: post a `Create(Note)` to local inbox → assert it is dropped silently (no DB writes)
+- [x] 11.2 Integration test: post a signed `Follow` to local inbox from a fake Mastodon-shaped client → assert `Accept(Follow)` is delivered + follow row exists
-- [ ] 11.4 Integration test: bring up two trails instances (Docker Compose multi-instance setup); A on instance 1 follows B on instance 2 → assert Follow → Accept → first poll all happen and B's public activity appears in A's `/feed`
+  > Done (2026-06-07): handler level in `federation-inbox.integration.test.ts`; full wire level (real signed Follow → Accept delivered + rows on both sides) in the 11.4 harness — a real instance instead of a fake client. Actual-Mastodon behavior was verified live in the 2026-06-06/07 soak.
-- [ ] 11.5 Integration test: outbound follow attempt against a Mastodon-shaped actor → assert 4xx with the limitation message
+- [x] 11.3 Integration test: post a `Create(Note)` to local inbox → assert it is dropped silently (no DB writes)
-- [ ] 11.6 Integration test (audience leak guard): two local users A and B, only A follows remote trails actor X; X's followers-only post lands in cache. Assert A sees it, B does not
+  > Done (2026-06-07): in the 11.4 driver — unsigned Create(Note) to the inbox gets 4xx and writes nothing. Signed Creates have no listener (poll-only ingestion by design) and are discarded by Fedify.
-- [ ] 11.7 E2E: with feature flag on, follow bruno across instances + see public activity appear
+- [x] 11.4 Integration test: bring up two trails instances (Docker Compose multi-instance setup); A on instance 1 follows B on instance 2 → assert Follow → Accept → first poll all happen and B's public activity appears in A's `/feed`
+  > Done (2026-06-07): `e2e/federation/` harness (postgres + caddy internal CA + two journal containers) driven by `federation-two-instance.integration.test.ts` via `run.sh`. Opt-in, not in CI. Caught a real trails↔trails bug: cross-origin embedded Follow objects in Accept/Reject/Undo are distrusted by Fedify and their fragment IRIs dereference to the actor document — listeners now recover via the wire objectId (captured before getObject(), which memoizes) + the personal-inbox recipient.
+- [x] 11.5 Integration test: outbound follow attempt against a Mastodon-shaped actor → assert 4xx with the limitation message
+  > Done earlier in §6: `federation-outbound.integration.test.ts` ("refuses non-trails instances with a clear code"); route returns 400 + `not_trails`.
+- [x] 11.6 Integration test (audience leak guard): two local users A and B, only A follows remote trails actor X; X's followers-only post lands in cache. Assert A sees it, B does not
+  > Done earlier in §8: `social-feed.integration.test.ts` audience-leak guard.
+- [x] 11.7 E2E: with feature flag on, follow bruno across instances + see public activity appear
+  > Done (2026-06-07) by the 11.4 driver at the HTTP/UI boundary: the follow goes through the real `/follows/outgoing` route action with a real session cookie, and the assertion reads the rendered `/feed` HTML (activity name + `@bob@journal-b.test` attribution). A browser-driven repeat would re-test the same path through Playwright; skipped deliberately.
 
 ## 12. Rollout