improvement(ci): share production builds across HTTP and Electron suites

Author: waleedlatif1

.github/scripts/http-e2e.sh

@@ -1,9 +1,9 @@
 #!/usr/bin/env bash
 # Runs one end-to-end suite group over real HTTP, each against its own app: `next dev`, except
-# mobile and Project Files in CI, which reuse the production build from `e2e-build`.
+# CI groups, which reuse the production build from `e2e-build`.
 # The job extracts it into apps/sim/.next before running these suites.
 #
-# Usage: http-e2e.sh <scim|cli|stop-after|desktop-inbox|project-files|mobile>   (run from apps/sim)
+# Usage: http-e2e.sh <scim|cli|stop-after|desktop-inbox|project-files|mobile|seo>   (run from apps/sim)
 #
 # The job provides DATABASE_URL, BETTER_AUTH_SECRET and ENCRYPTION_KEY; each group sets the rest of
 # its app's environment here. Reports and server logs land in $RUNNER_TEMP/e2e.
@@ -20,7 +20,7 @@
 # telemetry flush runs detached and still writes .next/dev).
 set -euo pipefail
 
-group=${1:?usage: http-e2e.sh <scim|cli|stop-after|desktop-inbox|project-files|mobile>}
+group=${1:?usage: http-e2e.sh <scim|cli|stop-after|desktop-inbox|project-files|mobile|seo>}
 report_dir="$RUNNER_TEMP/e2e"
 ready_timeout_seconds=600
 mkdir -p "$report_dir"
@@ -34,7 +34,7 @@ realtime_tag=''
 realtime_log=''
 browsers_pid=''
 # `dev` or `start`; a group sets it before start_app.
-app_server=dev
+app_server=${E2E_APP_SERVER:-dev}
 
 finish() {
   local status=$?
@@ -174,7 +174,7 @@ case "$group" in
       bun --no-env-file scripts/test-mcp-host-e2e.ts
     CLI_LATENCY_E2E_BASE_URL="$NEXT_PUBLIC_APP_URL" \
     CLI_LATENCY_E2E_DATABASE_URL="$DATABASE_URL" \
-    CLI_LATENCY_E2E_RUNS=3 \
+    CLI_LATENCY_E2E_RUNS=1 \
     CLI_LATENCY_E2E_WARMUP=1 \
     CLI_LATENCY_E2E_REPORT_PATH="$report_dir/cli-run-latency-report.json" \
       bun run test:cli-run-latency:e2e
@@ -246,7 +246,13 @@ case "$group" in
     MOBILE_E2E_AUTH_SECRET="$BETTER_AUTH_SECRET" \
     MOBILE_E2E_REPORT_PATH="$report_dir/mobile-e2e-report.json" \
       bun run test:mobile:e2e
-    # The same production build answers as a crawler sees it: statuses, robots, sitemap, JSON-LD.
+    ;;
+
+  seo)
+    app_server=start
+    export NEXT_PUBLIC_FORCE_HOSTED=false
+    start_app seo 3024 'landing SEO' record-http-status
+    # Plain HTTP assertions have no browser-specific behavior.
     LANDING_SEO_E2E_BASE_URL="$NEXT_PUBLIC_APP_URL" \
     LANDING_SEO_E2E_REPORT_PATH="$report_dir/landing-seo-e2e-report.json" \
       bun run test:landing-seo:e2e

.github/workflows/checks.yml

@@ -167,11 +167,10 @@ jobs:
   # The suites exercise HTTP behavior rather than a provisioning path, so they run once, against
   # the production (migrate) path. Each group's app environment lives in http-e2e.sh.
   #
-  # SCIM's hosted suites use 8 vCPU; the other self-hosted suites fit on 4. The mobile browser
-  # suite runs in e2e-production below.
+  # All groups reuse the production build; compiling routes is not part of HTTP acceptance.
   e2e:
     name: e2e (${{ matrix.group }})
-    needs: changes
+    needs: [changes, e2e-build]
     if: needs.changes.outputs.full == 'true'
     runs-on: ${{ (vars.CI_PROVIDER == '' || vars.CI_PROVIDER == 'blacksmith') && matrix.runner || 'ubuntu-latest' }}
     timeout-minutes: 20
@@ -180,13 +179,15 @@ jobs:
       matrix:
         include:
           - group: scim
-            runner: blacksmith-8vcpu-ubuntu-2404
+            runner: blacksmith-4vcpu-ubuntu-2404
           - group: cli
             runner: blacksmith-4vcpu-ubuntu-2404
           - group: stop-after
             runner: blacksmith-4vcpu-ubuntu-2404
           - group: desktop-inbox
             runner: blacksmith-4vcpu-ubuntu-2404
+          - group: seo
+            runner: blacksmith-4vcpu-ubuntu-2404
     services:
       postgres:
         image: pgvector/pgvector:pg17
@@ -214,6 +215,7 @@ jobs:
     env:
       DATABASE_URL: postgresql://postgres:postgres@127.0.0.1:5432/sim_test
       BETTER_AUTH_SECRET: http-e2e-ci-secret-at-least-32-characters
+      E2E_APP_SERVER: start
       ENCRYPTION_KEY: '0000000000000000000000000000000000000000000000000000000000000000'
 
     steps:
@@ -241,12 +243,39 @@ jobs:
         working-directory: packages/db
         run: bun run db:migrate
 
+      - &download-e2e-build
+        name: Download the production build
+        uses: actions/download-artifact@9000827ccba6bdab643e8b6fd33ac0654aef8333 # v8.0.2
+        with:
+          name: next-build
+          path: ${{ runner.temp }}
+
+      # A hosted build (NEXT_PUBLIC_APP_URL on sim.ai) prerenders every post; the build production
+      # images use prerenders none. The URL itself is no marker: next.config's redirects name
+      # www.sim.ai in every build.
+      - &extract-e2e-build
+        name: Extract the production build
+        run: |
+          tar -I zstd -xf "$RUNNER_TEMP/next-build.tar.zst" -C apps/sim
+          jq -e '[.routes[].srcRoute] | any(. == "/blog/[slug]" or . == "/library/[slug]") | not' \
+            apps/sim/.next/prerender-manifest.json > /dev/null ||
+            { echo "::error::The build prerendered posts (a hosted build) or has no prerender manifest."; exit 1; }
+
       # No step timeout: the job's bound covers a hang without cutting a slow but healthy suite
       # short of writing its report.
       - name: Run end-to-end suites
         working-directory: apps/sim
         run: bash ../../.github/scripts/http-e2e.sh "${{ matrix.group }}"
 
+      - name: Upload acceptance timing and coverage reports
+        if: always()
+        uses: actions/upload-artifact@cf430e030ddbb5b0abf93d22962f4752f3646cd9 # v7.0.2
+        with:
+          name: e2e-reports-${{ matrix.group }}
+          path: ${{ runner.temp }}/e2e/**/*.json
+          if-no-files-found: ignore
+          retention-days: 7
+
       - name: Upload end-to-end reports and server logs
         if: failure()
         uses: actions/upload-artifact@cf430e030ddbb5b0abf93d22962f4752f3646cd9 # v7.0.2
@@ -340,21 +369,8 @@ jobs:
         working-directory: packages/db
         run: bun run db:migrate
 
-      - name: Download the production build
-        uses: actions/download-artifact@9000827ccba6bdab643e8b6fd33ac0654aef8333 # v8.0.2
-        with:
-          name: next-build
-          path: ${{ runner.temp }}
-
-      # A hosted build (NEXT_PUBLIC_APP_URL on sim.ai) prerenders every post; the build production
-      # images use prerenders none. The URL itself is no marker: next.config's redirects name
-      # www.sim.ai in every build.
-      - name: Extract the production build
-        run: |
-          tar -I zstd -xf "$RUNNER_TEMP/next-build.tar.zst" -C apps/sim
-          jq -e '[.routes[].srcRoute] | any(. == "/blog/[slug]" or . == "/library/[slug]") | not' \
-            apps/sim/.next/prerender-manifest.json > /dev/null ||
-            { echo "::error::The build prerendered posts (a hosted build) or has no prerender manifest."; exit 1; }
+      - *download-e2e-build
+      - *extract-e2e-build
 
       # JOBS=4 caps node-gyp parallelism as docker/app.Dockerfile does (laverdet/isolated-vm#428).
       - name: Build native isolate addon for Project file rendering
@@ -393,15 +409,14 @@ jobs:
           if-no-files-found: ignore
           retention-days: 7
 
-  # Desktop tools in the real Electron app against a local app, on its own runner: the
-  # Electron app, the dev app and its realtime server together outgrow the e2e runner.
+  # Desktop tools exercise the real Electron app against the shared production build.
   # Two runners: the lease-lapse test sleeps through a long import and then a lease (~3 min), as
   # long as the rest of the spec together, so on one runner it set the time for every run.
   desktop-live:
     name: desktop-live (${{ matrix.tests }})
-    needs: changes
+    needs: [changes, e2e-build]
     if: needs.changes.outputs.desktop_live == 'true'
-    runs-on: ${{ (vars.CI_PROVIDER == '' || vars.CI_PROVIDER == 'blacksmith') && 'blacksmith-16vcpu-ubuntu-2404' || 'ubuntu-latest' }}
+    runs-on: ${{ (vars.CI_PROVIDER == '' || vars.CI_PROVIDER == 'blacksmith') && 'blacksmith-8vcpu-ubuntu-2404' || 'ubuntu-latest' }}
     timeout-minutes: 30
     strategy:
       fail-fast: false
@@ -452,22 +467,8 @@ jobs:
         working-directory: packages/db
         run: bun run db:migrate
 
-      # Turbopack's dev cache turns the spec's route warm-up from a cold compile (~4 min) into a
-      # restore. It is content-addressed, so a pull request's changed modules still recompile; the
-      # key carries the installed Next version so an upgrade starts from an empty cache, and the
-      # event and fork segments keep untrusted runs off the cache trusted runs read.
-      - name: Resolve Turbopack dev cache key
-        id: next-cache
-        run: echo "key=${GITHUB_REPOSITORY}-next-dev-desktop-live-${GITHUB_EVENT_NAME}${FORK_SUFFIX}-$(jq -r .version node_modules/next/package.json)" >> "$GITHUB_OUTPUT"
-        env:
-          FORK_SUFFIX: ${{ github.event.pull_request.head.repo.fork && '-fork' || '' }}
-
-      - name: Mount Turbopack dev cache
-        uses: ./.github/actions/cache
-        with:
-          provider: ${{ vars.CI_PROVIDER }}
-          key: ${{ steps.next-cache.outputs.key }}
-          path: ./apps/sim/.next/dev
+      - *download-e2e-build
+      - *extract-e2e-build
 
       # Chat switches, Stop, sign-out, approval, the dormant-executor round trip, and background
       # runs across a chat switch and a network cut. The spec runs the recording proxy (the app's
@@ -496,37 +497,12 @@ jobs:
           # every process it started has exited.
           realtime_tag="desktop-realtime-$GITHUB_RUN_ID-$GITHUB_RUN_ATTEMPT-$$"
           server_tag="desktop-tools-$GITHUB_RUN_ID-$GITHUB_RUN_ATTEMPT-$$"
-          # Keeps the restored cache under the cap the dev scripts apply locally.
-          (cd apps/sim && bun run dev:cache:cap)
           start_sim() {
-            (cd apps/sim && E2E_APP="$server_tag" exec setsid node ../../node_modules/next/dist/bin/next dev --hostname 127.0.0.1 \
+            (cd apps/sim && E2E_APP="$server_tag" LOG_LEVEL=INFO exec setsid node ../../node_modules/next/dist/bin/next start --hostname 127.0.0.1 \
               --port 3021 >> "$next_log" 2>&1) &
             server_pid=$!
           }
-          # A corrupted cache aborts Turbopack instead of falling back. Only then is the shared cache
-          # dropped: a failing test must not cost every later run its warm cache.
-          cache_broken() {
-            grep -qiE 'cache corruption|turbopack.*panic|panicked' "$next_log" 2>/dev/null
-          }
-          # The cache directory is a mount point: empty it rather than remove it. Absolute, because
-          # the EXIT trap runs after the step has moved into apps/desktop.
-          clear_cache() {
-            find "$GITHUB_WORKSPACE/apps/sim/.next/dev" -mindepth 1 -maxdepth 1 -exec rm -rf {} +
-          }
-          # A sticky disk keeps nothing from a failed job, so a clear here sticks only when the job
-          # passes. The marker names the cache for desktop-live-cache-reset, which empties it after
-          # a failed job; written on each panic, it survives the restart truncating the log.
-          mark_cache_broken() {
-            jq -r .version "$GITHUB_WORKSPACE/node_modules/next/package.json" > "$report_dir/turbopack-cache-broken"
-          }
-          # SIGINT first: `next dev` SIGKILLs its server 100ms after SIGTERM, which discards a cache
-          # write in flight. stop-session.sh then removes anything still running.
-          # SIGINT is best-effort; the cleanup always runs, since workers and the detached telemetry
-          # flush can outlive a server that has already exited.
           stop_sim() {
-            if kill -INT "$server_pid" 2>/dev/null; then
-              for _ in $(seq 1 30); do kill -0 "$server_pid" 2>/dev/null || break; sleep 1; done
-            fi
             bash "$GITHUB_WORKSPACE/.github/scripts/stop-session.sh" "$server_pid" "$server_tag"
           }
           (cd apps/realtime && PORT=3023 SIM_DB_ROLE=realtime ALLOWED_ORIGINS="$NEXT_PUBLIC_APP_URL" \
@@ -538,11 +514,6 @@ jobs:
             stop_sim || status=1
             bash "$GITHUB_WORKSPACE/.github/scripts/stop-session.sh" "$realtime_pid" "$realtime_tag" || status=1
             wait "$server_pid" "$realtime_pid" 2>/dev/null || true
-            if cache_broken; then
-              echo "::warning::Turbopack reported a broken dev cache; clearing it for the next run."
-              mark_cache_broken
-              clear_cache
-            fi
             exit "$status"
           }
           trap finish EXIT
@@ -554,21 +525,10 @@ jobs:
           # which a Linux run does not use.
           (cd apps/desktop && bun run scripts/build.ts)
           started=$SECONDS
-          retried=0
           until curl --fail --silent --max-time 10 http://127.0.0.1:3021/api/health > /dev/null &&
             curl --fail --silent --max-time 10 http://127.0.0.1:3023/health > /dev/null; do
             if ! kill -0 "$server_pid" 2>/dev/null; then
               tail -n 200 "$next_log"
-              if [ "$retried" = 0 ] && cache_broken; then
-                echo "::warning::Turbopack rejected the restored dev cache; restarting from an empty cache."
-                bash "$GITHUB_WORKSPACE/.github/scripts/stop-session.sh" "$server_pid" "$server_tag"
-                mark_cache_broken
-                clear_cache
-                : > "$next_log"
-                retried=1
-                start_sim
-                continue
-              fi
               exit 1
             fi
             kill -0 "$realtime_pid" 2>/dev/null || { tail -n 200 "$report_dir/desktop-tools-realtime.log"; exit 1; }
@@ -595,63 +555,6 @@ jobs:
           if-no-files-found: ignore
           retention-days: 7
 
-  # Empties a dev cache Turbopack panicked on in a failed desktop-live run. The run's own clear is
-  # discarded with the rest of its failed job, so every later run would restore the same broken
-  # cache and fail; emptied in this passing job, the empty cache is what the next run mounts. A
-  # failure the cache did not cause leaves no marker, and the warm cache stays.
-  desktop-live-cache-reset:
-    name: desktop-live cache reset
-    needs: desktop-live
-    if: ${{ always() && needs.desktop-live.result == 'failure' }}
-    runs-on: *runner-2vcpu
-    timeout-minutes: 5
-    steps:
-      - name: Download desktop-live results
-        uses: actions/download-artifact@9000827ccba6bdab643e8b6fd33ac0654aef8333 # v8.0.2
-        with:
-          pattern: desktop-live-e2e-results-*
-          path: ${{ runner.temp }}/desktop-live
-          merge-multiple: true
-
-      # The marker carries only the Next version; the rest of the key comes from this run's own
-      # context, as in desktop-live, so a fork's marker can name no cache but its own.
-      - name: Resolve the broken cache's key
-        id: broken
-        env:
-          MARKER: ${{ runner.temp }}/desktop-live/turbopack-cache-broken
-          FORK_SUFFIX: ${{ github.event.pull_request.head.repo.fork && '-fork' || '' }}
-        run: |
-          if [ ! -s "$MARKER" ]; then
-            echo "No broken dev cache reported; keeping the warm cache."
-            exit 0
-          fi
-          next_version="$(head -n 1 "$MARKER")"
-          if ! [[ "$next_version" =~ ^[0-9]+\.[0-9]+\.[0-9]+[0-9A-Za-z.-]*$ ]]; then
-            echo "::error::The broken cache marker holds no Next version: '$next_version'"
-            exit 1
-          fi
-          echo "key=${GITHUB_REPOSITORY}-next-dev-desktop-live-${GITHUB_EVENT_NAME}${FORK_SUFFIX}-${next_version}" >> "$GITHUB_OUTPUT"
-
-      - *checkout-mirror
-      - *checkout-plain
-
-      # A desktop-live leg that passed may have saved this run's actions/cache entry already; the
-      # suffix gives the emptied cache its own, newer entry for the next run's prefix restore.
-      - name: Mount the broken dev cache
-        if: steps.broken.outputs.key != ''
-        uses: ./.github/actions/cache
-        with:
-          provider: ${{ vars.CI_PROVIDER }}
-          key: ${{ steps.broken.outputs.key }}
-          path: ./apps/sim/.next/dev
-          save-suffix: -reset
-
-      - name: Empty the dev cache
-        if: steps.broken.outputs.key != ''
-        run: |
-          mkdir -p apps/sim/.next/dev
-          find apps/sim/.next/dev -mindepth 1 -maxdepth 1 -exec rm -rf {} +
-
   # Lint, audits, type-check and the schema sync check: about two minutes of mostly cached work,
   # kept off the test shards so neither waits on the other.
   lint:
@@ -985,7 +888,7 @@ jobs:
   e2e-build:
     name: build (e2e)
     needs: changes
-    if: needs.changes.outputs.full == 'true'
+    if: needs.changes.outputs.full == 'true' || needs.changes.outputs.desktop_live == 'true'
     runs-on: ${{ (vars.CI_PROVIDER == '' || vars.CI_PROVIDER == 'blacksmith') && 'blacksmith-16vcpu-ubuntu-2404' || 'linux-x64-8-core' }}
     timeout-minutes: 25
 

apps/sim/lib/mothership/tools/client/computer-tool-unload.test.ts

@@ -8,6 +8,7 @@ import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest'
 import { copilotConfirmBodySchema } from '@/lib/api/contracts/copilot'
 import { executeComputerToolOnClient } from '@/lib/mothership/tools/client/computer-tool-execution'
 import { clearUserData } from '@/stores'
+import '@/stores/reset-all-stores'
 
 const native = vi.hoisted(() => ({ execute: vi.fn(), cancel: vi.fn() }))
 vi.mock('@/lib/computer-use/transport', () => ({