Merge pull request 'fix(web): remove prompt registry shortcut' (#57)

fix(web): remove prompt registry shortcut

.forgejo/workflows/dimensionlab-website.yml

@@ -0,0 +1,85 @@
+name: Dimension Lab website
+
+on:
+  pull_request:
+    types:
+      - opened
+      - synchronize
+      - reopened
+  push:
+    branches:
+      - main
+  workflow_dispatch:
+
+concurrency:
+  group: dimensionlab-website-${{ github.ref }}
+  cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
+
+jobs:
+  ci:
+    runs-on: docker
+    timeout-minutes: 30
+    steps:
+      - name: Checkout
+        uses: https://data.forgejo.org/actions/checkout@v4
+        with:
+          fetch-depth: 0
+          submodules: false
+
+      - name: Initialize submodules
+        run: |
+          git config --global url."https://git.dimensionlab.net/".insteadOf "ssh://git@git.dimensionlab.net/"
+          git submodule update --init --recursive
+
+      - name: Install Bun
+        run: |
+          curl -fsSL https://bun.sh/install | bash -s "bun-v1.3.14"
+          "$HOME/.bun/bin/bun" --version
+
+      - name: Check, test, and build
+        run: |
+          export BUN_INSTALL="$HOME/.bun"
+          export PATH="$BUN_INSTALL/bin:$PATH"
+          bun install --frozen-lockfile
+          bun run check
+          bun run test
+          bun run build
+
+  deploy:
+    needs: ci
+    if: github.event_name == 'push' && github.ref == 'refs/heads/main'
+    runs-on: deploy
+    timeout-minutes: 30
+    steps:
+      - name: Checkout
+        run: |
+          if [ -d .git ]; then
+            git remote set-url origin git@git.dimensionlab.net:vince/dimensionlab-website.git
+          else
+            git init
+            git remote add origin git@git.dimensionlab.net:vince/dimensionlab-website.git
+          fi
+          git fetch --force --prune --depth=1 origin "$GITHUB_SHA"
+          git checkout --force --detach "$GITHUB_SHA"
+          git clean -ffdx
+
+      - name: Initialize submodules
+        run: |
+          git config --global url."https://git.dimensionlab.net/".insteadOf "ssh://git@git.dimensionlab.net/"
+          git submodule update --init --recursive
+
+      - name: Verify Podman deployment socket
+        run: |
+          command -v podman
+          command -v systemctl
+          unit="$(timeout 15s podman inspect dimensionlab-website --format '{{ index .Config.Labels "PODMAN_SYSTEMD_UNIT" }}')"
+          test "$unit" = "dimensionlab-website.service"
+
+      - name: Deploy production website
+        env:
+          DEPLOY_CONTAINER_CLI: podman
+          DEPLOY_EVENT_NAME: ${{ github.event_name }}
+          DEPLOY_REF: ${{ github.ref }}
+          DEPLOY_RESTART_STRATEGY: quadlet-container
+          DEPLOY_SHA: ${{ github.sha }}
+        run: scripts/deploy-dimensionlab-website.sh

README.md

@@ -161,3 +161,27 @@ The env file must provide `AGENT_CONFIG_TOKEN`. Runtime defaults inside the
 image set `HOST=0.0.0.0`, `PORT=3000`,
 `DATABASE_URL=file:/data/dimensionlab.sqlite`, and
 `DASHBOARD_MIGRATIONS_DIR=/repo/apps/web/drizzle`.
+
+### Forgejo Actions Deployment
+
+Merges to `main` run `.forgejo/workflows/dimensionlab-website.yml`. Pull
+requests run check, test, and build only; the deploy job is guarded to run only
+for `push` events on `refs/heads/main`.
+
+The workflow uses two runner classes. Pull request CI runs on the containerized
+`docker` runner. Production deployment runs on a separate host runner with the
+`deploy:host` label so the guarded deploy script can use the user's rootless
+`podman` and `systemctl --user` commands directly. The deploy job uses a
+shell-only `git fetch` checkout so the host runner does not need a Node runtime
+for checkout actions.
+
+```yaml
+runner:
+  labels:
+    - deploy:host
+```
+
+The deploy job also performs a host preflight against the
+`dimensionlab-website.service` Podman label before it builds or restarts the
+production container. Do not give the general pull request runner deployment
+socket access; keep deploy privileges on the dedicated `deploy` runner.

apps/web/package.json

@@ -8,8 +8,8 @@
     "build": "rm -rf build && vite build && bun build src/server/index.ts --target bun --outdir build",
     "preview": "HOST=0.0.0.0 PORT=4173 bun build/index.js",
     "check": "tsc --noEmit",
-    "test": "vitest run",
+    "test": "bun --bun vitest run",
-    "test:unit": "vitest run",
+    "test:unit": "bun --bun vitest run",
     "test:e2e": "env -u NO_COLOR playwright test",
     "db:generate": "drizzle-kit generate",
     "db:check": "drizzle-kit check"

apps/web/src/App.test.tsx

@@ -161,6 +161,68 @@ describe("React app dashboard state view", () => {
     });
   });
 
+  test("does not restore aggregate health snapshots over the fresh shell", () => {
+    const restored = restoreDashboardTileSnapshots(
+      {
+        state: "ready",
+        document: dimensionLabDashboardFixture,
+        schemaVersion: "dashboard.v1",
+        currentRevisionId: "revision-a",
+      },
+      [
+        {
+          ageMs: 60_000,
+          response: {
+            state: "ready",
+            tile: { kind: "status", stripId: "footer-status", id: "system-status" },
+            item: {
+              id: "system-status",
+              label: "System Status",
+              value: "20 services down",
+              severity: "stale",
+            },
+          },
+        },
+        {
+          ageMs: 60_000,
+          response: {
+            state: "ready",
+            tile: { kind: "module", id: "runtime-health-summary" },
+            item: {
+              id: "runtime-health-summary",
+              kind: "summary",
+              title: "Runtime Health",
+              value: "20 services down",
+              detail: "0 warnings - 8 services ok - stale 60s",
+              severity: "stale",
+            },
+          },
+        },
+      ],
+    );
+
+    expect(restored.restoredItemIds).toEqual(new Set());
+    if (restored.dashboard.state !== "ready") {
+      throw new Error("Expected dashboard to be ready");
+    }
+    expect(
+      restored.dashboard.document.statusStrips[0].items.find((item) =>
+        item.id === "system-status"
+      ),
+    ).toMatchObject({
+      id: "system-status",
+      value: "Fallback operational",
+    });
+    expect(
+      restored.dashboard.document.modules?.find((module) =>
+        module.id === "runtime-health-summary"
+      ),
+    ).toMatchObject({
+      id: "runtime-health-summary",
+      value: "fallback",
+    });
+  });
+
   test("uses structured tile match keys for delimiter-bearing ids", () => {
     expect(
       dashboardTileMatchKey({ kind: "service", groupId: "a:b", id: "c" }),

apps/web/src/App.tsx

@@ -15,8 +15,10 @@ import {
   createDashboardTileSnapshotStore,
   createDashboardTileBackoff,
   createDashboardRequestAborter,
+  isPersistableDashboardTileSnapshot,
   runViewportAwareDashboardHydrationQueue,
   shouldPauseDashboardRefresh,
+  subscribeToDashboardTileEvents,
   waitForDashboardHydrationIdle,
   type DashboardIntersectionObserverFactory,
   type DashboardTileReference,
@@ -177,6 +179,7 @@ export default function App() {
   useEffect(() => {
     let cancelled = false;
     let refreshTimer: number | undefined;
+    let unsubscribeTileEvents: (() => void) | undefined;
     let lastRefreshIntervalMs = dashboardFallbackRefreshIntervalMs;
     let hydrationRun = 0;
     const requestAborter = createDashboardRequestAborter();
@@ -213,6 +216,8 @@ export default function App() {
 
     function pauseRefreshes() {
       clearRefreshTimer();
+      unsubscribeTileEvents?.();
+      unsubscribeTileEvents = undefined;
       requestAborter.abortActiveRequests();
       setHydratingItemIds(new Set());
     }
@@ -266,6 +271,14 @@ export default function App() {
               schemaVersion: restored.dashboard.schemaVersion,
             },
           );
+          subscribeDashboardTileEvents(
+            currentRun,
+            tileSignal,
+            {
+              currentRevisionId: restored.dashboard.currentRevisionId,
+              schemaVersion: restored.dashboard.schemaVersion,
+            },
+          );
         } else {
           setHydratingItemIds(new Set());
         }
@@ -358,6 +371,30 @@ export default function App() {
       });
     }
 
+    function subscribeDashboardTileEvents(
+      run: number,
+      signal: AbortSignal,
+      snapshotContext: DashboardTileSnapshotStoreContext,
+    ) {
+      unsubscribeTileEvents?.();
+      unsubscribeTileEvents = subscribeToDashboardTileEvents({
+        onTile: (event) => {
+          const tileResponse = event as DashboardTileResponse;
+          if (!isDashboardTileResponse(tileResponse)) return;
+          applyDashboardTileHydrationResponse(
+            tileResponse.tile,
+            tileResponse,
+            run,
+            signal,
+            snapshotContext,
+          );
+        },
+        onUnavailable: () => {
+          unsubscribeTileEvents = undefined;
+        },
+      });
+    }
+
     async function hydrateDashboardTileBatch(
       tiles: DashboardTileReference[],
       run: number,
@@ -526,6 +563,7 @@ export default function App() {
     return () => {
       cancelled = true;
       clearRefreshTimer();
+      unsubscribeTileEvents?.();
       requestAborter.abortActiveRequests();
       detachRefreshLifecycle();
     };
@@ -625,6 +663,20 @@ function isAbortError(error: unknown): boolean {
   );
 }
 
+function isDashboardTileResponse(value: unknown): value is DashboardTileResponse {
+  return (
+    typeof value === "object" &&
+    value !== null &&
+    "state" in value &&
+    (value.state === "ready" ||
+      value.state === "not_found" ||
+      value.state === "disabled") &&
+    "tile" in value &&
+    typeof value.tile === "object" &&
+    value.tile !== null
+  );
+}
+
 export function restoreDashboardTileSnapshots(
   dashboard: DashboardRuntimeState,
   snapshots: RestoredDashboardTileSnapshot[],
@@ -639,7 +691,9 @@ export function restoreDashboardTileSnapshots(
     };
   }
 
-  return snapshots.reduce(
+  return snapshots.filter((snapshot) =>
+    isPersistableDashboardTileSnapshot(snapshot.response.tile)
+  ).reduce(
     (current, snapshot) => ({
       dashboard: {
         ...current.dashboard,

apps/web/src/lib/client/dashboard-refresh.test.ts

@@ -13,6 +13,7 @@ import {
   runViewportAwareDashboardHydrationQueue,
   shouldPauseDashboardRefresh,
   splitDashboardHydrationItemsByVisibility,
+  subscribeToDashboardTileEvents,
   waitForDashboardHydrationIdle,
   type DashboardIntersectionEntry,
 } from "./dashboard-refresh";
@@ -302,6 +303,69 @@ describe("dashboard refresh lifecycle", () => {
     ]);
   });
 
+  test("subscribes to dashboard tile events", () => {
+    const received: unknown[] = [];
+    let listener: ((event: MessageEvent<string>) => void) | undefined;
+    let closed = false;
+
+    const unsubscribe = subscribeToDashboardTileEvents({
+      createEventSource: (url) => {
+        expect(url).toBe("/api/dashboard/events");
+        return {
+          addEventListener(_type, eventListener) {
+            listener = eventListener;
+          },
+          close() {
+            closed = true;
+          },
+          onerror: null,
+        };
+      },
+      onTile: (tile) => received.push(tile),
+    });
+
+    listener?.({ data: JSON.stringify({ state: "ready" }) } as MessageEvent<string>);
+    expect(received).toEqual([{ state: "ready" }]);
+    unsubscribe();
+    expect(closed).toBe(true);
+  });
+
+  test("falls back when dashboard tile events are unavailable or fail", () => {
+    let unavailableCount = 0;
+
+    subscribeToDashboardTileEvents({
+      createEventSource: undefined,
+      onTile: () => undefined,
+      onUnavailable: () => {
+        unavailableCount += 1;
+      },
+    });
+
+    let errorHandler: (() => void) | null = null;
+    const unsubscribe = subscribeToDashboardTileEvents({
+      createEventSource: () => ({
+        addEventListener: () => undefined,
+        close: () => undefined,
+        get onerror() {
+          return errorHandler;
+        },
+        set onerror(handler) {
+          errorHandler = handler;
+        },
+      }),
+      onTile: () => undefined,
+      onUnavailable: () => {
+        unavailableCount += 1;
+      },
+    });
+
+    if (!errorHandler) throw new Error("expected error handler");
+    const triggerError = errorHandler as unknown as () => void;
+    triggerError();
+    unsubscribe();
+    expect(unavailableCount).toBe(2);
+  });
+
   test("backs off failed tile keys and resets after success", () => {
     const backoff = createDashboardTileBackoff();
 
@@ -454,6 +518,77 @@ describe("dashboard refresh lifecycle", () => {
       },
     ]);
   });
+
+  test("ignores restored aggregate health snapshots", () => {
+    const storage = createMemoryStorage();
+    storage.setItem(
+      "dimensionlab.dashboard.tiles.v1",
+      JSON.stringify({
+        currentRevisionId: "revision-a",
+        schemaVersion: "dashboard.v1",
+        tiles: [
+          {
+            item: {
+              id: "system-status",
+              label: "System Status",
+              value: "20 services down",
+              severity: "danger",
+            },
+            savedAt: 1_000,
+            tile: { kind: "status", stripId: "footer-status", id: "system-status" },
+          },
+          {
+            item: {
+              id: "runtime-health-summary",
+              kind: "summary",
+              title: "Runtime Health",
+              value: "20 services down",
+              detail: "0 warnings - 8 services ok",
+              severity: "danger",
+            },
+            savedAt: 1_000,
+            tile: { kind: "module", id: "runtime-health-summary" },
+          },
+          {
+            item: {
+              id: "infra-ram",
+              label: "Infra RAM",
+              value: { kind: "percent", value: 42 },
+              detail: "live",
+              severity: "ok",
+            },
+            savedAt: 1_000,
+            tile: { kind: "telemetry", id: "infra-ram" },
+          },
+        ],
+        version: 1,
+      }),
+    );
+
+    const store = createDashboardTileSnapshotStore(storage, {
+      now: () => 16_000,
+    });
+
+    expect(store.restore({
+      currentRevisionId: "revision-a",
+      schemaVersion: "dashboard.v1",
+    })).toEqual([
+      {
+        ageMs: 15_000,
+        response: {
+          state: "ready",
+          tile: { kind: "telemetry", id: "infra-ram" },
+          item: {
+            id: "infra-ram",
+            label: "Infra RAM",
+            value: { kind: "percent", value: 42 },
+            detail: "live - stale 15s",
+            severity: "stale",
+          },
+        },
+      },
+    ]);
+  });
 });
 
 async function waitFor(predicate: () => boolean) {

apps/web/src/lib/client/dashboard-refresh.ts

@@ -489,6 +489,51 @@ export function createDashboardPerformanceMarks(
   };
 }
 
+export interface DashboardTileEventSource {
+  addEventListener(
+    type: "dashboard-tile",
+    listener: (event: MessageEvent<string>) => void,
+  ): void;
+  close(): void;
+  onerror: (() => void) | null;
+}
+
+export interface DashboardTileEventSubscriptionOptions {
+  createEventSource?: (url: string) => DashboardTileEventSource;
+  onTile: (data: unknown) => void;
+  onUnavailable?: () => void;
+  url?: string;
+}
+
+export function subscribeToDashboardTileEvents(
+  options: DashboardTileEventSubscriptionOptions,
+): () => void {
+  const createEventSource = options.createEventSource ||
+    (typeof globalThis.EventSource !== "undefined"
+      ? (url: string) => new globalThis.EventSource(url)
+      : undefined);
+
+  if (!createEventSource) {
+    options.onUnavailable?.();
+    return () => undefined;
+  }
+
+  const source = createEventSource(options.url || "/api/dashboard/events");
+  source.addEventListener("dashboard-tile", (event) => {
+    try {
+      options.onTile(JSON.parse(event.data));
+    } catch {
+      // Ignore malformed diagnostics from an optional live transport.
+    }
+  });
+  source.onerror = () => {
+    source.close();
+    options.onUnavailable?.();
+  };
+
+  return () => source.close();
+}
+
 const dashboardTileBackoffDelaysMs = [15_000, 30_000, 60_000, 120_000];
 
 export function createDashboardTileBackoff() {
@@ -572,7 +617,9 @@ export function createDashboardTileSnapshotStore(
       return {
         currentRevisionId: payload.currentRevisionId,
         schemaVersion: payload.schemaVersion,
-        tiles: payload.tiles.filter(isDashboardTileSnapshotRecord),
+        tiles: payload.tiles
+          .filter(isDashboardTileSnapshotRecord)
+          .filter((record) => isPersistableDashboardTileSnapshot(record.tile)),
         version: 1,
       };
     } catch {
@@ -615,6 +662,8 @@ export function createDashboardTileSnapshotStore(
       response: Extract<DashboardTileSnapshotResponse, { state: "ready" }>;
     },
   ): void {
+    if (!isPersistableDashboardTileSnapshot(input.response.tile)) return;
+
     const payload = matchingPayload(input);
     const key = dashboardTileSnapshotKey(input.response.tile);
     const nextRecord: DashboardTileSnapshotRecord = {
@@ -675,6 +724,14 @@ function dashboardTileSnapshotKey(tile: DashboardTileReference): string {
   return JSON.stringify(tile);
 }
 
+export function isPersistableDashboardTileSnapshot(
+  tile: DashboardTileReference,
+): boolean {
+  if (tile.kind === "status" && tile.id === "system-status") return false;
+  if (tile.kind === "module" && tile.id === "runtime-health-summary") return false;
+  return true;
+}
+
 function isDashboardTileSnapshotRecord(
   value: unknown,
 ): value is DashboardTileSnapshotRecord {

apps/web/src/lib/dashboard-seed/dimensionlab.test.ts

@@ -128,7 +128,6 @@ const verifiedSeedIconIds = new Set([
   "mdi:pulse",
   "mdi:robot-outline",
   "mdi:router-network",
-  "mdi:text-box-search",
   "mdi:thermometer",
   "mdi:web",
   "mdi:weather-sunny",

apps/web/src/lib/dashboard-seed/dimensionlab.ts

@@ -431,14 +431,6 @@ export const dimensionLabDashboardFixture: DashboardDocument = {
         href: "https://models.dimensionlab.net",
         datasource: uptimeMonitor(7),
       }),
-      service({
-        id: "prompt-registry",
-        label: "Prompt Registry",
-        description: "Shared prompts, traces, evals",
-        icon: "mdi:text-box-search",
-        href: "https://prompts.dimensionlab.net",
-        datasource: uptimeMonitor(20),
-      }),
     ]),
     group("systems", "Systems", [
       service({

apps/web/src/lib/server/dashboard.test.ts

@@ -66,9 +66,6 @@ describe("dashboard runtime loader", () => {
     expect(runtime.document.statusStrips[0]?.items.map((item) => item.id)).toContain(
       "auto-refresh",
     );
-    expect(runtime.document.serviceGroups.flatMap((group) => group.services).map((service) => service.id)).toContain(
-      "prompt-registry",
-    );
     expect(store.listRevisions()).toHaveLength(2);
     expect(store.getActiveDashboard()?.revision.actor).toBe("initial-seed");
   });
@@ -138,13 +135,5 @@ function olderDimensionLabSeed() {
     ...strip,
     items: strip.items.filter((item) => item.id !== "auto-refresh"),
   }));
-  document.serviceGroups = document.serviceGroups.map((group) =>
-    group.id === "ai-automation"
-      ? {
-          ...group,
-          services: group.services.filter((service) => service.id !== "prompt-registry"),
-        }
-      : group,
-  );
   return document;
 }

apps/web/src/lib/workspace-boundary.test.ts

@@ -1,4 +1,6 @@
-import { existsSync, readFileSync } from "node:fs";
+import { spawnSync } from "node:child_process";
+import { chmodSync, existsSync, mkdtempSync, readFileSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
 import { join } from "node:path";
 import { describe, expect, test } from "vitest";
 
@@ -42,6 +44,8 @@ describe("workspace boundaries", () => {
       "turbo run check test:unit build @dimensionlab/ui#build-storybook @dimensionlab/web#test:e2e",
     );
     expect(webPackage.scripts).not.toHaveProperty("test:qa");
+    expect(webPackage.scripts?.test).toBe("bun --bun vitest run");
+    expect(webPackage.scripts?.["test:unit"]).toBe("bun --bun vitest run");
     expect(turboConfig.tasks).not.toHaveProperty("test:qa");
     expect(turboConfig.globalDependencies).toEqual(
       expect.arrayContaining(["bun.lock", "tsconfig.base.json"]),
@@ -253,6 +257,186 @@ describe("workspace boundaries", () => {
       "rm -rf build && vite build && bun build src/server/index.ts --target bun --outdir build",
     );
   });
+
+  test("defines Forgejo CI and main-branch deploy automation", () => {
+    const workflow = readFileSync(
+      join(root, ".forgejo/workflows/dimensionlab-website.yml"),
+      "utf8",
+    );
+
+    expect(workflow).toContain("name: Dimension Lab website");
+    expect(workflow).toContain("pull_request:");
+    expect(workflow).toContain("push:");
+    expect(workflow).toContain("branches:");
+    expect(workflow).toContain("- main");
+    expect(workflow).toContain("runs-on: docker");
+    expect(workflow).toContain("bun install --frozen-lockfile");
+    expect(workflow).toContain("bun run check");
+    expect(workflow).toContain("bun run test");
+    expect(workflow).toContain("bun run build");
+    expect(workflow).toContain("needs: ci");
+    expect(workflow).toContain("runs-on: deploy");
+    expect(workflow).toContain("github.event_name == 'push'");
+    expect(workflow).toContain("github.ref == 'refs/heads/main'");
+    expect(workflow).toContain(
+      "git remote add origin git@git.dimensionlab.net:vince/dimensionlab-website.git",
+    );
+    expect(workflow).toContain('git fetch --force --prune --depth=1 origin "$GITHUB_SHA"');
+    expect(workflow).toContain("podman inspect dimensionlab-website");
+    expect(workflow).toContain("DEPLOY_CONTAINER_CLI: podman");
+    expect(workflow).toContain("PODMAN_SYSTEMD_UNIT");
+    expect(workflow).toContain("scripts/deploy-dimensionlab-website.sh");
+  });
+
+  test("keeps production deployment behind a guarded script", () => {
+    const deployScript = readFileSync(
+      join(root, "scripts/deploy-dimensionlab-website.sh"),
+      "utf8",
+    );
+
+    expect(deployScript).toContain("refs/heads/main");
+    expect(deployScript).toContain("dimensionlab-website.service");
+    expect(deployScript).toContain("localhost/dimensionlab-website");
+    expect(deployScript).toContain("apps/web/Containerfile");
+    expect(deployScript).toContain("rollback-");
+    expect(deployScript).toContain("https://dimensionlab.net");
+    expect(deployScript).toContain("/api/dashboard/tiles");
+    expect(deployScript).toContain("--dry-run");
+    expect(deployScript).toContain("DEPLOY_RESTART_STRATEGY");
+  });
+
+  test("rolls back the latest image when production smoke checks fail", () => {
+    const result = runDeployScriptWithFakes(
+      {
+        curl: failingCurlCommand,
+        git: fakeGitCommand,
+        podman: fakePodmanCommand("dimensionlab-website.service"),
+      },
+      {
+        DEPLOY_CONTAINER_CLI: "podman",
+        DEPLOY_EVENT_NAME: "push",
+        DEPLOY_REF: "refs/heads/main",
+        DEPLOY_RESTART_STRATEGY: "quadlet-container",
+        DEPLOY_SHA: "1234567890abcdef",
+        DEPLOY_SMOKE_TIMEOUT_SECONDS: "0",
+      },
+    );
+
+    expect(result.status).toBe(1);
+    expect(result.stderr).toContain("smoke checks failed");
+    expect(result.stderr).toContain("rolling back to localhost/dimensionlab-website:rollback-");
+    expect(result.log).toMatch(
+      /tag localhost\/dimensionlab-website:latest localhost\/dimensionlab-website:rollback-\d{14}/,
+    );
+    expect(result.log).toMatch(
+      /tag localhost\/dimensionlab-website:rollback-\d{14} localhost\/dimensionlab-website:latest/,
+    );
+    expect(result.log.match(/^stop dimensionlab-website$/gm)).toHaveLength(2);
+  });
+
+  test("rolls back the latest image when the container fails to restart", () => {
+    const result = runDeployScriptWithFakes(
+      {
+        curl: passingCurlCommand,
+        git: fakeGitCommand,
+        podman: fakePodmanCommand("dimensionlab-website.service"),
+      },
+      {
+        DEPLOY_CONTAINER_CLI: "podman",
+        DEPLOY_CONTAINER_START_TIMEOUT_SECONDS: "1",
+        DEPLOY_TEST_CONTAINER_IMAGE: "wrong",
+        DEPLOY_EVENT_NAME: "push",
+        DEPLOY_REF: "refs/heads/main",
+        DEPLOY_RESTART_STRATEGY: "quadlet-container",
+        DEPLOY_SHA: "1234567890abcdef",
+      },
+    );
+
+    expect(result.status).toBe(1);
+    expect(result.stderr).toContain("did not restart on localhost/dimensionlab-website:latest");
+    expect(result.stderr).toContain("rolling back to localhost/dimensionlab-website:rollback-");
+    expect(result.stderr).toContain("rollback image is running");
+    expect(result.log).toMatch(
+      /tag localhost\/dimensionlab-website:rollback-\d{14} localhost\/dimensionlab-website:latest/,
+    );
+    expect(result.log.match(/^stop dimensionlab-website$/gm)).toHaveLength(2);
+  });
+
+  test.each([
+    {
+      env: { DEPLOY_EVENT_NAME: "pull_request", DEPLOY_REF: "refs/heads/main" },
+      message: "production deploys only run for push",
+    },
+    {
+      env: { DEPLOY_EVENT_NAME: "push", DEPLOY_REF: "refs/heads/codex/test" },
+      message: "expected refs/heads/main",
+    },
+  ])("refuses guarded deploy contexts before host mutations", ({ env, message }) => {
+    const result = runDeployScriptWithFakes(
+      {
+        curl: passingCurlCommand,
+        git: fakeGitCommand,
+        podman: fakePodmanCommand("dimensionlab-website.service"),
+      },
+      {
+        DEPLOY_CONTAINER_CLI: "podman",
+        DEPLOY_SHA: "1234567890abcdef",
+        ...env,
+      },
+    );
+
+    expect(result.status).toBe(1);
+    expect(result.stderr).toContain(message);
+    expect(result.log).toBe("");
+  });
+
+  test("refuses stop-based deploys unless the container belongs to the expected unit", () => {
+    const result = runDeployScriptWithFakes(
+      {
+        curl: passingCurlCommand,
+        git: fakeGitCommand,
+        podman: fakePodmanCommand("other.service"),
+      },
+      {
+        DEPLOY_CONTAINER_CLI: "podman",
+        DEPLOY_EVENT_NAME: "push",
+        DEPLOY_REF: "refs/heads/main",
+        DEPLOY_RESTART_STRATEGY: "quadlet-container",
+        DEPLOY_SHA: "1234567890abcdef",
+      },
+    );
+
+    expect(result.status).toBe(1);
+    expect(result.stderr).toContain(
+      "refusing to stop dimensionlab-website; expected PODMAN_SYSTEMD_UNIT=dimensionlab-website.service",
+    );
+    expect(result.log).not.toContain("build ");
+    expect(result.log).not.toContain("stop dimensionlab-website");
+  });
+
+  test("checks the expected unit before auto falls back to stopping the container", () => {
+    const result = runDeployScriptWithFakes(
+      {
+        curl: passingCurlCommand,
+        git: fakeGitCommand,
+        podman: fakePodmanCommand("other.service"),
+        systemctl: fakeSystemctlCommand({ active: false, show: true }),
+      },
+      {
+        DEPLOY_CONTAINER_CLI: "podman",
+        DEPLOY_EVENT_NAME: "push",
+        DEPLOY_REF: "refs/heads/main",
+        DEPLOY_RESTART_STRATEGY: "auto",
+        DEPLOY_SHA: "1234567890abcdef",
+      },
+    );
+
+    expect(result.status).toBe(1);
+    expect(result.stderr).toContain(
+      "refusing to stop dimensionlab-website; expected PODMAN_SYSTEMD_UNIT=dimensionlab-website.service",
+    );
+    expect(result.log).not.toContain("stop dimensionlab-website");
+  });
 });
 
 type WorkspacePackageExport =
@@ -269,3 +453,126 @@ function expectPackageExport(
 ): void {
   expect(actual).toMatchObject(expected);
 }
+
+function runDeployScriptWithFakes(
+  commands: Record<string, string>,
+  env: Record<string, string>,
+): { log: string; status: number | null; stderr: string; stdout: string } {
+  const tempDir = mkdtempSync(join(tmpdir(), "dimensionlab-deploy-test-"));
+  const logPath = join(tempDir, "commands.log");
+
+  for (const [name, source] of Object.entries(commands)) {
+    const commandPath = join(tempDir, name);
+    writeFileSync(commandPath, source);
+    chmodSync(commandPath, 0o755);
+  }
+
+  const result = spawnSync("bash", [join(root, "scripts/deploy-dimensionlab-website.sh")], {
+    cwd: root,
+    encoding: "utf8",
+    env: {
+      ...process.env,
+      ...env,
+      DEPLOY_TEST_LOG: logPath,
+      PATH: `${tempDir}:${process.env.PATH ?? ""}`,
+    },
+  });
+
+  return {
+    log: existsSync(logPath) ? readFileSync(logPath, "utf8") : "",
+    status: result.status,
+    stderr: result.stderr,
+    stdout: result.stdout,
+  };
+}
+
+const fakeGitCommand = `#!/usr/bin/env bash
+case "$1" in
+  branch)
+    echo main
+    ;;
+  rev-parse)
+    echo 1234567890abcdef
+    ;;
+  config|submodule)
+    exit 0
+    ;;
+esac
+`;
+
+function fakePodmanCommand(systemdUnit: string): string {
+  return `#!/usr/bin/env bash
+state_file="$DEPLOY_TEST_LOG.state"
+[ -f "$state_file" ] || printf 'initial' > "$state_file"
+printf '%s\\n' "$*" >> "$DEPLOY_TEST_LOG"
+if [ "$1" = "image" ] && [ "$2" = "inspect" ]; then
+  if [ "$4" = "--format" ]; then
+    case "$3" in
+      *:rollback-*)
+        echo sha256:old
+        ;;
+      *)
+        if [ "$(cat "$state_file")" = "rollback" ]; then
+          echo sha256:old
+        else
+          echo sha256:new
+        fi
+        ;;
+    esac
+  fi
+  exit 0
+fi
+if [ "$1" = "tag" ] && [ "$2" != "localhost/dimensionlab-website:latest" ]; then
+  printf 'rollback' > "$state_file"
+fi
+if [ "$1" = "inspect" ]; then
+  case "$*" in
+    *PODMAN_SYSTEMD_UNIT*)
+      echo ${systemdUnit}
+      ;;
+    *State.Running*)
+      echo true
+      ;;
+    *'.Image'*|*'{{.Image}}'*)
+      if [ "$(cat "$state_file")" = "rollback" ]; then
+        echo sha256:old
+      elif [ "\${DEPLOY_TEST_CONTAINER_IMAGE:-new}" = "wrong" ]; then
+        echo sha256:wrong
+      else
+        echo sha256:new
+      fi
+      ;;
+  esac
+fi
+`;
+}
+
+const failingCurlCommand = `#!/usr/bin/env bash
+printf 'curl %s\\n' "$*" >> "$DEPLOY_TEST_LOG"
+exit 22
+`;
+
+const passingCurlCommand = `#!/usr/bin/env bash
+printf 'curl %s\\n' "$*" >> "$DEPLOY_TEST_LOG"
+if [ "$*" = *'/api/dashboard/tiles'* ]; then
+  printf '{"state":"ready","tiles":[]}'
+fi
+`;
+
+function fakeSystemctlCommand(options: { active: boolean; show: boolean }): string {
+  const activeStatus = options.active ? 0 : 3;
+  const showStatus = options.show ? 0 : 1;
+
+  return `#!/usr/bin/env bash
+printf 'systemctl %s\\n' "$*" >> "$DEPLOY_TEST_LOG"
+if [ "$1" = "--user" ] && [ "$2" = "is-active" ]; then
+  exit ${activeStatus}
+fi
+if [ "$1" = "--user" ] && [ "$2" = "show" ]; then
+  exit ${showStatus}
+fi
+if [ "$1" = "--user" ] && [ "$2" = "restart" ]; then
+  exit 0
+fi
+`;
+}

apps/web/src/server/index.test.ts

@@ -50,4 +50,28 @@ describe("server request routing", () => {
     expect(response.status).toBe(405);
     expect(response.headers.get("allow")).toBe("POST");
   });
+
+  test("routes dashboard event stream requests", async () => {
+    const response = await handleRequest(
+      new Request("https://example.test/api/dashboard/events", {
+        method: "GET",
+      }),
+    );
+
+    expect(response.status).toBe(200);
+    expect(response.headers.get("content-type")).toBe(
+      "text/event-stream; charset=utf-8",
+    );
+  });
+
+  test("rejects non-get dashboard event stream requests", async () => {
+    const response = await handleRequest(
+      new Request("https://example.test/api/dashboard/events", {
+        method: "POST",
+      }),
+    );
+
+    expect(response.status).toBe(405);
+    expect(response.headers.get("allow")).toBe("GET");
+  });
 });

apps/web/src/server/index.ts

@@ -1,6 +1,7 @@
 import { extname, normalize } from "node:path";
 import { handleAgentDashboardRoute } from "./routes/agent-dashboard";
 import {
+  handleDashboardEventsRoute,
   handleDashboardRoute,
   handleDashboardTileRoute,
   handleDashboardTilesRoute,
@@ -32,6 +33,11 @@ export async function handleRequest(request: Request): Promise<Response> {
     return handleDashboardTilesRoute(request);
   }
 
+  if (url.pathname === "/api/dashboard/events") {
+    if (request.method !== "GET") return methodNotAllowed(["GET"]);
+    return handleDashboardEventsRoute({ signal: request.signal });
+  }
+
   if (url.pathname.startsWith("/api/dashboard/tile/")) {
     if (request.method !== "GET") return methodNotAllowed(["GET"]);
     return handleDashboardTileRoute(url.pathname);

apps/web/src/server/routes/dashboard.test.ts

@@ -3,6 +3,7 @@ import { dimensionLabDashboardFixture } from "$lib/dashboard-seed/dimensionlab";
 import {
   createDashboardTileCache,
   dashboardTileCacheKey,
+  handleDashboardEventsRoute,
   handleDashboardTilesRoute,
   handleDashboardTileRoute,
   loadDashboardResponse,
@@ -605,6 +606,45 @@ describe("dashboard API route", () => {
     });
   });
 
+  test("streams ready dashboard tile events", async () => {
+    const controller = new AbortController();
+    const response = await handleDashboardEventsRoute({
+      refreshSeedDocument: true,
+      seedIfEmpty: true,
+      signal: controller.signal,
+      tileCache: {
+        async resolve(key) {
+          controller.abort();
+          const tile = JSON.parse(key);
+
+          return {
+            cache: "miss",
+            coalesced: false,
+            response: {
+              state: "ready",
+              tile,
+              item: {
+                id: tile.id,
+                label: "Status",
+                severity: "ok",
+                value: "ok",
+              },
+            },
+          };
+        },
+      },
+    });
+
+    expect(response.headers.get("content-type")).toBe(
+      "text/event-stream; charset=utf-8",
+    );
+    expect(response.headers.get("cache-control")).toBe("no-cache");
+    const body = await response.text();
+    expect(body).toContain("event: dashboard-tile");
+    expect(body).toContain('"state":"ready"');
+    expect(body).toContain('"tile"');
+  });
+
   test("rejects invalid batch tile requests", async () => {
     const response = await handleDashboardTilesRoute(
       new Request("https://example.test/api/dashboard/tiles", {

apps/web/src/server/routes/dashboard.ts

@@ -25,6 +25,10 @@ export interface LoadDashboardResponseOptions
   tileCache?: DashboardTileCache;
 }
 
+export interface DashboardEventsRouteOptions extends LoadDashboardResponseOptions {
+  signal?: AbortSignal;
+}
+
 interface DashboardTileCacheEntry {
   expiresAt: number;
   response: DashboardTileResolution;
@@ -293,6 +297,48 @@ export async function handleDashboardTilesRoute(
   });
 }
 
+export async function handleDashboardEventsRoute(
+  options: DashboardEventsRouteOptions = {},
+): Promise<Response> {
+  const stream = new ReadableStream<Uint8Array>({
+    async start(controller) {
+      const encoder = new TextEncoder();
+      const dashboard = await loadDashboardResponse({
+        ...options,
+        hydrateLiveDatasources: false,
+      });
+
+      if (dashboard.state !== "ready" || options.signal?.aborted) {
+        controller.close();
+        return;
+      }
+
+      try {
+        for (const tile of dashboardEventTiles(dashboard.document)) {
+          if (options.signal?.aborted) break;
+          const resolution = await loadDashboardTileResponse(tile, options);
+          if (resolution.state === "ready") {
+            controller.enqueue(
+              encoder.encode(dashboardTileEventChunk(resolution)),
+            );
+          }
+        }
+      } finally {
+        controller.close();
+      }
+    },
+  });
+
+  return new Response(stream, {
+    headers: {
+      "Cache-Control": "no-cache",
+      "Connection": "keep-alive",
+      "Content-Type": "text/event-stream; charset=utf-8",
+      "X-Accel-Buffering": "no",
+    },
+  });
+}
+
 export function dashboardTileCacheKey(tile: DashboardTileReference): string {
   return JSON.stringify(tile);
 }
@@ -422,6 +468,45 @@ function parseDashboardTileReference(value: unknown): DashboardTileReference | n
   return null;
 }
 
+function dashboardEventTiles(document: DashboardDocument): DashboardTileReference[] {
+  const status = document.statusStrips.flatMap((strip) =>
+    strip.items.map((item): DashboardTileReference => ({
+      kind: "status",
+      stripId: strip.id,
+      id: item.id,
+    })),
+  );
+  const telemetry = document.telemetry
+    .filter((card) => card.datasource?.type === "external")
+    .map((card): DashboardTileReference => ({ kind: "telemetry", id: card.id }));
+  const modules = (document.modules || [])
+    .filter((module) =>
+      module.datasource?.type === "external" ||
+      module.id === "runtime-health-summary"
+    )
+    .map((module): DashboardTileReference => ({ kind: "module", id: module.id }));
+  const services = document.serviceGroups.flatMap((group) =>
+    group.services
+      .filter((service) => service.datasource?.type === "external")
+      .map((service): DashboardTileReference => ({
+        kind: "service",
+        groupId: group.id,
+        id: service.id,
+      })),
+  );
+
+  return [...status, ...telemetry, ...modules, ...services];
+}
+
+function dashboardTileEventChunk(resolution: DashboardTileResolution): string {
+  return [
+    "event: dashboard-tile",
+    `data: ${JSON.stringify(resolution)}`,
+    "",
+    "",
+  ].join("\n");
+}
+
 class DashboardTileCacheResolutionError extends Error {
   readonly cache: "hit" | "miss";
   readonly coalesced: boolean;

apps/web/tests/e2e/dashboard.spec.ts

@@ -16,7 +16,6 @@ const linkedServiceIds = [
   "open-webui",
   "comfyui",
   "models",
-  "prompt-registry",
   "adminer",
   "assistant",
   "suna",
@@ -104,7 +103,7 @@ test.describe("dashboard page QA gate", () => {
     expect(metrics.runtimeBottom).toBeLessThanOrEqual(956);
     expect(metrics.footerBottom).toBeLessThanOrEqual(956);
     expect(metrics.telemetryCardCount).toBe(16);
-    expect(metrics.serviceRowCount).toBe(28);
+    expect(metrics.serviceRowCount).toBe(27);
     expect(metrics.footerCellCount).toBe(5);
     expect(metrics.clippedItems).toEqual([]);
   });

scripts/deploy-dimensionlab-website.sh

@@ -0,0 +1,362 @@
+#!/usr/bin/env bash
+set -Eeuo pipefail
+
+APP_NAME="${APP_NAME:-dimensionlab-website}"
+SERVICE_NAME="${SERVICE_NAME:-dimensionlab-website.service}"
+CONTAINER_NAME="${CONTAINER_NAME:-dimensionlab-website}"
+IMAGE_REPO="${IMAGE_REPO:-localhost/dimensionlab-website}"
+CONTAINERFILE="${CONTAINERFILE:-apps/web/Containerfile}"
+PUBLIC_URL="${PUBLIC_URL:-https://dimensionlab.net/}"
+TILE_BATCH_URL="${TILE_BATCH_URL:-https://dimensionlab.net/api/dashboard/tiles}"
+DEPLOY_RESTART_STRATEGY="${DEPLOY_RESTART_STRATEGY:-auto}"
+DEPLOY_SMOKE_TIMEOUT_SECONDS="${DEPLOY_SMOKE_TIMEOUT_SECONDS:-120}"
+DEPLOY_CONTAINER_START_TIMEOUT_SECONDS="${DEPLOY_CONTAINER_START_TIMEOUT_SECONDS:-90}"
+
+dry_run=false
+rollback_tag=""
+release_tag=""
+latest_tag="${IMAGE_REPO}:latest"
+container_cli=""
+deployment_started=false
+rollback_done=false
+rollback_in_progress=false
+
+usage() {
+  cat <<USAGE
+Usage: $0 [--dry-run]
+
+Build and deploy ${APP_NAME} for the production ${SERVICE_NAME} unit.
+
+Options:
+  --dry-run  Print the deployment actions without changing the host.
+USAGE
+}
+
+log() {
+  printf '[deploy:%s] %s\n' "$APP_NAME" "$*"
+}
+
+fail() {
+  printf '[deploy:%s] ERROR: %s\n' "$APP_NAME" "$*" >&2
+  if [ "${deployment_started:-false}" = "true" ] && [ "${rollback_in_progress:-false}" != "true" ]; then
+    rollback || true
+  fi
+  exit 1
+}
+
+run() {
+  if "$dry_run"; then
+    printf '[deploy:%s] DRY-RUN:' "$APP_NAME"
+    printf ' %q' "$@"
+    printf '\n'
+    return 0
+  fi
+
+  "$@"
+}
+
+for arg in "$@"; do
+  case "$arg" in
+    --dry-run)
+      dry_run=true
+      ;;
+    -h|--help)
+      usage
+      exit 0
+      ;;
+    *)
+      usage >&2
+      fail "unknown argument: $arg"
+      ;;
+  esac
+done
+
+deployment_ref() {
+  printf '%s' "${DEPLOY_REF:-${GITHUB_REF:-${FORGEJO_REF:-}}}"
+}
+
+deployment_event() {
+  printf '%s' "${DEPLOY_EVENT_NAME:-${GITHUB_EVENT_NAME:-${FORGEJO_EVENT_NAME:-}}}"
+}
+
+require_main_push() {
+  local event
+  local ref
+
+  event="$(deployment_event)"
+  ref="$(deployment_ref)"
+
+  if [ -n "$event" ] && [ "$event" != "push" ]; then
+    fail "refusing to deploy for event '$event'; production deploys only run for push"
+  fi
+
+  if [ -n "$ref" ]; then
+    [ "$ref" = "refs/heads/main" ] || fail "refusing to deploy ref '$ref'; expected refs/heads/main"
+    return 0
+  fi
+
+  local branch
+  branch="$(git branch --show-current 2>/dev/null || true)"
+  [ "$branch" = "main" ] || fail "refusing to deploy branch '$branch'; expected main"
+}
+
+select_container_cli() {
+  if [ -n "${DEPLOY_CONTAINER_CLI:-}" ]; then
+    command -v "$DEPLOY_CONTAINER_CLI" >/dev/null 2>&1 || fail "container CLI not found: $DEPLOY_CONTAINER_CLI"
+    container_cli="$DEPLOY_CONTAINER_CLI"
+    return 0
+  fi
+
+  if command -v podman >/dev/null 2>&1; then
+    container_cli="podman"
+    return 0
+  fi
+
+  if command -v docker >/dev/null 2>&1; then
+    container_cli="docker"
+    return 0
+  fi
+
+  fail "podman or docker is required"
+}
+
+current_sha() {
+  if [ -n "${DEPLOY_SHA:-${GITHUB_SHA:-}}" ]; then
+    printf '%s' "${DEPLOY_SHA:-${GITHUB_SHA:-}}"
+    return 0
+  fi
+
+  git rev-parse HEAD
+}
+
+tag_existing_latest_for_rollback() {
+  rollback_tag="${IMAGE_REPO}:rollback-$(date -u +%Y%m%d%H%M%S)"
+  if "$container_cli" image inspect "$latest_tag" >/dev/null 2>&1; then
+    log "tagging current latest image as $rollback_tag"
+    run "$container_cli" tag "$latest_tag" "$rollback_tag"
+  else
+    log "no existing $latest_tag image found; rollback image tag will not be created"
+    rollback_tag=""
+  fi
+}
+
+container_systemd_unit() {
+  "$container_cli" inspect "$CONTAINER_NAME" \
+    --format '{{ index .Config.Labels "PODMAN_SYSTEMD_UNIT" }}' 2>/dev/null || true
+}
+
+require_container_managed_by_service() {
+  local unit
+
+  if "$dry_run"; then
+    log "DRY-RUN: would require $CONTAINER_NAME to be managed by $SERVICE_NAME"
+    return 0
+  fi
+
+  unit="$(container_systemd_unit)"
+  [ "$unit" = "$SERVICE_NAME" ] || fail "refusing to stop $CONTAINER_NAME; expected PODMAN_SYSTEMD_UNIT=$SERVICE_NAME, got '${unit:-unset}'"
+}
+
+validate_restart_strategy() {
+  case "$DEPLOY_RESTART_STRATEGY" in
+    systemctl)
+      if ! "$dry_run" && ! systemctl --user show "$SERVICE_NAME" >/dev/null 2>&1; then
+        fail "systemctl --user cannot access $SERVICE_NAME"
+      fi
+      ;;
+    quadlet-container|kill-container)
+      require_container_managed_by_service
+      ;;
+    auto)
+      if "$dry_run"; then
+        log "DRY-RUN: would validate automatic restart strategy"
+      elif ! command -v systemctl >/dev/null 2>&1 || ! systemctl --user show "$SERVICE_NAME" >/dev/null 2>&1; then
+        require_container_managed_by_service
+      fi
+      ;;
+    *)
+      fail "unknown DEPLOY_RESTART_STRATEGY: $DEPLOY_RESTART_STRATEGY"
+      ;;
+  esac
+}
+
+initialize_submodules() {
+  log "initializing submodules"
+  run git config --global url."https://git.dimensionlab.net/".insteadOf "ssh://git@git.dimensionlab.net/"
+  run git submodule update --init --recursive
+}
+
+build_image() {
+  local sha
+  local short_sha
+
+  sha="$(current_sha)"
+  short_sha="${sha:0:12}"
+  release_tag="${IMAGE_REPO}:${short_sha}"
+
+  [ -f "$CONTAINERFILE" ] || fail "containerfile not found: $CONTAINERFILE"
+
+  log "building $release_tag and $latest_tag from $CONTAINERFILE"
+  run "$container_cli" build -f "$CONTAINERFILE" -t "$release_tag" -t "$latest_tag" .
+}
+
+restart_service() {
+  log "restarting $SERVICE_NAME with strategy $DEPLOY_RESTART_STRATEGY"
+
+  case "$DEPLOY_RESTART_STRATEGY" in
+    systemctl)
+      run systemctl --user restart "$SERVICE_NAME"
+      ;;
+    quadlet-container|kill-container)
+      require_container_managed_by_service
+      run "$container_cli" stop "$CONTAINER_NAME"
+      ;;
+    auto)
+      if command -v systemctl >/dev/null 2>&1 && systemctl --user is-active "$SERVICE_NAME" >/dev/null 2>&1; then
+        run systemctl --user restart "$SERVICE_NAME"
+      else
+        require_container_managed_by_service
+        run "$container_cli" stop "$CONTAINER_NAME"
+      fi
+      ;;
+    *)
+      fail "unknown DEPLOY_RESTART_STRATEGY: $DEPLOY_RESTART_STRATEGY"
+      ;;
+  esac
+}
+
+latest_image_id() {
+  "$container_cli" image inspect "$latest_tag" --format '{{.Id}}' 2>/dev/null || true
+}
+
+container_image_id() {
+  "$container_cli" inspect "$CONTAINER_NAME" --format '{{.Image}}' 2>/dev/null || true
+}
+
+container_running() {
+  local running
+  running="$("$container_cli" inspect "$CONTAINER_NAME" --format '{{.State.Running}}' 2>/dev/null || true)"
+  [ "$running" = "true" ]
+}
+
+wait_for_container_restart() {
+  local expected_image
+
+  if "$dry_run"; then
+    log "DRY-RUN: would wait for $CONTAINER_NAME to run $latest_tag"
+    return 0
+  fi
+
+  expected_image="$(latest_image_id)"
+  [ -n "$expected_image" ] || fail "could not resolve image id for $latest_tag"
+
+  wait_for_container_image "$expected_image" "new image" || fail "$CONTAINER_NAME did not restart on $latest_tag within ${DEPLOY_CONTAINER_START_TIMEOUT_SECONDS}s"
+}
+
+wait_for_container_image() {
+  local expected_image="$1"
+  local label="$2"
+  local deadline
+
+  deadline=$((SECONDS + DEPLOY_CONTAINER_START_TIMEOUT_SECONDS))
+
+  while [ "$SECONDS" -lt "$deadline" ]; do
+    if container_running && [ "$(container_image_id)" = "$expected_image" ]; then
+      log "$CONTAINER_NAME is running the $label"
+      return 0
+    fi
+
+    sleep 2
+  done
+
+  return 1
+}
+
+smoke_get() {
+  local url="$1"
+  curl -fsS --max-time 10 -o /dev/null "$url"
+}
+
+smoke_tiles() {
+  local response
+
+  response="$(
+    curl -fsS --max-time 20 \
+      -H "content-type: application/json" \
+      --data '{"tiles":[{"kind":"status","stripId":"footer-status","id":"system-status"}]}' \
+      "$TILE_BATCH_URL"
+  )"
+
+  [[ "$response" == *'"state":"ready"'* ]]
+}
+
+wait_for_smoke() {
+  local deadline
+
+  if "$dry_run"; then
+    log "DRY-RUN: would smoke check $PUBLIC_URL and $TILE_BATCH_URL"
+    return 0
+  fi
+
+  deadline=$((SECONDS + DEPLOY_SMOKE_TIMEOUT_SECONDS))
+  until smoke_get "$PUBLIC_URL" && smoke_tiles; do
+    if [ "$SECONDS" -ge "$deadline" ]; then
+      fail "smoke checks failed for $PUBLIC_URL and $TILE_BATCH_URL"
+    fi
+
+    sleep 3
+  done
+
+  log "smoke checks passed"
+}
+
+rollback() {
+  local rollback_image
+
+  if [ "$deployment_started" != "true" ] || [ -z "$rollback_tag" ] || [ "$rollback_done" = "true" ]; then
+    return 0
+  fi
+
+  rollback_done=true
+  rollback_in_progress=true
+  printf '[deploy:%s] rolling back to %s\n' "$APP_NAME" "$rollback_tag" >&2
+  rollback_image="$("$container_cli" image inspect "$rollback_tag" --format '{{.Id}}' 2>/dev/null || true)"
+  "$container_cli" tag "$rollback_tag" "$latest_tag" || true
+  if container_running; then
+    restart_service || true
+  else
+    printf '[deploy:%s] waiting for %s to recover with rollback image\n' "$APP_NAME" "$SERVICE_NAME" >&2
+  fi
+  if [ -n "$rollback_image" ] && wait_for_container_image "$rollback_image" "rollback image"; then
+    printf '[deploy:%s] rollback image is running\n' "$APP_NAME" >&2
+  else
+    printf '[deploy:%s] ERROR: rollback image did not become healthy\n' "$APP_NAME" >&2
+  fi
+  rollback_in_progress=false
+}
+
+on_error() {
+  local status=$?
+  rollback
+  exit "$status"
+}
+
+trap on_error ERR
+
+require_main_push
+select_container_cli
+validate_restart_strategy
+
+log "using container CLI: $container_cli"
+log "target image: $latest_tag"
+log "target service: $SERVICE_NAME"
+
+initialize_submodules
+tag_existing_latest_for_rollback
+build_image
+deployment_started=true
+restart_service
+wait_for_container_restart
+wait_for_smoke
+
+log "deployment finished"