Test

fix(marketing): improve hero layout and restore projector demo #14280

Workflow file for this run

	name: Test

	on:
	push:
	branches:
	- main
	paths:
	- "web/**"
	- "electron/**"
	- "packages/**"
	- "reliability/**"
	- "mobile/**"
	- "examples/**"
	- "scripts/**"
	- "Dockerfile"
	- ".dockerignore"
	- "package.json"
	- "package-lock.json"
	- "tsconfig.json"
	- "tsconfig.*.json"
	- "turbo.json"
	- ".nvmrc"
	- ".github/workflows/test.yml"
	- ".github/workflows/quality-checks.yml"
	- ".github/actions/setup-build/**"
	pull_request:
	branches:
	- main
	# No `paths` filter here: this workflow's "Quality Gate" job is a REQUIRED
	# status check. A path-filtered required check is left Pending (never runs)
	# when a PR push/merge touches no matching path — e.g. a clean "merge main"
	# commit — which blocks the PR and auto-merge indefinitely. Running the gate
	# on every PR keeps the required check from getting stuck. (The `push` to
	# main above keeps its path filter — there is no required-check to block.)

	# Cancel superseded runs for the same branch/PR; the latest push is what matters.
	concurrency:
	group: ${{ github.workflow }}-${{ github.ref }}
	cancel-in-progress: true

	jobs:
	# Does this push touch anything but prose? A documentation-only change has
	# nothing for typecheck, lint, the test suites, the image build or the
	# browser E2E runs to say, and every leg it starts holds a runner other PRs
	# queue behind (the account cap is ~20 concurrent jobs). The workflow still
	# triggers on every PR — see the `paths` note above — so the required check
	# is always reported; the legs inside it skip instead, and a skipped job
	# counts as a pass for a required check.
	#
	# The prose set is deliberately narrow: `docs/**` (the Jekyll site), the
	# Markdown at the repo root, the agent instruction files anywhere, and the
	# Markdown under `.github/`. Markdown that code reads at run time — a
	# sandbox pack's `SKILL.md`, a package `README.md` — is NOT in it and keeps
	# running the full gate.
	#
	# Fails open: the step is `continue-on-error` and the output falls back to
	# "true", so a filter that cannot resolve a base runs everything rather than
	# waving a change through.
	#
	# It also decides whether the diff can reach the root workspace at all.
	# `marketing/` is a separate npm project — its own lockfile, its own install,
	# built and deployed by marketing-ci.yml — and nothing in this workflow
	# touches it. A marketing-only PR used to build ~55 backend packages,
	# typecheck three apps, run every suite, build the container image and drive
	# two browser E2E jobs; with an account-wide cap of ~20 concurrent jobs, that
	# is time the rest of the queue spends waiting. `shared` is false for such a
	# diff and every leg skips, exactly as it does for prose.
	#
	# `mobile/` is NOT carved out, though it is also a separate npm project. Its
	# typecheck resolves `@nodetool-ai/websocket`'s router types (not a mobile
	# dependency at all) and `@nodetool-ai/protocol` through the root workspace,
	# so it is not separable from this gate without changing how mobile resolves
	# those packages. See the workflows README.
	changes:
	name: changes
	runs-on: ubuntu-latest
	timeout-minutes: 5
	outputs:
	code: ${{ steps.filter.outputs.code \|\| 'true' }}
	shared: ${{ steps.filter.outputs.shared \|\| 'true' }}
	steps:
	- name: Checkout code
	uses: actions/checkout@v7
	with:
	fetch-depth: 1

	- name: Detect non-documentation changes
	id: filter
	continue-on-error: true
	uses: dorny/paths-filter@ceb8a2b8f2d89434be7ff52d3de7ec3738c5cc9d # v4.0.3
	with:
	filters: \|
	code:
	- '!{docs/*,.md,/AGENTS.md,/CLAUDE.md,.github/*/.md}'
	shared:
	- '!{docs/*,.md,/AGENTS.md,/CLAUDE.md,.github/*/.md,marketing/**,.github/workflows/marketing-ci.yml}'

	quality:
	name: Quality Gate (npm run check)
	needs: changes
	# `always()` plus the explicit `== 'false'` test below so a `changes` job
	# that never reported (infrastructure failure, not a filter result) runs the
	# full gate. Only an explicit "this diff is prose" skips work — a required
	# check must never go green because the job that decides was broken.
	if: always()
	uses: ./.github/workflows/quality-checks.yml
	# A called workflow's permissions are capped by the caller's, so the gate's
	# `docker` leg only gets its read-only GHCR buildcache access if it is
	# granted here too.
	permissions:
	contents: read
	packages: read
	with:
	# false → every check leg (deps/typecheck/lint/tests) runs to completion
	# so each reports its own red/green instead of the first failure masking
	# the rest. The required check is the reusable workflow's `quality` gate
	# job, surfaced unchanged as "Quality Gate (npm run check) / quality".
	fail-fast: false
	skip-tests: false
	# A documentation-only push skips every leg inside the gate. The gate job
	# itself still runs and reports, so the required check lands green.
	docs-only: ${{ needs.changes.outputs.code == 'false' }}
	# Same fail-safe reading as `docs-only`: the output defaults to 'true',
	# so a filter that could not resolve a base runs the whole gate. False
	# only for a diff confined to marketing/, whose gate is marketing-ci.yml.
	shared: ${{ needs.changes.outputs.shared != 'false' }}

	# Heavy jobs run concurrently with the quality gate (both wait only on the
	# seconds-long `changes` job) rather than after it, so
	# the critical path is max(quality, integration, e2e) instead of
	# quality + max(integration, e2e). Trade-off: a PR that fails a static check
	# still spends these runner minutes. They build packages themselves (they are
	# in this workflow, not the reusable quality-checks one, so they can't consume
	# its build artifact without depending on the whole gate — which would undo
	# the decoupling).
	integration:
	name: Workflow Integration Tests
	needs: changes
	# Same fail-safe reading as the gate above: skip only on an explicit
	# "prose", or an explicit "this diff never leaves marketing/" — it runs
	# shipped-workflow execution out of packages/base-nodes, which marketing
	# cannot reach.
	if: ${{ always() && needs.changes.outputs.code != 'false' && needs.changes.outputs.shared != 'false' }}
	runs-on: ubuntu-latest
	timeout-minutes: 45

	steps:
	- name: Checkout code
	uses: actions/checkout@v7

	- name: Set up Node, install deps, build packages
	uses: ./.github/actions/setup-build

	- name: Run shipped-workflow execution tests
	run: npm run test:integration

	workflow-runner-e2e:
	name: Workflow Runner Browser E2E
	needs: changes
	if: ${{ always() && needs.changes.outputs.code != 'false' && needs.changes.outputs.shared != 'false' }}
	runs-on: ubuntu-latest
	timeout-minutes: 45

	steps:
	- name: Checkout code
	uses: actions/checkout@v7

	- name: Set up Node, install deps, build packages
	# mesa-vulkan-drivers ships lavapipe (CPU Vulkan ICD) so headless
	# Chromium can acquire a WebGPU adapter on GPU-less CI runners. The
	# workflow-runner browser harness uses it to drive shader-pool tests
	# (color.brightnessContrast, etc.) against the same module catalog
	# the server uses on Node/Dawn.
	uses: ./.github/actions/setup-build
	with:
	extra-apt: mesa-vulkan-drivers

	- name: Cache Playwright browsers
	id: playwright-cache
	uses: actions/cache@v6
	with:
	path: ~/.cache/ms-playwright
	key: ${{ runner.os }}-playwright-${{ hashFiles('**/package-lock.json') }}
	restore-keys: \|
	${{ runner.os }}-playwright-

	- name: Install Playwright Chromium browser
	run: npx playwright install --with-deps chromium
	working-directory: packages/workflow-runner

	- name: Run workflow-runner browser E2E tests
	run: npm run test:e2e --workspace=@nodetool-ai/workflow-runner

	- name: Upload Playwright report on failure
	if: failure()
	uses: actions/upload-artifact@v7
	with:
	name: workflow-runner-playwright-report
	path: packages/workflow-runner/playwright-report
	retention-days: 7

Navigation Menu

Uh oh!

fix(marketing): improve hero layout and restore projector demo #14280

Workflow file

fix(marketing): improve hero layout and restore projector demo #14280

Uh oh!

Workflow file for this run