Repository navigation

chore(release): version packages #2163

Workflow file for this run

	name: CI

	# KEEP IN SYNC: the `paths-ignore` lists below must exactly mirror the
	# `paths:` list in ci-docs-noop.yml — the no-op twin that reports the required
	# lint/typecheck/test/editor-ux contexts on docs-only PRs, which would
	# otherwise stay BLOCKED because this workflow never runs for them.

	on:
	push:
	branches: [main]
	paths-ignore:
	- '*.md'
	- '*/.md'
	- 'docs/**'
	- '.github/*.md'
	pull_request:
	branches: [main]
	paths-ignore:
	- '*.md'
	- '*/.md'
	- 'docs/**'
	- '.github/*.md'
	# Manual trigger, and the probe for the fail-closed event dispatch below
	# (0372): `workflow_dispatch` is deliberately an event none of the jobs name,
	# so running it by hand exercises the unknown-event path end to end. Every
	# gate in this file must do real work under it — a green dispatch run that
	# executed nothing is the bug, not the pass.
	workflow_dispatch:

	# Cancel superseded PR runs (group keyed on the PR number). Pushes to main key
	# on the commit SHA so each merge gets a unique group — never cancelled, never
	# queued behind the previous merge's run. (Exploration 0193.)
	concurrency:
	group: ci-${{ github.event.pull_request.number \|\| github.sha }}
	cancel-in-progress: true

	jobs:
	# Fast checks that need NO build: prettier, eslint (not type-aware), and the
	# cloud license-boundary grep. Kept under the name `lint` so the existing
	# required status check still resolves. These used to sit serially behind the
	# ~250s build+typecheck step; now a formatting/lint failure surfaces in under
	# a minute instead of after the build (exploration 0193).
	lint:
	runs-on: ubuntu-latest
	steps:
	# fetch-depth: 0 — check:exploration-fallow dates each exploration from
	# the commit that introduced it. A shallow clone has no such commit, so
	# every document would look new and the ratchet would pass vacuously; it
	# fails loudly below a coverage floor rather than reporting a false green.
	- uses: actions/checkout@v4
	with:
	fetch-depth: 0
	- uses: ./.github/actions/setup

	- name: Format check
	run: pnpm prettier --check "packages/*/.{ts,tsx}" "apps/*/.{ts,tsx}"

	- name: Lint
	run: pnpm lint

	- name: Cloud license boundary
	run: pnpm check:cloud-boundary

	- name: Plugin license policy (0196)
	run: pnpm check:plugin-licenses

	# The negative control for the values-as-code gates (0430). Their green
	# state is otherwise unfalsifiable: a regex that stopped matching after a
	# rename looks exactly like a clean tree, and a footprint metric that
	# stopped being measurable looks exactly like a lean app. This plants
	# violations they MUST flag, so green means something.
	- name: Gate negative controls (0430)
	run: pnpm check:gate-controls

	- name: Motion vocabulary (0199)
	run: pnpm check:motion-vocab

	- name: Surface tokens (0299)
	run: pnpm check:surface-tokens

	- name: Humane patterns (0234)
	run: pnpm check:humane-patterns

	# POSSE's machine-checkable half (0432): a syndicated post is a mirror,
	# so every announced post must carry a link home to xnet.fyi, fit
	# Bluesky's 300-grapheme budget, and have link facets whose byte ranges
	# actually decode to their URL.
	- name: Syndication invariants (0432)
	run: pnpm check:syndication

	# Charter §Agency's second half: a capability you cannot see is not a
	# degree of freedom you have. Every other check here asks whether the
	# code is correct — none asks whether a shipped capability is findable,
	# so a feature can land reachable-but-invisible and nothing notices.
	# Advisory for now; goes --strict once the flag population has held
	# green for a release. (0428.)
	- name: Capability surfaces (0428)
	run: pnpm check:capability-surface

	# Every agent lane must build its retrieval. The omission is invisible in
	# review — the call reads fine, it is what it doesn't pass that drops the
	# lane to a keyword scan with no graph stage. (0415.)
	- name: AI surfaces wire retrieval (0415)
	run: pnpm check:ai-retrieval

	# The dev-only point-and-change source stamp must never reach a
	# production bundle — it would publish the repo's directory layout in the
	# DOM of every page. A diff cannot show that; this can. (0399.)
	- name: Source stamp is dev-only (0399)
	run: pnpm guard:no-source-stamp

	# Exploration status lives in the filename, so checking a doc off renames
	# it and silently breaks every reference spelling the old name. 25 had
	# rotted before this check existed — including four public site/ legal
	# pages and three package READMEs. Absolute, not a ratchet: zero is
	# reachable today, so every new break is a fresh mistake. (0421.)
	- name: Exploration references resolve (0421)
	run: pnpm check:exploration-links

	# The backlog grows ~85 documents a month and nothing closes one. Ratchet
	# only — the count may not EXCEED the committed baseline, so a green repo
	# stays green and the fix is always a one-line frontmatter edit. Needs the
	# full history checked out below: creation dates are what make a document
	# "stale", and a shallow clone has none. (0421.)
	- name: Exploration backlog ratchet (0421)
	run: pnpm check:exploration-fallow

	- name: Publish closure (0220)
	run: pnpm check:publish-closure

	# `changeset version` only runs on push to main, so a malformed changeset
	# lands green on its own PR and reds the release instead — the Version
	# Packages PR stops refreshing and nothing publishes. Four changesets
	# mixing ignored and publishable packages did exactly that for 12 days.
	# changesets aborts on the first offender, so this reports all of them.
	- name: Changesets assemble into a release plan (0265)
	run: pnpm check:changesets

	# site/public/llms-full.txt is generated from the docs but committed, and
	# CI does not build site/ on pull requests — so an MDX edit that skips the
	# regeneration ships a stale artifact and nothing says so. Commit 4c00be2c7
	# renumbered an exploration reference in decisions.mdx and left the file
	# quoting the dead number until an unrelated commit happened to regenerate
	# it. Byte equality against a fresh render; the fix is one command.
	- name: llms-full.txt matches the docs
	run: pnpm check:llms-full

	# The hub Dockerfile hand-lists the packages it COPYs. A dependency added
	# three packages away silently staled that list, and nothing failed until
	# a six-minute image build died on "Cannot find module" — the same build
	# Railway runs, so the first symptom can be a failed deploy. (0415.)
	- name: Hub image copies its dependency closure (0415)
	run: pnpm check:hub-image

	- name: Electron parity (0238)
	run: pnpm check:electron-parity

	# The storybook plugins shim must mirror every @xnetjs/plugins name the
	# story bundle imports; drift broke Visual UI Capture three times (0279,
	# 0280, #412) six minutes into an optional job. Fail here in seconds
	# instead — visual-capture is continue-on-error by contract. (0283.)
	- name: Storybook plugins shim conformance (0283)
	run: pnpm check:storybook-shim

	# Workflow files themselves: an unparseable workflow made GitHub emit a
	# phantom failed run on EVERY push for weeks (exploration 0283). Severity
	# floor `warning` keeps shellcheck's style/info pedantry out of the gate.
	- name: Lint workflow files (actionlint, 0283)
	env:
	SHELLCHECK_OPTS: --severity=warning
	run: \|
	bash <(curl -fsSL https://raw.githubusercontent.com/rhysd/actionlint/v1.7.12/scripts/download-actionlint.bash) 1.7.12
	./actionlint -color

	# Validate changelog fragments and dry-run PR-number resolution on PRs, so a
	# malformed entry (which would otherwise fail only the production deploy and
	# freeze the whole changelog) or one that would ship numberless fails the PR
	# instead (exploration 0203). Deliberately light: plain Node + on-the-fly tsx,
	# no workspace install — both scripts are dependency-free / type-only.
	changelog:
	runs-on: ubuntu-latest
	# Fail-closed (0372): named as an EXCLUSION of the one event we mean to
	# skip, not an allow-list of two. `push` has no PR to validate against;
	# every other event — known or not — runs the check rather than skipping
	# into a green it did not earn. See CHANGELOG_BASE_REF below: this job has
	# a second fail-open mouth, and flipping the gate without feeding it would
	# only move the silent pass one level down.
	if: github.event_name != 'push'
	steps:
	- uses: actions/checkout@v4
	with:
	fetch-depth: 0
	- uses: actions/setup-node@v4
	with:
	node-version: '23'

	- name: Validate changelog fragments
	run: npx --yes tsx site/scripts/validate-changelog.ts

	- name: Dry-run PR-number resolution
	run: node scripts/changelog/resolve-prs.mjs --check
	env:
	GITHUB_TOKEN: ${{ github.token }}
	# Fail-closed (0372): resolve-prs.mjs treats an EMPTY base ref as
	# "local run" and downgrades every unresolved fragment from an error
	# to a warning — so on a non-PR event this step would exit 0 having
	# gated nothing. `pull_request` is undefined off a PR, hence the
	# fallback chain, ending at `origin/main` (available because the
	# checkout above is fetch-depth: 0).
	CHANGELOG_BASE_REF: ${{ github.event.pull_request.base.sha \|\| github.event.merge_group.base_sha \|\| 'origin/main' }}

	# The heavy half of the old lint job: build the workspace and typecheck it.
	# typecheck genuinely needs `^build` (turbo builds dependencies first), so it
	# stays coupled to the build. Runs in parallel with `lint`, `test`, and
	# `editor-ux`; it is the pipeline's long pole.
	typecheck:
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@v4
	- uses: ./.github/actions/setup

	- name: Build & Type check
	run: pnpm turbo run build typecheck

	# The committed API reports (packages//etc/.api.md) are the record of
	# what we promise to keep working. api-extractor fails when the built
	# surface no longer matches the report, so a change to what we export
	# cannot land without showing up as a reviewable diff. Must run AFTER the
	# build -- it reads dist/*.d.ts, which is why it is not in `lint` (that
	# job is deliberately build-free, exploration 0193). (0370)
	- name: API report drift (0370)
	run: pnpm check:api-report

	# publint over every publishable package: can a consumer actually install
	# and import this? Upstream of any versioning promise -- a correct semver
	# bump on a package whose export map doesn't resolve is worth nothing.
	# Errors only; the suggestion tier is unpassable by design. (0370)
	- name: Packaging validity (0370)
	run: pnpm check:packaging

	# Charter §7 "Floor": software is how working computers become waste, so
	# a change that raises what a first load costs is a change that retires
	# someone's hardware. Ratchets the critical-path bytes against
	# footprint-baseline.json. Lives here rather than in `lint` for the same
	# reason check:api-report does -- it reads build output, and `lint` is
	# deliberately build-free (0193). Its negative control runs in `lint`
	# beside the other gate controls. (0434)
	- name: Footprint budget (0434)
	run: pnpm check:footprint-budget

	test:
	runs-on: ubuntu-latest
	strategy:
	matrix:
	shard: [1/3, 2/3, 3/3]
	steps:
	- uses: actions/checkout@v4
	with:
	fetch-depth: 0
	- uses: ./.github/actions/setup

	# No build step: vitest.config.ts aliases every @xnetjs/* import to the
	# package's src/, and no test imports from dist/, so the workspace build
	# is dead weight for the test run (exploration 0193). better-sqlite3 — the
	# only native dependency tests touch — is rebuilt in the setup composite.

	# No --coverage: CI never enforced thresholds (vitest.config.ts skips
	# them when CI=true because shards see partial coverage) and the codecov
	# upload had no consumer — no badge, no gate (exploration 0294). Local
	# full runs still enforce the configured thresholds.
	# FAIL-CLOSED EVENT DISPATCH (0372). This used to be two mutually
	# exclusive `if:` gates — `push` → full run, `pull_request` → changed-only.
	# On ANY other event (merge_group, workflow_call, workflow_dispatch,
	# schedule) neither matched, both steps were skipped, and the job
	# SUCCEEDED having executed zero tests. `test (1/3..3/3)` are required
	# checks in the `Protect main` ruleset and GitHub counts a skipped job as
	# passing, so the gate reported green on an untested tree — it failed
	# safe-looking, which is worse than failing loudly.
	#
	# The scope is now resolved in ONE place with the unknown event falling
	# through to the FULL suite. An unforeseen trigger costs CI minutes; it
	# never costs a false green. Narrowing (`--changed`) must be opted into
	# by name, never arrived at by default.
	- name: Resolve test scope
	id: scope
	env:
	EVENT_NAME: ${{ github.event_name }}
	BASE_REF: ${{ github.base_ref }}
	run: \|
	case "$EVENT_NAME" in
	pull_request) echo "args=--changed=origin/$BASE_REF --passWithNoTests" >> "$GITHUB_OUTPUT" ;;
	*) echo "args=" >> "$GITHUB_OUTPUT" ;;
	esac

	- name: Run tests (shard ${{ matrix.shard }})
	run: pnpm vitest run --shard=${{ matrix.shard }} ${{ steps.scope.outputs.args }}

	editor-ux:
	runs-on: ubuntu-latest
	timeout-minutes: 20
	# Fail-closed (0372): `editor-ux` is a REQUIRED check, so a skip reports
	# green. Excluding `push` (where the merged result is already covered by
	# the full `test` run) leaves every other event running the suite. Nothing
	# in this job reads PR context, so the unknown-event path is sound as-is.
	if: github.event_name != 'push'
	steps:
	- uses: actions/checkout@v4
	- uses: ./.github/actions/setup

	- name: Build packages
	run: pnpm turbo run build --filter='./packages/*'

	# Cache the Playwright browser binaries (~/.cache/ms-playwright), keyed on
	# the pinned @playwright/test version and the browser set this job needs.
	# The big cost is the browser download; OS deps are not in this path, so
	# on a cache hit we still run `install-deps` (apt only) to keep webkit's
	# libraries present. (Exploration 0193.)
	- name: Cache Playwright browsers
	id: playwright-cache
	uses: actions/cache@v4
	with:
	path: ~/.cache/ms-playwright
	key: playwright-${{ runner.os }}-1.58.1-chromium-webkit

	- name: Install Playwright browsers (cache miss)
	if: steps.playwright-cache.outputs.cache-hit != 'true'
	run: pnpm --filter @xnetjs/e2e-tests exec playwright install --with-deps chromium webkit

	- name: Install Playwright system deps (cache hit)
	if: steps.playwright-cache.outputs.cache-hit == 'true'
	run: pnpm --filter @xnetjs/e2e-tests exec playwright install-deps chromium webkit

	- name: Run focused editor interaction tests
	run: pnpm --filter @xnetjs/editor exec vitest run src/blocknote/collab-convergence.test.tsx src/blocknote/doc-utils.test.ts src/blocknote/legacy-import.test.ts

	- name: Start web app
	run: pnpm --filter xnet-web dev --host 127.0.0.1 --port 5173 > /tmp/xnet-web.log 2>&1 &

	- name: Wait for web app
	run: for _ in {1..60}; do curl -fsS http://127.0.0.1:5173 >/dev/null && exit 0; sleep 2; done; echo "web app did not start"; exit 1

	- name: Run editor UX smoke tests (desktop + mobile)
	run: pnpm --filter @xnetjs/e2e-tests exec playwright test src/editor-ux.spec.ts src/editor-ux-mobile.spec.ts src/safety-ui.spec.ts --project=chromium --project=mobile-chromium --fail-on-flaky-tests
	env:
	CI: 'true'
	PLAYWRIGHT_TEST_BASE_URL: 'http://127.0.0.1:5173'

	# Desktop only: the mobile shell doesn't render sections at all (0388
	# open question), so this spec has nothing to assert under mobile-chromium.
	- name: Run left-nav section tests (desktop)
	run: pnpm --filter @xnetjs/e2e-tests exec playwright test src/left-nav.spec.ts --project=chromium --fail-on-flaky-tests
	env:
	CI: 'true'
	PLAYWRIGHT_TEST_BASE_URL: 'http://127.0.0.1:5173'

	# Preserve the screenshots these smoke specs already take (they were
	# written to tmp/playwright and discarded). The richer changed-UI gallery
	# lives in the visual-capture workflow; this just keeps debug evidence.
	- name: Upload editor UX screenshots
	if: always()
	uses: actions/upload-artifact@v4
	with:
	name: editor-ux-screenshots
	# Playwright runs with cwd=tests/e2e, so screenshots land there; the
	# root glob is a fallback for any spec run from the repo root.
	path: \|
	tests/e2e/tmp/playwright/*/.png
	tmp/playwright/*/.png
	if-no-files-found: ignore
	retention-days: 14

	# Electron desktop e2e (exploration 0238, L2 + L3): the cross-client
	# convergence matrix (web↔web · electron↔web · electron↔electron, over ws and
	# best-effort webrtc, incl. offline→reconnect) and the `_electron.launch()` app
	# smoke (launch, no-console-errors, SQLite persist across restart, xnet://
	# deep-link routing, core canvas flow). Headless under xvfb.
	#
	# Native-module ABI: `deps:electron` rebuilds the shared better-sqlite3 for
	# Electron's ABI so the unpacked app can load it — which means plain Node can't.
	# The in-process hub therefore runs under Electron's OWN Node runtime
	# (ELECTRON_RUN_AS_NODE via XNET_HUB_ELECTRON_BIN), so the hub and the app share
	# the one Electron-ABI copy. `electron-cache: 'true'` caches ~/.cache/electron so
	# the rebuild is paid once. `--fail-on-flaky-tests` quarantines flakes.
	electron-e2e:
	runs-on: ubuntu-latest
	timeout-minutes: 30
	# Fail-closed (0372): same shape as `editor-ux` — skip only the `push`
	# case we deliberately exempt, run on everything else. This job reads no
	# PR context either (fetch-depth: 0 is for history, not for a base ref).
	if: github.event_name != 'push'
	steps:
	- uses: actions/checkout@v4
	with:
	fetch-depth: 0
	- uses: ./.github/actions/setup
	with:
	electron-cache: 'true'

	- name: Build packages + Electron renderer/main
	run: pnpm turbo run build --filter=xnet-desktop --filter=@xnetjs/hub

	- name: Rebuild native modules for Electron (unpacked app)
	run: pnpm --filter xnet-desktop run deps:electron

	- name: Install desktop test services
	run: sudo apt-get update && sudo apt-get install -y xvfb dbus-x11 gnome-keyring libsecret-1-0

	- name: Cache Playwright browser
	id: playwright-cache
	uses: actions/cache@v4
	with:
	path: ~/.cache/ms-playwright
	key: playwright-${{ runner.os }}-1.58.1-chromium

	- name: Install Playwright Chromium (cache miss)
	if: steps.playwright-cache.outputs.cache-hit != 'true'
	run: pnpm --filter @xnetjs/e2e-tests exec playwright install --with-deps chromium

	- name: Install Playwright system deps (cache hit)
	if: steps.playwright-cache.outputs.cache-hit == 'true'
	run: pnpm --filter @xnetjs/e2e-tests exec playwright install-deps chromium

	# Resolve the Electron binary and expose it as the hub's Node runtime, then
	# smoke the two ABI-sensitive paths before the full run: better-sqlite3 must
	# load under Electron's ABI, and the hub must start under ELECTRON_RUN_AS_NODE.
	- name: Probe Electron launch (diagnostic)
	run: \|
	ELECTRON_BIN=$(cd apps/electron && node -e "process.stdout.write(require('electron'))")
	echo "electron binary: $ELECTRON_BIN"
	echo "XNET_HUB_ELECTRON_BIN=$ELECTRON_BIN" >> "$GITHUB_ENV"
	echo "--- better-sqlite3 under Electron ABI ---"
	(cd apps/electron && ELECTRON_RUN_AS_NODE=1 "$ELECTRON_BIN" -e "new (require('better-sqlite3'))(':memory:'); console.log('better-sqlite3 OK under Electron')") \
	\|\| echo "PROBE FAILED: better-sqlite3 cannot load under Electron"
	echo "--- hub starts under ELECTRON_RUN_AS_NODE ---"
	ELECTRON_RUN_AS_NODE=1 timeout 25 "$ELECTRON_BIN" --import tsx packages/hub/src/cli.ts \
	--port 39997 --no-auth --storage memory 2>&1 \| head -25 \|\| true

	- name: Run convergence matrix + Electron app smoke (xvfb)
	run: \|
	bash scripts/with-linux-test-keyring.sh \
	xvfb-run --auto-servernum pnpm --filter @xnetjs/e2e-tests exec \
	playwright test src/sync-matrix.spec.ts src/electron-smoke.spec.ts \
	--project=chromium --project=electron --fail-on-flaky-tests
	env:
	CI: 'true'
	XNET_ELECTRON_NO_SANDBOX: '1'

	- name: Upload Electron e2e artifacts
	if: failure()
	uses: actions/upload-artifact@v4
	with:
	name: electron-e2e-artifacts
	path: \|
	tests/e2e/test-results/**
	tests/e2e/tmp/**
	if-no-files-found: ignore
	retention-days: 14

	# Rust core conformance (exploration 0257): the portable kernel in
	# rust/xnet-core must reproduce the shared golden vectors in conformance/vectors
	# byte-for-byte — the same L0–L3 set the TypeScript core is gated on. Without
	# this job a non-TS implementation can drift from the protocol undetected.
	# cargo is preinstalled on ubuntu-latest runners, so no toolchain action needed.
	conformance-rust:
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@v4
	- name: Cache cargo
	uses: actions/cache@v4
	with:
	path: \|
	~/.cargo/registry
	~/.cargo/git
	rust/xnet-core/target
	key: cargo-${{ runner.os }}-${{ hashFiles('rust/xnet-core/Cargo.lock') }}
	- name: Rust kernel conformance vectors
	run: cargo test --manifest-path rust/xnet-core/Cargo.toml --locked

Navigation Menu

Repository navigation

chore(release): version packages #2163

Workflow file

chore(release): version packages #2163

Uh oh!

Workflow file for this run