Files
screenci__screenci/cli.ts
T
2026-06-29 16:17:56 +03:00

3852 lines
120 KiB
TypeScript

import { spawn } from 'child_process'
import type { ChildProcess } from 'child_process'
import { createReadStream } from 'fs'
import {
existsSync,
mkdirSync,
readFileSync,
readdirSync,
realpathSync,
rmSync,
} from 'fs'
import { createHash, randomUUID } from 'crypto'
import { createRequire } from 'module'
import { appendFile, readdir, readFile, stat, writeFile } from 'fs/promises'
import {
delimiter,
dirname,
isAbsolute,
relative as pathRelative,
resolve,
} from 'path'
import { fileURLToPath, pathToFileURL } from 'url'
import { Command, CommanderError } from 'commander'
import pc from 'picocolors'
import { logger } from './src/logger.js'
import {
determinePackageManager,
parsePackageManager,
runInit,
} from './src/init.js'
import type {
RecordingCustomVoiceRef,
RecordingData,
VideoCueTranslationFile,
} from './src/events.js'
import {
SCREENCI_DISABLE_RECORDING_TIMINGS_ENV,
SCREENCI_MOCK_RECORD_ENV,
SCREENCI_LANGUAGES_ENV,
SCREENCI_VALUES_OVERRIDES_ENV,
SCREENCI_RECORD_OPTIONS_ENV,
isUploadExistingEnabled,
} from './src/runtimeMode.js'
import { DEFAULT_RECORD_UPLOAD_POLICY } from './src/defaults.js'
import type { VoiceKey } from './src/voices.js'
import type { RecordUploadPolicy, ScreenCIConfig } from './src/types.js'
import {
findDuplicateTitles,
formatDuplicateTitlesMessage,
} from './src/titleValidation.js'
import {
createLinkSessionSpec,
deletePersistedLinkSessionSpec,
getCliLinkSessionApiUrl,
getDevBackendUrl,
getDevFrontendUrl,
getLinkSessionFilePath,
getScreenCIEnvironment,
isStoredLinkSessionReusable,
readPersistedLinkSessionSpec,
SCREENCI_LINK_SESSION_FILE,
writePersistedLinkSessionSpec,
} from './src/linkSession.js'
import type {
LinkSessionStatus,
PersistedLinkSessionSpec,
} from './src/linkSession.js'
import { openUrlInBrowser } from './src/openBrowser.js'
import { maybeExtractVoiceSampleAudio } from './src/voiceSampleAudio.js'
// Re-export the environment-aware URL helpers so existing importers (and tests)
// can keep importing them from the CLI entrypoint.
export { getCliLinkSessionApiUrl, getDevBackendUrl, getDevFrontendUrl }
const SCREENCI_MOCK_RECORD_DOCS_URL =
'https://screenci.com/docs/reference/cli/#--mock-record'
const SCREENCI_RECORD_DOCS_URL =
'https://screenci.com/docs/reference/cli/#screenci-record'
// Records the recordId of the most recent `screenci record` upload so
// `screenci info` can report exactly the run that was just made.
const SCREENCI_LAST_RECORD_FILE = 'last-record.json'
const SCREENCI_LINK_SESSION_POLL_INTERVAL_MS = 2_000
// `record --poll` keeps a non-interactive session (an agent or CI) waiting for
// sign-in. We poll on a slower cadence than the interactive loop so a long wait
// for a human to click the link does not hammer the backend.
const SCREENCI_LINK_SESSION_POLL_FLAG_INTERVAL_MS = 5_000
// A waiting `record` does not wait forever: after this long without a completed
// sign-in we stop polling so a non-interactive agent run does not hang. The link
// stays valid, so the command can be rerun. The default can be overridden with
// SCREENCI_POLL_AUTH_TIMEOUT_MS (milliseconds).
const SCREENCI_LINK_SESSION_POLL_FLAG_TIMEOUT_MS = 5 * 60 * 1_000
// An interactive `record` (a human at a terminal) waits longer before giving up,
// since a person may take a while to open the link and finish signing in. Also
// overridable with SCREENCI_POLL_AUTH_TIMEOUT_MS.
const SCREENCI_LINK_SESSION_INTERACTIVE_TIMEOUT_MS = 15 * 60 * 1_000
const require = createRequire(import.meta.url)
type PlaywrightListReportSuite = {
title?: string
specs?: Array<{ title: string }>
suites?: PlaywrightListReportSuite[]
}
type PlaywrightListReport = {
suites?: PlaywrightListReportSuite[]
errors?: Array<{
message?: string
snippet?: string
}>
}
/**
* Reports whether the current session can complete an interactive browser
* sign-in. A session is interactive only when both stdin and stdout are
* attached to a terminal and no signal marks the run as automated. This is the
* proxy for "a human is present to open the sign-in link" — it does not attempt
* to identify any particular caller (CI, a piped shell, or an automated tool).
*
* Dependency-injected so tests can force a value without a real terminal.
*/
export function detectInteractiveSession(
env: NodeJS.ProcessEnv = process.env,
stdout: { isTTY?: boolean } = process.stdout,
stdin: { isTTY?: boolean } = process.stdin
): boolean {
if (env.SCREENCI_NONINTERACTIVE === '1') return false
if (env.CI === 'true') return false
return Boolean(stdout.isTTY) && Boolean(stdin.isTTY)
}
export function collectPlaywrightListTitles(
suites: readonly PlaywrightListReportSuite[]
): string[] {
const titles: string[] = []
const visitSuite = (suite: PlaywrightListReportSuite) => {
for (const spec of suite.specs ?? []) {
titles.push(spec.title)
}
for (const child of suite.suites ?? []) {
visitSuite(child)
}
}
for (const suite of suites) {
visitSuite(suite)
}
return titles
}
function parsePlaywrightListReport(stdout: string): PlaywrightListReport {
return JSON.parse(stdout) as PlaywrightListReport
}
function extractPlaywrightDiscoveryError(output: string): string | null {
try {
const report = parsePlaywrightListReport(output)
const firstError = report.errors?.find(
(entry) =>
typeof entry.message === 'string' && entry.message.trim() !== ''
)
if (!firstError?.message) {
return null
}
const message = firstError.message.trim()
const snippet = firstError.snippet?.trim()
return snippet ? `${message}\n\n${snippet}` : message
} catch {
return null
}
}
function logScreenCISecretGuide(): void {
logger.info(`Guide: ${pc.cyan(SCREENCI_RECORD_DOCS_URL)}`)
}
function getSuggestedScreenciCommand(
command: 'record' | 'test',
flags = ''
): string {
const suffix = flags ? ` ${flags}` : ''
const pm = determinePackageManager()
if (pm === 'pnpm') return `pnpm exec screenci ${command}${suffix}`
if (pm === 'yarn') return `yarn screenci ${command}${suffix}`
return `npx screenci ${command}${suffix}`
}
async function collectDiscoveredTestTitles(
configPath: string,
additionalArgs: string[],
env: NodeJS.ProcessEnv
): Promise<string[]> {
const listArgs = [
'test',
'--config',
configPath,
...additionalArgs,
'--list',
'--reporter=json',
]
const spawnSpec = resolvePlaywrightSpawnSpec(listArgs, dirname(configPath))
return await new Promise<string[]>((resolve, reject) => {
const child = spawn(spawnSpec.command, spawnSpec.args, {
stdio: ['inherit', 'pipe', 'pipe'],
...(spawnSpec.shell !== undefined ? { shell: spawnSpec.shell } : {}),
...(spawnSpec.windowsVerbatimArguments !== undefined
? {
windowsVerbatimArguments: spawnSpec.windowsVerbatimArguments,
}
: {}),
env,
})
const childSignals = forwardChildSignals(
child,
'screenci title validation',
{
killTree: process.platform !== 'win32',
exitParentOnForward: true,
}
)
let stdout = ''
let stderr = ''
child.stdout?.setEncoding?.('utf8')
child.stderr?.setEncoding?.('utf8')
child.stdout?.on('data', (chunk) => {
stdout += chunk
})
child.stderr?.on('data', (chunk) => {
stderr += chunk
})
child.on('close', (code, signal) => {
const forwardedSignal = childSignals.getForwardedSignal()
childSignals.cleanup()
if (forwardedSignal) {
process.kill(process.pid, forwardedSignal)
return
}
if (signal) {
process.kill(process.pid, signal)
return
}
if (code !== 0) {
if (stderr.trim() === '' && stdout.trim() === '') {
resolve([])
return
}
const parsedDiscoveryError =
stderr.trim() === '' ? extractPlaywrightDiscoveryError(stdout) : null
reject(
new Error(
stderr.trim() ||
parsedDiscoveryError ||
stdout.trim() ||
'Playwright test discovery failed'
)
)
return
}
try {
if (stdout.trim() === '') {
resolve([])
return
}
const report = parsePlaywrightListReport(stdout)
resolve(collectPlaywrightListTitles(report.suites ?? []))
} catch (error) {
reject(error)
}
})
child.on('error', (err) => {
childSignals.cleanup()
reject(err)
})
})
}
async function validateUniqueDiscoveredTestTitles(
configPath: string,
additionalArgs: string[],
env: NodeJS.ProcessEnv
): Promise<void> {
const titles = await collectDiscoveredTestTitles(
configPath,
additionalArgs,
env
)
const duplicates = findDuplicateTitles(titles)
if (duplicates.length > 0) {
throw new Error(formatDuplicateTitlesMessage(duplicates))
}
}
function resolveRecordingFileCandidates(
filePath: string,
configDir: string,
sourceFilePath?: string
): string[] {
const sourceFileCandidate =
typeof sourceFilePath === 'string'
? resolve(configDir, dirname(sourceFilePath), filePath)
: null
return [
filePath,
...(sourceFileCandidate ? [sourceFileCandidate] : []),
resolve(configDir, 'recordings', filePath),
resolve(configDir, pathRelative('/app', filePath)),
]
}
async function readRecordingFile(
filePath: string,
configDir: string,
sourceFilePath?: string
): Promise<{ buffer: Buffer; resolvedPath: string } | null> {
for (const candidate of resolveRecordingFileCandidates(
filePath,
configDir,
sourceFilePath
)) {
try {
return { buffer: await readFile(candidate), resolvedPath: candidate }
} catch {
// try next candidate
}
}
return null
}
function contentTypeForPath(filePath: string): string {
const ext = filePath.split('.').pop()?.toLowerCase() ?? 'bin'
const contentTypeMap: Record<string, string> = {
png: 'image/png',
jpg: 'image/jpeg',
jpeg: 'image/jpeg',
gif: 'image/gif',
webp: 'image/webp',
mp4: 'video/mp4',
webm: 'video/webm',
mp3: 'audio/mpeg',
wav: 'audio/wav',
m4a: 'audio/mp4',
aac: 'audio/aac',
svg: 'image/svg+xml',
}
return contentTypeMap[ext] ?? 'application/octet-stream'
}
type CustomVoiceRefLike = {
assetHash: string
assetPath: string
// SHA-256 of the original sample file. The backend keys the clone cache on
// this so re-encoded uploads (audio extracted from a video) do not re-clone.
sampleHash?: string
}
/** What an uploaded file actually is, used to label it in the upload log. */
type UploadAssetKind = 'overlay' | 'audio' | 'clip' | 'voice'
/** Human label per upload kind, shown as `<label> uploaded: <path>`. */
const UPLOAD_ASSET_LABEL: Record<UploadAssetKind, string> = {
overlay: 'Overlay',
audio: 'Audio',
// A `videoCueStart` asset: a media file placed on the narration timeline
// (e.g. via `narration({ en: { intro: { media } } })`).
clip: 'Narration',
voice: 'Voice',
}
type PreparedUploadAsset = {
kind: UploadAssetKind
fileHash: string
path: string
size: number
name?: string
fileBuffer?: Buffer
contentType?: string
// Skip the existence check and always push to the server.
// Used for per-recording ephemeral captures (screen audio) that are never
// shared across recordings and must not be silently skipped.
alwaysUpload?: boolean
// The local file was absent when assets were collected, so its bytes (and, for
// overlays, its content hash) are not available. The asset's identity must be
// recovered from a previous upload of this video (matched by path/name) before
// the recording is started. Once recovered, fileHash/size/contentType are
// filled in and `assumedUploaded` is set. See resolveMissingUploadAssets.
needsResolve?: boolean
// Set once a missing local file has been matched to a previously uploaded
// asset. The asset is referenced by its known hash with no local bytes; the
// backend existence check confirms it is still stored.
assumedUploaded?: boolean
}
type UploadCandidate = {
entry: string
videoName: string
data: RecordingData
preparedUploadAssets: PreparedUploadAsset[]
}
export type UploadStudioInfo = { held: true } | { applied: true }
export type StudioUploadNotice = {
videoName: string
videoId: string | null
studio: UploadStudioInfo
}
export function formatStudioUrl(
appUrl: string,
projectId: string,
videoId: string
): string {
// The video hub resolves `?studio` to the right language page and scrolls it
// to Studio, so we never need to guess the language in the printed link.
return `${appUrl}/project/${projectId}/video/${videoId}?studio`
}
type OrgPlan = 'free' | 'starter' | 'business'
type UploadJobResult = {
projectId: string | null
videoId: string | null
hadFailure: boolean
videoName: string
failureMessage?: string
recordId: string
studio?: UploadStudioInfo
plan?: OrgPlan
}
type UploadProgressStatus = 'success' | 'failure' | 'cancelled'
class UploadAssetError extends Error {
constructor(message: string) {
super(message)
this.name = 'UploadAssetError'
}
}
class UploadCancelledError extends Error {
constructor(message = 'Upload cancelled') {
super(message)
this.name = 'UploadCancelledError'
}
}
class PartialUploadError extends Error {
constructor(message = 'Not all recordings succeeded to upload.') {
super(message)
this.name = 'PartialUploadError'
}
}
class RecordFailureHintError extends Error {
readonly cause: Error
constructor(cause: Error) {
super(cause.message)
this.name = cause.name
this.cause = cause
}
}
// Thrown when we wait for a browser sign-in and the deadline passes before it
// completes. The link stays valid, so the message tells the user to sign in and
// rerun. A timed-out sign-in means recording did not happen, so the caller maps
// this to a non-zero exit (it is a failure, not the clean print-and-exit handoff
// used by CI).
export class SignInTimeoutError extends Error {
constructor(message: string) {
super(message)
this.name = 'SignInTimeoutError'
}
}
function isUploadCancelledError(err: unknown): boolean {
return (
err instanceof UploadCancelledError ||
(err instanceof Error &&
(err.name === 'AbortError' || err.name === 'UploadCancelledError'))
)
}
function isPartialUploadError(err: unknown): boolean {
return err instanceof PartialUploadError
}
function isRecordFailureHintError(err: unknown): err is RecordFailureHintError {
return err instanceof RecordFailureHintError
}
function isUploadAssetError(err: unknown): boolean {
return err instanceof UploadAssetError
}
export async function withUploadRetry<T>(
fn: () => Promise<T>,
signal: AbortSignal | undefined,
maxAttempts = 3
): Promise<T> {
let lastErr: unknown
for (let attempt = 0; attempt < maxAttempts; attempt++) {
try {
return await fn()
} catch (err) {
if (isUploadCancelledError(err)) throw err
lastErr = err
}
}
throw lastErr
}
function formatUploadProgressLine(
videoName: string,
status?: UploadProgressStatus
): string {
switch (status) {
case undefined:
return `${pc.cyan('...')} Uploading "${videoName}"`
case 'success':
return `${pc.green('✔')} Uploaded "${videoName}"`
case 'failure':
return `${pc.red('✖')} Failed to upload "${videoName}"`
case 'cancelled':
return `${pc.yellow('!')} Cancelled "${videoName}"`
default: {
const exhaustiveCheck: never = status
return exhaustiveCheck
}
}
}
function createUploadProgressReporter(
videoNames: readonly string[],
_verbose: boolean
): {
complete: (index: number, status: UploadProgressStatus) => void
info: (message: string) => void
} {
return {
complete(index, status) {
logger.info(
formatUploadProgressLine(videoNames[index] ?? 'unknown', status)
)
},
info(message) {
logger.info(message)
},
}
}
function shouldKeepRecordedArtifacts(): boolean {
return process.env.DEBUG === 'true'
}
function cleanupUploadedRecordingDir(screenciDir: string, entry: string): void {
if (shouldKeepRecordedArtifacts()) return
try {
rmSync(resolve(screenciDir, entry), { recursive: true, force: true })
} catch (error) {
logger.warn(
`Uploaded recording cleanup failed for "${entry}": ${error instanceof Error ? error.message : String(error)}`
)
}
}
async function loadUploadCandidate(
screenciDir: string,
entry: string,
verbose: boolean
): Promise<UploadCandidate | null> {
const dataJsonPath = resolve(screenciDir, entry, 'data.json')
if (!existsSync(dataJsonPath)) {
if (verbose) logger.info(`Skipping "${entry}": no data.json found`)
return null
}
let data: RecordingData
try {
const raw = await readFile(dataJsonPath, 'utf-8')
data = JSON.parse(raw) as RecordingData
} catch {
logger.warn(`Failed to read ${dataJsonPath}, skipping`)
return null
}
const videoName = data.metadata?.videoName ?? entry
const preparedUploadAssets = await collectUploadAssets(
data,
resolve(screenciDir, '..')
)
// The recording data is annotated with asset hashes later, in
// uploadRecordingCandidate, after any locally missing assets have been
// resolved against a previous upload. Annotating here would strip the asset
// paths the resolve step needs to match on.
return {
entry,
videoName,
data,
preparedUploadAssets,
}
}
async function uploadRecordingCandidate(
candidate: UploadCandidate,
screenciDir: string,
projectName: string,
apiUrl: string,
secret: string,
elevenLabsApiKey: string | undefined,
verbose: boolean,
uploadAbort: ReturnType<typeof createUploadAbortController>,
progressReporter: {
complete: (index: number, status: UploadProgressStatus) => void
info: (message: string) => void
},
progressIndex: number,
recordId: string,
expectedScreenshotCount: number
): Promise<UploadJobResult> {
const { entry, videoName, data: rawData, preparedUploadAssets } = candidate
let projectId: string | null = null
let videoId: string | null = null
let plan: OrgPlan | null = null
try {
uploadAbort.throwIfAborted()
// A screenshot recording uploads its raw page capture (always a PNG) through
// the same recording endpoint a video uses; the renderer reads those bytes as
// the capture. Videos upload recording.mp4. Output kind defaults to 'video'.
const isScreenshot = rawData.output === 'screenshot'
const recordingFileName = isScreenshot
? (rawData.screenshot?.path ?? 'screenshot.png')
: 'recording.mp4'
const recordingContentType = isScreenshot ? 'image/png' : 'video/mp4'
const recordingPath = resolve(screenciDir, entry, recordingFileName)
if (!existsSync(recordingPath)) {
progressReporter.complete(progressIndex, 'failure')
return {
projectId: null,
videoId: null,
hadFailure: true,
videoName,
failureMessage: `Missing ${recordingFileName} for "${videoName}"`,
recordId,
}
}
// Locally missing assets (e.g. gitignored media on CI) carry no bytes and,
// for overlays, no hash. Recover their identity from a previous upload of
// this video before starting, so the recording data references them by hash
// and the backend existence check confirms they are still stored.
const unresolved = await resolveMissingUploadAssets(
preparedUploadAssets,
projectName,
videoName,
apiUrl,
secret,
uploadAbort.signal,
progressReporter
)
if (unresolved.length > 0) {
progressReporter.complete(progressIndex, 'failure')
return {
projectId: null,
videoId: null,
hadFailure: true,
videoName,
failureMessage: formatUnresolvedAssetMessage(videoName, unresolved),
recordId,
}
}
const data = annotateRecordingDataWithAssetHashes(
rawData,
preparedUploadAssets
)
const recordingHash = await hashFile(recordingPath)
const startResponse = await withUploadRetry(
() =>
fetch(`${apiUrl}/cli/upload/start`, {
method: 'POST',
headers: {
'Content-Type': 'application/json',
'X-ScreenCI-Secret': secret,
...(elevenLabsApiKey
? { 'X-ElevenLabs-Api-Key': elevenLabsApiKey }
: {}),
},
body: JSON.stringify({
projectName,
videoName,
data,
recordingHash,
recordId,
...(isScreenshot ? { expectedScreenshotCount } : {}),
expectedAssets: preparedUploadAssets.map((asset) => ({
fileHash: asset.fileHash,
size: asset.size,
path: asset.path,
...(typeof asset.contentType === 'string'
? { contentType: asset.contentType }
: {}),
...(typeof asset.name === 'string' ? { name: asset.name } : {}),
})),
}),
signal: uploadAbort.signal,
}),
uploadAbort.signal
)
if (!startResponse.ok) {
const text = await startResponse.text()
progressReporter.complete(progressIndex, 'failure')
return {
projectId: null,
videoId: null,
hadFailure: true,
videoName,
failureMessage: formatUploadStartFailureMessage(
videoName,
startResponse.status,
text,
secret
),
recordId,
}
}
const startBody = (await startResponse.json()) as {
recordingId: string
projectId: string
videoId?: string
studio?: UploadStudioInfo
plan?: OrgPlan
dependencyErrors?: Array<{
targetName: string
reason: string
detail: string
}>
}
const { recordingId } = startBody
projectId = startBody.projectId
videoId = startBody.videoId ?? null
plan = startBody.plan ?? null
const studio = startBody.studio
// Render-dependency validation failures are reported on the upload response
// so the author sees them at record time. These are enforced server-side:
// the render for this video fails with the dependency error until fixed.
if (startBody.dependencyErrors && startBody.dependencyErrors.length > 0) {
for (const depError of startBody.dependencyErrors) {
logger.error(
`Render dependency error in "${videoName}": ${depError.detail}. This render will fail until it is fixed.`
)
}
}
if (verbose) {
logger.info(`recordingId=${recordingId} projectId=${projectId}`)
logger.info(
`assets=${preparedUploadAssets.length} recordingHash=${recordingHash ?? 'none'}`
)
}
await uploadAssets(
preparedUploadAssets,
apiUrl,
secret,
recordingId,
uploadAbort.signal,
uploadAbort.throwIfAborted,
progressReporter
)
uploadAbort.throwIfAborted()
const fileStat = await stat(recordingPath)
if (verbose) {
logger.info(
`Uploading ${recordingFileName} size=${(fileStat.size / 1024 / 1024).toFixed(1)}MB`
)
}
const recordingResponse = await withUploadRetry(async () => {
const stream = createReadStream(recordingPath)
const abortStream = () => {
stream.destroy(
new UploadCancelledError(`Upload cancelled for "${videoName}"`)
)
}
uploadAbort.signal.addEventListener('abort', abortStream, { once: true })
try {
return await fetch(`${apiUrl}/cli/upload/${recordingId}/recording`, {
method: 'PUT',
headers: {
'Content-Type': recordingContentType,
'Content-Length': String(fileStat.size),
'X-ScreenCI-Secret': secret,
...(elevenLabsApiKey
? { 'X-ElevenLabs-Api-Key': elevenLabsApiKey }
: {}),
},
body: stream as unknown as BodyInit,
signal: uploadAbort.signal,
// @ts-expect-error Node.js fetch supports duplex for streaming
duplex: 'half',
})
} finally {
uploadAbort.signal.removeEventListener('abort', abortStream)
}
}, uploadAbort.signal)
if (!recordingResponse.ok) {
const text = await recordingResponse.text()
progressReporter.complete(progressIndex, 'failure')
return {
projectId,
videoId,
hadFailure: true,
videoName,
failureMessage: `Failed to upload recording for "${videoName}": ${recordingResponse.status} ${text}${hint401(recordingResponse.status, secret)}`,
recordId,
...(plan !== null && { plan }),
}
}
progressReporter.complete(progressIndex, 'success')
cleanupUploadedRecordingDir(screenciDir, entry)
return {
projectId,
videoId,
hadFailure: false,
videoName,
recordId,
...(studio !== undefined && { studio }),
...(plan !== null && { plan }),
}
} catch (err) {
if (isUploadCancelledError(err)) {
progressReporter.complete(progressIndex, 'cancelled')
throw err
}
if (isUploadAssetError(err)) {
progressReporter.complete(progressIndex, 'failure')
return {
projectId,
videoId,
hadFailure: true,
videoName,
failureMessage: err instanceof Error ? err.message : String(err),
recordId,
...(plan !== null && { plan }),
}
}
progressReporter.complete(progressIndex, 'failure')
return {
projectId,
videoId,
hadFailure: true,
videoName,
failureMessage: `Network error uploading "${videoName}": ${err instanceof Error ? err.message : String(err)}`,
recordId,
...(plan !== null && { plan }),
}
}
}
export function attachUploadAbortStdinListener(
input: Pick<NodeJS.ReadStream, 'on' | 'off' | 'pause'>,
onAbort: (signal: NodeJS.Signals) => void
): () => void {
const handleStdinData = (chunk: Buffer | string) => {
const bytes =
typeof chunk === 'string'
? Buffer.from(chunk, 'utf8')
: Buffer.from(chunk)
if (bytes.includes(0x03)) {
onAbort('SIGINT')
}
}
input.on('data', handleStdinData)
return () => {
input.off('data', handleStdinData)
input.pause()
}
}
function createUploadAbortController(activityLabel: string): {
signal: AbortSignal
throwIfAborted: () => void
cleanup: () => void
} {
const controller = new AbortController()
let cleanedUp = false
const cleanupStdinListener = attachUploadAbortStdinListener(
process.stdin,
(signal) => abortUpload(signal)
)
const cleanup = () => {
if (cleanedUp) return
cleanedUp = true
process.off('SIGINT', handleSigint)
process.off('SIGTERM', handleSigterm)
cleanupStdinListener()
}
function abortUpload(signal: NodeJS.Signals) {
if (controller.signal.aborted) return
logger.info(`Received ${signal}, stopping ${activityLabel}...`)
cleanup()
controller.abort(new UploadCancelledError(`${activityLabel} cancelled`))
process.kill(process.pid, signal)
}
const handleSigint = () => abortUpload('SIGINT')
const handleSigterm = () => abortUpload('SIGTERM')
process.on('SIGINT', handleSigint)
process.on('SIGTERM', handleSigterm)
return {
signal: controller.signal,
throwIfAborted: () => {
if (controller.signal.aborted) {
throw controller.signal.reason instanceof Error
? controller.signal.reason
: new UploadCancelledError(`${activityLabel} cancelled`)
}
},
cleanup,
}
}
function resolveSpawnSpec(
cmd: string,
args: string[]
): {
command: string
args: string[]
shell?: boolean
windowsVerbatimArguments?: boolean
} {
if (process.platform !== 'win32') {
return { command: cmd, args }
}
const windowsCmdShims = new Set(['npm', 'npx', 'playwright', 'pnpm'])
if (!windowsCmdShims.has(cmd)) {
return { command: cmd, args }
}
return {
command: process.env.comspec ?? 'cmd.exe',
args: ['/d', '/s', '/c', `"${buildWindowsBatchCommandLine(cmd, args)}"`],
windowsVerbatimArguments: true,
}
}
function quoteWindowsBatchArg(arg: string): string {
if (arg.length === 0) {
return '""'
}
return `"${arg
.replace(/(\\*)"/g, '$1$1\\"')
.replace(/(\\+)$/g, '$1$1')
.replace(/%/g, '%%')}"`
}
function buildWindowsBatchCommandLine(cmd: string, args: string[]): string {
return [resolveWindowsCmdShim(cmd), ...args]
.map(quoteWindowsBatchArg)
.join(' ')
}
function resolveWindowsCmdShim(cmd: string): string {
const shimName = `${cmd}.cmd`
const pathEntries = process.env.PATH?.split(delimiter) ?? []
for (const entry of pathEntries) {
if (!entry) continue
const shimPath = resolve(entry, shimName)
if (existsSync(shimPath)) {
return shimPath
}
}
const bundledShimCommands = new Set(['npm', 'npx'])
if (bundledShimCommands.has(cmd)) {
const bundledShimPath = resolve(dirname(process.execPath), shimName)
if (existsSync(bundledShimPath)) {
return bundledShimPath
}
}
return shimName
}
function isModuleNotFoundError(error: unknown): boolean {
return (
error instanceof Error &&
(error as NodeJS.ErrnoException).code === 'MODULE_NOT_FOUND'
)
}
function resolvePlaywrightCliEntrypoint(searchFrom: string): string {
// Prefer the @playwright/test installed alongside the user's config, since it
// is declared as a peer dependency of the project being recorded.
try {
return require.resolve('@playwright/test/cli', { paths: [searchFrom] })
} catch (error) {
if (!isModuleNotFoundError(error)) throw error
}
// Fall back to the copy resolvable from the screenci CLI's own install. This
// keeps discovery working when Playwright is hoisted to a parent install or
// bundled with the CLI rather than next to the config file.
return require.resolve('@playwright/test/cli')
}
function resolvePlaywrightSpawnSpec(
args: string[],
searchFrom: string
): {
command: string
args: string[]
shell?: boolean
windowsVerbatimArguments?: boolean
} {
const cliEntrypoint = resolvePlaywrightCliEntrypoint(searchFrom)
return {
command: process.execPath,
args: [cliEntrypoint, ...args],
}
}
function forwardChildSignals(
child: ChildProcess,
activityLabel: string,
options: { killTree?: boolean; exitParentOnForward?: boolean } = {}
): { cleanup: () => void; getForwardedSignal: () => NodeJS.Signals | null } {
let forwardedSignal: NodeJS.Signals | null = null
let forceKillTimer: NodeJS.Timeout | null = null
const killTree = options.killTree ?? false
const exitParentOnForward = options.exitParentOnForward ?? false
const killChild = (signal: NodeJS.Signals): void => {
if (child.pid === undefined) return
if (killTree && process.platform !== 'win32') {
try {
process.kill(-child.pid, signal)
return
} catch {
// Fall back to direct child kill below.
}
}
if (!child.killed) {
child.kill(signal)
}
}
const forwardSignal = (signal: NodeJS.Signals) => {
if (forwardedSignal !== null) return
forwardedSignal = signal
if (process.env.SCREENCI_SIGNAL_LOGGING !== 'silent') {
logger.info(`Received ${signal}, stopping ${activityLabel}...`)
}
killChild(signal)
if (exitParentOnForward) {
cleanup()
process.exit(signal === 'SIGINT' ? 130 : 143)
}
forceKillTimer = setTimeout(() => {
if (child.exitCode === null) {
if (process.env.SCREENCI_SIGNAL_LOGGING !== 'silent') {
logger.info(`Forcing ${activityLabel} to stop after timeout...`)
}
killChild('SIGKILL')
process.exit(signal === 'SIGINT' ? 130 : 143)
}
}, 3000)
forceKillTimer.unref()
}
const handleSigint = () => forwardSignal('SIGINT')
const handleSigterm = () => forwardSignal('SIGTERM')
const cleanupStdinListener = attachUploadAbortStdinListener(
process.stdin,
(signal) => {
if (process.env.SCREENCI_SIGNAL_LOGGING !== 'silent') {
logger.info(`Received ${signal}, stopping ${activityLabel}...`)
}
forwardSignal(signal)
}
)
const cleanup = () => {
if (forceKillTimer !== null) {
clearTimeout(forceKillTimer)
}
process.off('SIGINT', handleSigint)
process.off('SIGTERM', handleSigterm)
cleanupStdinListener()
}
process.on('SIGINT', handleSigint)
process.on('SIGTERM', handleSigterm)
return {
cleanup,
getForwardedSignal: () => forwardedSignal,
}
}
function clearRecordingDirectories(dir: string): void {
mkdirSync(dir, { recursive: true })
for (const entry of readdirSync(dir)) {
if (entry === SCREENCI_LINK_SESSION_FILE) continue
rmSync(resolve(dir, entry), { recursive: true, force: true })
}
}
type ScreenCIConfigResolution =
| { kind: 'found'; path: string }
| { kind: 'island-not-entered'; islandConfigPath: string }
| { kind: 'not-found' }
function findScreenCIConfig(customPath?: string): ScreenCIConfigResolution {
if (customPath) {
const resolvedPath = resolve(process.cwd(), customPath)
return existsSync(resolvedPath)
? { kind: 'found', path: resolvedPath }
: { kind: 'not-found' }
}
// Walk up from the current directory looking for a flat `screenci.config.ts`,
// which is what's present when the command runs from inside the `screenci/`
// island. We deliberately do NOT auto-use a nested
// `screenci/screenci.config.ts`: running the CLI from outside the island
// resolves the `screenci` binary from the registry (npx download) rather than
// the version-pinned island install, so it would silently run a different
// version. Instead we detect the island and ask the user to `cd` into it.
let current = process.cwd()
let islandConfigPath: string | undefined
while (true) {
const flatConfig = resolve(current, 'screenci.config.ts')
if (existsSync(flatConfig)) {
return { kind: 'found', path: flatConfig }
}
if (islandConfigPath === undefined) {
const islandConfig = resolve(current, 'screenci', 'screenci.config.ts')
if (existsSync(islandConfig)) {
islandConfigPath = islandConfig
}
}
const parent = dirname(current)
if (parent === current) break
current = parent
}
if (islandConfigPath !== undefined) {
return { kind: 'island-not-entered', islandConfigPath }
}
return { kind: 'not-found' }
}
// Resolve the config path, or log a helpful message and exit. Centralizes the
// `cd screenci` guidance so every command (test/record/info/...) behaves the
// same when invoked from outside the island.
function resolveScreenCIConfigPathOrExit(customPath?: string): string {
const resolution = findScreenCIConfig(customPath)
switch (resolution.kind) {
case 'found':
return resolution.path
case 'island-not-entered': {
const islandDir = dirname(resolution.islandConfigPath)
const relDir = pathRelative(process.cwd(), islandDir) || '.'
logger.error(
`Error: no screenci.config.ts found here, but found ${pc.cyan(
`${relDir}/screenci.config.ts`
)}. Run ${pc.cyan(`cd ${relDir}`)} and rerun the command from there.`
)
return process.exit(1)
}
case 'not-found': {
logger.error(
customPath
? `Error: Config file not found: ${customPath}`
: 'Error: screenci.config.ts not found in the current directory or any parent.'
)
return process.exit(1)
}
}
}
async function hashFile(filePath: string): Promise<string> {
return new Promise((resolveHash, reject) => {
const hash = createHash('sha256')
const stream = createReadStream(filePath)
stream.on('data', (chunk) => hash.update(chunk))
stream.on('error', reject)
stream.on('end', () => resolveHash(hash.digest('hex')))
})
}
async function prepareCustomVoiceAssets(
data: RecordingData,
configDir: string
): Promise<PreparedUploadAsset[]> {
const sourceFilePath = data.metadata?.sourceFilePath
const customVoiceRefsByPath = new Map<string, CustomVoiceRefLike[]>()
for (const event of data.events) {
if (event.type === 'cueStart' && event.translations) {
for (const translation of Object.values(event.translations)) {
if (
typeof translation.voice === 'object' &&
translation.voice !== null &&
'assetPath' in translation.voice &&
typeof translation.voice.assetPath === 'string'
) {
const voiceRef = translation.voice as CustomVoiceRefLike
const refs = customVoiceRefsByPath.get(voiceRef.assetPath) ?? []
refs.push(voiceRef)
customVoiceRefsByPath.set(voiceRef.assetPath, refs)
}
}
}
if (event.type === 'videoCueStart' && event.translations) {
for (const translation of Object.values(event.translations)) {
if (
'text' in translation &&
typeof translation.voice === 'object' &&
translation.voice !== null &&
'assetPath' in translation.voice &&
typeof translation.voice.assetPath === 'string'
) {
const voiceRef = translation.voice as CustomVoiceRefLike
const refs = customVoiceRefsByPath.get(voiceRef.assetPath) ?? []
refs.push(voiceRef)
customVoiceRefsByPath.set(voiceRef.assetPath, refs)
}
}
}
}
const preparedAssets: PreparedUploadAsset[] = []
for (const [voicePath, refs] of customVoiceRefsByPath) {
const resolvedFile = await readRecordingFile(
voicePath,
configDir,
sourceFilePath
)
if (resolvedFile === null) {
const existingHash = refs.find(
(ref) => typeof ref.assetHash === 'string'
)?.assetHash
if (existingHash) {
// The recording already carries this voice's content hash, so reference
// it by that hash. The backend check confirms it is still stored.
for (const ref of refs) {
ref.assetHash = existingHash
}
preparedAssets.push({
kind: 'voice',
fileHash: existingHash,
path: voicePath,
size: 0,
contentType: contentTypeForPath(voicePath),
assumedUploaded: true,
})
continue
}
// No cached hash either. Recover the voice's identity from a previous
// upload of this video, matched by path. resolveMissingUploadAssets fills
// in the hash and writes it back onto the cue refs.
preparedAssets.push({
kind: 'voice',
fileHash: '',
path: voicePath,
size: 0,
contentType: contentTypeForPath(voicePath),
needsResolve: true,
})
continue
}
const { buffer: fileBuffer, resolvedPath } = resolvedFile
// Identify the voice by the ORIGINAL file, so the clone cache survives any
// re-encode of the uploaded bytes (see backend resolveCustomVoice).
const sampleHash = createHash('sha256').update(fileBuffer).digest('hex')
// A voice clone only needs audio, and the clone service caps samples at
// ~11 MB. For video containers (or any oversized file) strip to a small MP3
// before upload; otherwise upload the audio unchanged.
const extracted = await maybeExtractVoiceSampleAudio(
resolvedPath,
fileBuffer.byteLength
)
const uploadBuffer = extracted?.buffer ?? fileBuffer
const contentType =
extracted?.contentType ?? contentTypeForPath(resolvedPath)
const assetHash = createHash('sha256').update(uploadBuffer).digest('hex')
for (const ref of refs) {
ref.assetHash = assetHash
ref.sampleHash = sampleHash
}
preparedAssets.push({
kind: 'voice',
fileHash: assetHash,
path: voicePath,
size: uploadBuffer.byteLength,
fileBuffer: uploadBuffer,
contentType,
})
}
return preparedAssets
}
export async function collectUploadAssets(
data: RecordingData,
configDir: string
): Promise<PreparedUploadAsset[]> {
const sourceFilePath = data.metadata?.sourceFilePath
const assets = new Map<string, PreparedUploadAsset>()
for (const event of data.events) {
if (event.type === 'assetStart') {
// Studio assets have no local file — they are uploaded from the Studio
// page and merged into the recording by the backend.
if ('studio' in event && event.studio === true) continue
// Render dependencies (selected(...)) have no local file: the backend
// resolves the target render's output and injects it at dispatch time.
if ('dependency' in event) continue
if (assets.has(`name:${event.name}`)) continue
const resolvedFile = await readRecordingFile(
event.path,
configDir,
sourceFilePath
)
if (resolvedFile === null) {
// The local file is gone (e.g. gitignored media on CI). Reference it so
// its identity can be recovered from a previous upload of this video.
assets.set(`name:${event.name}`, {
kind: 'overlay',
fileHash: '',
path: event.path,
name: event.name,
size: 0,
needsResolve: true,
})
continue
}
const { buffer: fileBuffer, resolvedPath } = resolvedFile
assets.set(`name:${event.name}`, {
kind: 'overlay',
fileHash: createHash('sha256').update(fileBuffer).digest('hex'),
path: event.path,
name: event.name,
size: fileBuffer.byteLength,
fileBuffer,
contentType: contentTypeForPath(resolvedPath),
})
continue
}
if (event.type === 'audioStart') {
// Studio audio tracks have no local file.
if ('studio' in event && event.studio === true) continue
// Prefer the record-time content hash as the dedup key; a missing local
// file may have been emitted without one, so fall back to the path.
const dedupKey = event.fileHash
? `hash:${event.fileHash}`
: `path:${event.path}`
if (assets.has(dedupKey)) continue
const resolvedFile = await readRecordingFile(
event.path,
configDir,
sourceFilePath
)
if (resolvedFile === null) {
// The local file is gone. If the recording still carries its content
// hash, reference it by that hash (the backend check confirms it is
// stored). Otherwise recover its identity from a previous upload by path.
// Captured screen audio (`__screen`) is per-recording and can never be
// recovered from a prior upload, so it is simply skipped when missing.
if (event.fileHash) {
assets.set(dedupKey, {
kind: 'audio',
fileHash: event.fileHash,
path: event.path,
size: 0,
assumedUploaded: true,
...(event.name === '__screen' && { alwaysUpload: true }),
})
} else if (event.name !== '__screen') {
assets.set(dedupKey, {
kind: 'audio',
fileHash: '',
path: event.path,
size: 0,
needsResolve: true,
})
}
continue
}
assets.set(dedupKey, {
kind: 'audio',
fileHash:
event.fileHash ??
createHash('sha256').update(resolvedFile.buffer).digest('hex'),
path: event.path,
size: resolvedFile.buffer.byteLength,
fileBuffer: resolvedFile.buffer,
contentType: contentTypeForPath(resolvedFile.resolvedPath),
...(event.name === '__screen' && { alwaysUpload: true }),
})
continue
}
if (event.type === 'videoCueStart') {
// Single-language: hash already computed during recording, use assetPath to read file
if (
typeof event.assetHash === 'string' &&
!assets.has(`hash:${event.assetHash}`)
) {
const resolvedFile =
typeof event.assetPath === 'string'
? await readRecordingFile(
event.assetPath,
configDir,
sourceFilePath
)
: null
assets.set(`hash:${event.assetHash}`, {
kind: 'clip',
fileHash: event.assetHash,
path: event.assetPath ?? event.assetHash,
size: resolvedFile?.buffer.byteLength ?? 0,
...(resolvedFile !== null
? {
fileBuffer: resolvedFile.buffer,
contentType: contentTypeForPath(resolvedFile.resolvedPath),
}
: { assumedUploaded: true }),
})
} else if (
typeof event.assetHash !== 'string' &&
typeof event.assetPath === 'string' &&
!assets.has(`path:${event.assetPath}`)
) {
// The media file was gone at record time, so it carries no hash. Recover
// its identity from a previous upload of this video, matched by path.
const resolvedFile = await readRecordingFile(
event.assetPath,
configDir,
sourceFilePath
)
if (resolvedFile === null) {
assets.set(`path:${event.assetPath}`, {
kind: 'clip',
fileHash: '',
path: event.assetPath,
size: 0,
needsResolve: true,
})
}
}
// Multi-language: each translation carries its own hash
if (event.translations) {
for (const translation of Object.values(event.translations)) {
if (typeof translation !== 'object' || translation === null) continue
const assetHash =
'assetHash' in translation &&
typeof translation.assetHash === 'string'
? translation.assetHash
: undefined
const assetPath =
'assetPath' in translation &&
typeof translation.assetPath === 'string'
? translation.assetPath
: undefined
if (assetHash !== undefined) {
if (assets.has(`hash:${assetHash}`)) continue
const resolvedFile =
assetPath !== undefined
? await readRecordingFile(assetPath, configDir, sourceFilePath)
: null
assets.set(`hash:${assetHash}`, {
kind: 'clip',
fileHash: assetHash,
path: assetPath ?? assetHash,
size: resolvedFile?.buffer.byteLength ?? 0,
...(resolvedFile !== null
? {
fileBuffer: resolvedFile.buffer,
contentType: contentTypeForPath(resolvedFile.resolvedPath),
}
: { assumedUploaded: true }),
})
} else if (
assetPath !== undefined &&
!assets.has(`path:${assetPath}`)
) {
const resolvedFile = await readRecordingFile(
assetPath,
configDir,
sourceFilePath
)
if (resolvedFile === null) {
assets.set(`path:${assetPath}`, {
kind: 'clip',
fileHash: '',
path: assetPath,
size: 0,
needsResolve: true,
})
}
}
}
}
}
}
for (const asset of await prepareCustomVoiceAssets(data, configDir)) {
assets.set(`path:${asset.path}`, asset)
}
return [...assets.values()]
}
export function stripVoicePath(
voice: VoiceKey | RecordingCustomVoiceRef,
byPath?: Map<string, string>
): VoiceKey | RecordingCustomVoiceRef {
if (typeof voice !== 'string') {
// A voice recovered from a previous upload has no record-time assetHash; fill
// it in from the resolved-by-path map before the path is dropped.
const assetHash =
voice.assetHash ??
(byPath !== undefined && typeof voice.assetPath === 'string'
? byPath.get(voice.assetPath)
: undefined)
return { assetHash: assetHash as string }
}
return voice
}
export function annotateRecordingDataWithAssetHashes(
data: RecordingData,
assets: PreparedUploadAsset[]
): RecordingData {
// Overlays are matched to their hash by name; every other asset kind (audio,
// narration clip, custom voice) is matched by its file path. Skip placeholder
// hashes that were never resolved so a missing entry stays untouched.
const byName = new Map<string, string>()
const byPath = new Map<string, string>()
for (const asset of assets) {
if (asset.fileHash.length === 0) continue
if (typeof asset.name === 'string') byName.set(asset.name, asset.fileHash)
byPath.set(asset.path, asset.fileHash)
}
return {
...data,
events: data.events.map((event) => {
if (event.type === 'assetStart') {
if ('studio' in event || 'dependency' in event) return event
const fileHash = byName.get(event.name) ?? event.fileHash
return fileHash ? { ...event, fileHash } : event
}
if (event.type === 'audioStart') {
// Studio audio tracks carry no local path; leave them untouched.
if (!('path' in event)) return event
if (event.fileHash) return event
const fileHash = byPath.get(event.path)
return fileHash ? { ...event, fileHash } : event
}
if (event.type === 'cueStart' && event.translations) {
const translations = Object.fromEntries(
Object.entries(event.translations).map(([language, translation]) => {
if (translation.voice === undefined) {
return [language, translation]
}
return [
language,
{
...translation,
voice: stripVoicePath(translation.voice, byPath),
} as typeof translation,
]
})
)
return { ...event, translations }
}
if (event.type !== 'videoCueStart') return event
// Strip assetPath from translations. The hash was either computed during
// recording or recovered from a previous upload (matched by that path).
if (event.translations) {
const translations = Object.fromEntries(
Object.entries(event.translations).map(([language, translation]) => {
if ('assetHash' in translation || 'assetPath' in translation) {
const file = translation as VideoCueTranslationFile
const assetHash =
file.assetHash ??
(file.assetPath !== undefined
? byPath.get(file.assetPath)
: undefined)
const { assetPath: _removed, ...rest } = file
return [
language,
assetHash !== undefined ? { ...rest, assetHash } : rest,
]
}
if ('voice' in translation) {
return [
language,
{
...translation,
...(translation.voice !== undefined
? { voice: stripVoicePath(translation.voice, byPath) }
: {}),
},
]
}
return [language, translation]
})
)
return { ...event, translations }
}
// Single-language: keep the assetHash (recovering it by path if needed) and
// drop the now-redundant assetPath.
const assetHash =
event.assetHash ??
(typeof event.assetPath === 'string'
? byPath.get(event.assetPath)
: undefined)
if (typeof assetHash === 'string') {
const { assetPath: _removed, ...rest } = event
return { ...rest, assetHash }
}
return event
}),
}
}
function hint401(status: number, secret: string): string {
if (status !== 401 || !secret) return ''
const frontendUrl = getDevFrontendUrl()
return `\nThe secret may have been deleted. Check your secrets at ${frontendUrl}/secrets`
}
export function formatUploadStartFailureMessage(
videoName: string,
status: number,
responseText: string,
secret: string
): string {
if (responseText.trim().length > 0) {
try {
const parsed = JSON.parse(responseText) as { error?: unknown }
if (typeof parsed.error === 'string' && parsed.error.trim().length > 0) {
return `${parsed.error}${hint401(status, secret)}`
}
} catch {
// fall back to raw response text
}
return responseText
}
return `Failed to start upload for "${videoName}": ${status}${hint401(status, secret)}`
}
const EXPRESSIVE_TIER_ERROR_PREFIX =
'Expressive narration and style prompts require the Business tier.'
export function formatFailedVideoMessage(
videoName: string,
message: string
): string {
if (message.startsWith(EXPRESSIVE_TIER_ERROR_PREFIX)) {
return [
`${videoName}: ${message}`,
"If you want to keep using the current tier, remove `voice.style` or `modelType: 'expressive'` from the localize `voice`.",
].join('\n')
}
return `${videoName}: ${message}`
}
export function printUploadStartFailureMessage(
videoName: string,
status: number,
responseText: string,
secret: string
): void {
const message = formatUploadStartFailureMessage(
videoName,
status,
responseText,
secret
)
if (responseText.trim().length > 0) {
process.stderr.write(`${message}\n`)
return
}
logger.warn(message)
}
export function displayAssetPath(assetPath: string): string {
if (!isAbsolute(assetPath)) {
return assetPath
}
const rel = pathRelative(process.cwd(), assetPath)
return rel.length > 0 ? rel : assetPath
}
type ResolveAssetRef = { path: string; name?: string | null; kind: string }
type ResolveAssetResult = {
path: string
name?: string | null
fileHash: string | null
size: number | null
contentType: string | null
}
/**
* The backend's response to a resolve-assets request. Each entry is aligned to
* the request's `assets` array; a null `fileHash` means no previously uploaded
* version of that asset was found for this video.
*/
export type ResolveAssetsResponse = { resolved: ResolveAssetResult[] }
/**
* Recovers the identity of assets whose local file was absent during collection
* (no bytes, and for overlays no hash) by matching them, by path or name,
* against a previous upload of the same video. Each resolved asset is mutated in
* place: its `fileHash`, `size`, and `contentType` are filled in and it is
* flagged `assumedUploaded` so the later existence check confirms the bytes are
* still stored rather than trying to upload absent bytes.
*
* Returns the assets that could not be resolved (no previous version), so the
* caller can fail the recording with actionable guidance.
*/
export async function resolveMissingUploadAssets(
assets: PreparedUploadAsset[],
projectName: string,
videoName: string,
apiUrl: string,
secret: string,
signal: AbortSignal,
progressReporter?: { info: (message: string) => void }
): Promise<PreparedUploadAsset[]> {
const pending = assets.filter((asset) => asset.needsResolve === true)
if (pending.length === 0) return []
const refs: ResolveAssetRef[] = pending.map((asset) => ({
path: asset.path,
...(typeof asset.name === 'string' ? { name: asset.name } : {}),
kind: asset.kind,
}))
const res = await withUploadRetry(
() =>
fetch(`${apiUrl}/cli/upload/resolve-assets`, {
method: 'POST',
headers: {
'Content-Type': 'application/json',
'X-ScreenCI-Secret': secret,
},
body: JSON.stringify({ projectName, videoName, assets: refs }),
signal,
}),
signal
)
if (!res.ok) {
const text = await res.text()
throw new UploadAssetError(
`Failed to resolve previously uploaded assets for "${videoName}": ${res.status} ${text}${hint401(res.status, secret)}`
)
}
const body = (await res.json()) as ResolveAssetsResponse
const resolved = Array.isArray(body.resolved) ? body.resolved : []
const logInfo = (message: string) => {
if (progressReporter) progressReporter.info(message)
else logger.info(message)
}
const unresolved: PreparedUploadAsset[] = []
pending.forEach((asset, index) => {
const match = resolved[index]
if (match && typeof match.fileHash === 'string') {
asset.fileHash = match.fileHash
asset.size = match.size ?? 0
if (typeof match.contentType === 'string') {
asset.contentType = match.contentType
}
asset.needsResolve = false
asset.assumedUploaded = true
logInfo(
`${pc.green('✔')} Locally missing ${UPLOAD_ASSET_LABEL[asset.kind].toLowerCase()}, reusing the previously uploaded version: ${displayAssetPath(asset.path)}`
)
} else {
unresolved.push(asset)
}
})
return unresolved
}
/**
* Builds the failure message shown when locally missing assets have no
* previously uploaded version to reuse.
*/
export function formatUnresolvedAssetMessage(
videoName: string,
unresolved: PreparedUploadAsset[]
): string {
const list = unresolved
.map((asset) => ` - ${UPLOAD_ASSET_LABEL[asset.kind]}: ${asset.path}`)
.join('\n')
return [
`Some asset files are missing locally and no previously uploaded version was found for "${videoName}":`,
list,
'Record once with these files present so they are uploaded, or commit them so they are available here.',
].join('\n')
}
async function uploadAssets(
assets: PreparedUploadAsset[],
apiUrl: string,
secret: string,
recordingId: string,
signal: AbortSignal,
throwIfAborted: () => void,
progressReporter?: { info: (message: string) => void }
): Promise<void> {
const logInfo = (message: string) => {
if (progressReporter) {
progressReporter.info(message)
} else {
logger.info(message)
}
}
for (const asset of assets) {
throwIfAborted()
const label = UPLOAD_ASSET_LABEL[asset.kind]
try {
if (!asset.alwaysUpload) {
const checkRes = await withUploadRetry(
() =>
fetch(`${apiUrl}/cli/upload/${recordingId}/asset/check`, {
method: 'POST',
headers: {
'Content-Type': 'application/json',
'X-ScreenCI-Secret': secret,
},
body: JSON.stringify({
fileHash: asset.fileHash,
contentType: asset.contentType,
size: asset.size,
path: asset.path,
...(typeof asset.name === 'string' ? { name: asset.name } : {}),
}),
signal,
}),
signal
)
if (!checkRes.ok) {
const text = await checkRes.text()
throw new UploadAssetError(
`Failed to check asset ${displayAssetPath(asset.path)}: ${checkRes.status} ${text}${hint401(checkRes.status, secret)}`
)
}
const checkBody = (await checkRes.json()) as { exists: boolean }
if (checkBody.exists) {
logInfo(
asset.assumedUploaded
? `${pc.green('✔')} Locally missing ${label.toLowerCase()}, already uploaded: ${displayAssetPath(asset.path)}`
: `${pc.green('✔')} ${label} already exists: ${displayAssetPath(asset.path)}`
)
continue
}
}
if (!asset.fileBuffer || !asset.contentType) {
// A locally missing asset matched a previous upload by path/name, but its
// bytes are not in this environment's storage. There is nothing to push.
if (asset.assumedUploaded) {
throw new UploadAssetError(
`${label} is missing locally and its previously uploaded bytes are no longer stored: ${displayAssetPath(asset.path)}. Record once with the file present so it is uploaded again, or commit the file.`
)
}
throw new UploadAssetError(
`Asset bytes not available for upload and backend does not have it yet: ${displayAssetPath(asset.path)}`
)
}
const fileBuffer = asset.fileBuffer
const contentType = asset.contentType
throwIfAborted()
const res = await withUploadRetry(
() =>
// Send the raw bytes, not base64-in-JSON: base64 builds a string that
// exceeds Node's max string length (~536MB) for large assets. Metadata
// travels in headers (URI-encoded so non-ASCII paths stay header-safe).
fetch(`${apiUrl}/cli/upload/${recordingId}/asset/stream`, {
method: 'PUT',
headers: {
'Content-Type': contentType,
'Content-Length': String(fileBuffer.byteLength),
'X-ScreenCI-Secret': secret,
'X-ScreenCI-File-Hash': asset.fileHash,
'X-ScreenCI-Asset-Size': String(asset.size),
'X-ScreenCI-Asset-Path': encodeURIComponent(asset.path),
},
body: fileBuffer as unknown as BodyInit,
signal,
}),
signal
)
if (!res.ok) {
const text = await res.text()
if (res.status === 409 && text.includes('already exists')) {
logInfo(
`${pc.green('✔')} ${label} already exists: ${displayAssetPath(asset.path)}`
)
} else {
throw new UploadAssetError(
`Failed to upload asset ${displayAssetPath(asset.path)}: ${res.status} ${text}${hint401(res.status, secret)}`
)
}
} else {
logInfo(
`${pc.green('✔')} ${label} uploaded: ${displayAssetPath(asset.path)}`
)
}
} catch (err) {
if (isUploadCancelledError(err)) {
throw err
}
if (isUploadAssetError(err)) {
throw err
}
throw new UploadAssetError(
`Network error uploading asset ${displayAssetPath(asset.path)}: ${err instanceof Error ? err.message : String(err)}`
)
}
}
}
export async function uploadRecordings(
screenciDir: string,
projectName: string,
apiUrl: string,
secret: string,
specificEntry?: string,
verbose = false
): Promise<{
projectId: string | null
recordId: string | null
hadFailures: boolean
failedVideoNames: string[]
failedVideoMessages: Array<{ videoName: string; message: string }>
studioNotices: StudioUploadNotice[]
plan: OrgPlan | null
}> {
const uploadAbort = createUploadAbortController('upload')
const recordId = randomUUID()
let entries: string[]
try {
entries = await readdir(screenciDir)
} catch {
logger.warn('No .screenci directory found, skipping upload')
return {
projectId: null,
recordId: null,
hadFailures: false,
failedVideoNames: [],
failedVideoMessages: [],
studioNotices: [],
plan: null,
}
}
if (specificEntry !== undefined) {
entries = entries.filter((e) => e === specificEntry)
}
let firstProjectId: string | null = null
const elevenLabsApiKey = process.env.ELEVENLABS_API_KEY?.trim() || undefined
try {
const candidates = (
await Promise.all(
entries.map(async (entry) => {
uploadAbort.throwIfAborted()
return await loadUploadCandidate(screenciDir, entry, verbose)
})
)
).filter((candidate): candidate is UploadCandidate => candidate !== null)
if (candidates.length === 0) {
return {
projectId: null,
recordId: null,
hadFailures: false,
failedVideoNames: [],
failedVideoMessages: [],
studioNotices: [],
plan: null,
}
}
const progressReporter = createUploadProgressReporter(
candidates.map((candidate) => candidate.videoName),
verbose
)
// Screenshots from this run render together on one machine; the backend
// waits for all of them to land before dispatching the batch, so it needs
// to know how many to expect.
const screenshotCount = candidates.filter(
(candidate) => candidate.data.output === 'screenshot'
).length
const results = await Promise.all(
candidates.map(
async (candidate, index) =>
await uploadRecordingCandidate(
candidate,
screenciDir,
projectName,
apiUrl,
secret,
elevenLabsApiKey,
verbose,
uploadAbort,
progressReporter,
index,
recordId,
screenshotCount
)
)
)
firstProjectId =
results.find((result) => result.projectId !== null)?.projectId ?? null
const resolvedPlan =
results.find((result) => result.plan !== undefined)?.plan ?? null
const hadFailures = results.some((result) => result.hadFailure)
const failedVideoNames = results
.filter((result) => result.hadFailure)
.map((result) => result.videoName)
const failedVideoMessages = results.flatMap((result) =>
result.hadFailure && typeof result.failureMessage === 'string'
? [{ videoName: result.videoName, message: result.failureMessage }]
: []
)
const studioNotices = results.flatMap((result) =>
!result.hadFailure && result.studio !== undefined
? [
{
videoName: result.videoName,
videoId: result.videoId,
studio: result.studio,
},
]
: []
)
return {
projectId: firstProjectId,
recordId,
hadFailures,
failedVideoNames,
failedVideoMessages,
studioNotices,
plan: resolvedPlan,
}
} finally {
uploadAbort.cleanup()
}
}
async function countCompletedRecordings(screenciDir: string): Promise<number> {
let entries: string[]
try {
entries = await readdir(screenciDir)
} catch {
return 0
}
return entries.filter((entry) =>
existsSync(resolve(screenciDir, entry, 'data.json'))
).length
}
function getScreenCISecretsUrl(): string {
return `${getDevFrontendUrl()}/secrets`
}
async function writeGitHubProjectOutput(projectUrl: string): Promise<void> {
const githubOutput = process.env.GITHUB_OUTPUT
if (!githubOutput) return
await appendFile(githubOutput, `screenci_project_url=${projectUrl}\n`)
}
async function loadScreenCIConfigAndEnv(configPath?: string): Promise<{
resolvedConfigPath: string
screenciConfig: ScreenCIConfig
}> {
const resolvedConfigPath = resolveScreenCIConfigPathOrExit(configPath)
let screenciConfig: ScreenCIConfig
try {
screenciConfig =
await loadRecordConfigWithoutPlaywrightCollision(resolvedConfigPath)
} catch (err) {
logger.error('Failed to load config:', err)
process.exit(1)
}
if (screenciConfig.envFile) {
loadEnvFile(
resolve(dirname(resolvedConfigPath), screenciConfig.envFile),
true
)
} else {
loadEnvFile(resolve(dirname(resolvedConfigPath), '.env'), false)
}
return { resolvedConfigPath, screenciConfig }
}
function loadEnvFile(envFilePath: string, warnOnFailure: boolean): void {
try {
const loadEnvFileCompat = (
process as NodeJS.Process & {
loadEnvFile?: (path: string | URL) => void
}
).loadEnvFile
if (typeof loadEnvFileCompat === 'function') {
loadEnvFileCompat(envFilePath)
return
}
loadEnvFileFallback(envFilePath)
} catch (err) {
if (warnOnFailure && !isMissingFileError(err)) {
logger.warn(`Failed to load env file ${envFilePath}:`, err)
}
}
}
function loadEnvFileFallback(envFilePath: string): void {
const envSource = readFileSync(envFilePath, 'utf8')
for (const [key, value] of parseEnvFile(envSource)) {
if (process.env[key] === undefined) {
process.env[key] = value
}
}
}
function parseEnvFile(envSource: string): Map<string, string> {
const parsed = new Map<string, string>()
const lines = envSource.replace(/^\uFEFF/, '').split(/\r?\n/)
for (const rawLine of lines) {
const line = rawLine.trim()
if (line === '' || line.startsWith('#')) continue
const normalizedLine = line.startsWith('export ')
? line.slice('export '.length).trimStart()
: line
const separatorIndex = normalizedLine.indexOf('=')
if (separatorIndex === -1) continue
const key = normalizedLine.slice(0, separatorIndex).trim()
if (!/^[A-Za-z_][A-Za-z0-9_]*$/.test(key)) continue
const rawValue = normalizedLine.slice(separatorIndex + 1).trim()
parsed.set(key, parseEnvValue(rawValue))
}
return parsed
}
function parseEnvValue(rawValue: string): string {
if (rawValue === '') return ''
if (
(rawValue.startsWith('"') && rawValue.endsWith('"')) ||
(rawValue.startsWith("'") && rawValue.endsWith("'"))
) {
const quote = rawValue[0]
const quotedValue = rawValue.slice(1, -1)
if (quote === '"') {
return quotedValue
.replace(/\\n/g, '\n')
.replace(/\\r/g, '\r')
.replace(/\\t/g, '\t')
.replace(/\\"/g, '"')
.replace(/\\\\/g, '\\')
}
return quotedValue
}
const inlineCommentIndex = rawValue.search(/\s#/)
if (inlineCommentIndex >= 0) {
return rawValue.slice(0, inlineCommentIndex).trimEnd()
}
return rawValue
}
function isMissingFileError(err: unknown): boolean {
return (
typeof err === 'object' &&
err !== null &&
'code' in err &&
err.code === 'ENOENT'
)
}
async function loadEnvFileFromConfigSource(
resolvedConfigPath: string,
warnOnFailure: boolean
): Promise<void> {
try {
const screenciConfig = await tryReadConfigFromSource(resolvedConfigPath)
loadEnvFile(
screenciConfig.envFile
? resolve(dirname(resolvedConfigPath), screenciConfig.envFile)
: resolve(dirname(resolvedConfigPath), '.env'),
warnOnFailure
)
} catch {
// Config import may require Playwright context or dynamic values. Continue with
// the existing process env; Playwright will still load the config normally.
}
}
async function resolveConfiguredEnvFilePath(
resolvedConfigPath: string
): Promise<string | undefined> {
try {
const screenciConfig = await tryReadConfigFromSource(resolvedConfigPath)
if (!screenciConfig.envFile) return undefined
return resolve(dirname(resolvedConfigPath), screenciConfig.envFile)
} catch {
return undefined
}
}
async function resolveProjectEnvFilePath(
resolvedConfigPath: string
): Promise<string> {
return (
(await resolveConfiguredEnvFilePath(resolvedConfigPath)) ??
resolve(dirname(resolvedConfigPath), '.env')
)
}
export function extractConfigStringLiteral(
configSource: string,
property: 'projectName' | 'envFile'
): string | undefined {
const singleQuoteMatch = configSource.match(
new RegExp(property + "\\s*:\\s*'([^'\\n]+)'")
)
if (singleQuoteMatch) return singleQuoteMatch[1]
const doubleQuoteMatch = configSource.match(
new RegExp(property + '\\s*:\\s*"([^"\\n]+)"')
)
if (doubleQuoteMatch) return doubleQuoteMatch[1]
const templateLiteralMatch = configSource.match(
new RegExp(property + '\\s*:\\s*`([^`\\n]+)`')
)
return templateLiteralMatch?.[1]
}
export function extractRecordUploadPolicyLiteral(
configSource: string
): RecordUploadPolicy | undefined {
const singleQuoteMatch = configSource.match(
/record\s*:\s*\{[\s\S]*?upload\s*:\s*'(passed-only|all-or-nothing)'/
)
if (singleQuoteMatch) {
return singleQuoteMatch[1] as RecordUploadPolicy
}
const doubleQuoteMatch = configSource.match(
/record\s*:\s*\{[\s\S]*?upload\s*:\s*"(passed-only|all-or-nothing)"/
)
if (doubleQuoteMatch) {
return doubleQuoteMatch[1] as RecordUploadPolicy
}
const templateLiteralMatch = configSource.match(
/record\s*:\s*\{[\s\S]*?upload\s*:\s*`(passed-only|all-or-nothing)`/
)
return templateLiteralMatch?.[1] as RecordUploadPolicy | undefined
}
export function extractMockRecordLiteral(
configSource: string
): boolean | undefined {
const match = configSource.match(
/test\s*:\s*\{[\s\S]*?mockRecord\s*:\s*(true|false)/
)
if (!match) return undefined
return match[1] === 'true'
}
function resolveRecordUploadPolicy(config: ScreenCIConfig): RecordUploadPolicy {
return config.record?.upload ?? DEFAULT_RECORD_UPLOAD_POLICY
}
async function tryReadConfigFromSource(resolvedConfigPath: string): Promise<
Pick<ScreenCIConfig, 'projectName'> & {
envFile?: string
record?: { upload?: RecordUploadPolicy }
test?: { mockRecord?: boolean }
}
> {
const configSource = await readFile(resolvedConfigPath, 'utf-8')
const projectName = extractConfigStringLiteral(configSource, 'projectName')
if (!projectName) {
throw new Error(
'Could not determine projectName from screenci.config.ts without importing it.'
)
}
const envFile = extractConfigStringLiteral(configSource, 'envFile')
const recordUpload = extractRecordUploadPolicyLiteral(configSource)
const mockRecord = extractMockRecordLiteral(configSource)
return {
projectName,
...(envFile !== undefined ? { envFile } : {}),
...(recordUpload !== undefined ? { record: { upload: recordUpload } } : {}),
...(mockRecord !== undefined ? { test: { mockRecord } } : {}),
}
}
export function getConfigModuleSpecifier(resolvedConfigPath: string): string {
if (
process.platform === 'win32' &&
/^[A-Za-z]:[\\/]/.test(resolvedConfigPath)
) {
return encodeURI(`file:///${resolvedConfigPath.replace(/\\/g, '/')}`)
}
return pathToFileURL(resolvedConfigPath).href
}
async function loadRecordConfigWithoutPlaywrightCollision(
resolvedConfigPath: string
): Promise<ScreenCIConfig> {
try {
const configModule = await import(
getConfigModuleSpecifier(resolvedConfigPath)
)
return configModule.default as ScreenCIConfig
} catch (err) {
const hasPlaywrightCollision =
err instanceof Error &&
err.message.includes('Requiring @playwright/test second time')
if (hasPlaywrightCollision) {
logger.warn(
'Playwright was loaded from multiple module paths. Falling back to static config parsing for upload metadata.'
)
}
try {
return (await tryReadConfigFromSource(
resolvedConfigPath
)) as ScreenCIConfig
} catch {
// Preserve the original import error when static parsing cannot recover.
}
throw err
}
}
async function requireScreenCISecret(
configPath?: string,
opts: { interactive?: boolean; pollAuth?: boolean; noPollAuth?: boolean } = {}
): Promise<{
resolvedConfigPath: string
screenciConfig: ScreenCIConfig
secret: string
apiUrl: string
}> {
const { resolvedConfigPath, screenciConfig } =
await loadScreenCIConfigAndEnv(configPath)
let secret: string | undefined
try {
secret =
process.env.SCREENCI_SECRET ??
(await ensureScreenciSecret(resolvedConfigPath, opts))
} catch (err) {
// A waited-out sign-in means recording did not happen: print the (already
// formatted) message and exit non-zero so the failure is loud, instead of
// looking like a successful run.
if (err instanceof SignInTimeoutError) {
logger.error(err.message)
process.exit(1)
}
throw err
}
if (!secret) {
// In a non-interactive session ensureScreenciSecret already printed the
// sign-in link and the next step, so we exit without repeating guidance.
// A pending sign-in is an expected handoff (surface the link, then rerun),
// not a failure, so exit cleanly (0) to avoid flagging the run as red.
if (opts.interactive === false) {
process.exit(0)
}
const envFilePath = await resolveProjectEnvFilePath(resolvedConfigPath)
logger.error(
`No SCREENCI_SECRET configured. Rerun ${pc.cyan(getSuggestedScreenciCommand('record'))} or add SCREENCI_SECRET manually to ${envFilePath} by following the guide at ${pc.cyan(SCREENCI_RECORD_DOCS_URL)}.`
)
process.exit(1)
}
return {
resolvedConfigPath,
screenciConfig,
secret,
apiUrl: getDevBackendUrl(),
}
}
async function updateVideoVisibility(
videoId: string,
isPublic: boolean,
configPath?: string
): Promise<void> {
const { secret, apiUrl } = await requireScreenCISecret(configPath, {
interactive: detectInteractiveSession(),
})
const method = isPublic ? 'PUT' : 'DELETE'
const res = await fetch(`${apiUrl}/cli/public-video/${videoId}`, {
method,
headers: {
'X-ScreenCI-Secret': secret,
},
})
if (!res.ok) {
const text = await res.text()
throw new Error(
`Failed to ${isPublic ? 'make public' : 'make private'}: ${res.status} ${text}${hint401(res.status, secret)}`
)
}
logger.info(`${isPublic ? 'Made public' : 'Made private'}: ${videoId}`)
}
// Extract a `--grep <value>` / `--grep=<value>` from the pass-through args so a
// remote trigger can forward it as a filter (records only matching videos).
export function extractGrep(args: string[]): string | undefined {
for (let i = 0; i < args.length; i++) {
const arg = args[i]
if (arg === undefined) continue
if (arg === '--grep' || arg === '-g') {
return args[i + 1]
}
if (arg.startsWith('--grep=')) {
return arg.slice('--grep='.length)
}
}
return undefined
}
// `screenci record --remote` triggers the project's GitHub Actions recording
// workflow instead of recording locally. The project is resolved from the
// existing SCREENCI_SECRET + config `projectName`, exactly like the other
// authenticated commands; the backend dispatches the workflow using the GitHub
// token stored for the project. An optional `--grep` records only matching
// videos/screenshots.
async function triggerRemoteRun(
configPath?: string,
grep?: string,
languages?: string
): Promise<void> {
const { screenciConfig, secret, apiUrl } = await requireScreenCISecret(
configPath,
{ interactive: detectInteractiveSession() }
)
const res = await fetch(`${apiUrl}/cli/trigger-run`, {
method: 'POST',
headers: {
'X-ScreenCI-Secret': secret,
'Content-Type': 'application/json',
},
body: JSON.stringify({
projectName: screenciConfig.projectName,
...(grep ? { grep } : {}),
...(languages ? { languages } : {}),
}),
})
if (!res.ok) {
const text = await res.text()
throw new Error(
`Failed to trigger remote run: ${res.status} ${text}${hint401(res.status, secret)}`
)
}
const filters = [
...(grep ? [`filter: ${grep}`] : []),
...(languages ? [`languages: ${languages}`] : []),
]
logger.info(
filters.length > 0
? `Triggered the remote recording workflow for "${screenciConfig.projectName}" (${filters.join(', ')}).`
: `Triggered the remote recording workflow for "${screenciConfig.projectName}".`
)
}
function getLastRecordFilePath(screenciDir: string): string {
return resolve(screenciDir, SCREENCI_LAST_RECORD_FILE)
}
/**
* Persists the recordId of the just-completed `screenci record` upload so a
* later `screenci info` can report exactly that run. Best-effort: a
* failure to write must not fail the record command.
*/
async function saveLastRecordId(
screenciDir: string,
recordId: string
): Promise<void> {
try {
mkdirSync(screenciDir, { recursive: true })
await writeFile(
getLastRecordFilePath(screenciDir),
`${JSON.stringify({ recordId, savedAt: new Date().toISOString() }, null, 2)}\n`
)
} catch (err) {
logger.warn(
`Failed to record run id for info: ${err instanceof Error ? err.message : String(err)}`
)
}
}
async function readLastRecordId(screenciDir: string): Promise<string | null> {
try {
const raw = await readFile(getLastRecordFilePath(screenciDir), 'utf-8')
const parsed = JSON.parse(raw) as { recordId?: unknown }
return typeof parsed.recordId === 'string' ? parsed.recordId : null
} catch (err) {
if (!isMissingFileError(err)) {
logger.warn(
`Ignoring invalid stored record at ${getLastRecordFilePath(screenciDir)}.`
)
}
return null
}
}
// `screenci info` prints every project video and its public URLs as JSON. When
// this machine has recorded a run (a recordId is stored in
// .screenci/last-record.json), the backend also attaches, to the videos from
// that run, a per-language `latestRecord` with render status and record-pinned
// URLs. Without a local run, only the project-wide listing with `static` URLs is
// returned. The server does the merge; the CLI just passes the recordId.
async function printInfo(configPath?: string): Promise<void> {
const { resolvedConfigPath, screenciConfig, secret, apiUrl } =
await requireScreenCISecret(configPath, {
interactive: detectInteractiveSession(),
})
const screenciDir = resolve(dirname(resolvedConfigPath), '.screenci')
const recordId = await readLastRecordId(screenciDir)
const url = new URL(`${apiUrl}/cli/info`)
url.searchParams.set('projectName', screenciConfig.projectName)
if (recordId) url.searchParams.set('record', recordId)
const res = await fetch(url.toString(), {
headers: { 'X-ScreenCI-Secret': secret },
})
if (!res.ok) {
const text = await res.text()
throw new Error(
`Failed to fetch info: ${res.status} ${text}${hint401(res.status, secret)}`
)
}
const data = await res.json()
process.stdout.write(`${JSON.stringify(data, null, 2)}\n`)
}
async function persistScreenCISecret(
envFilePath: string,
secret: string
): Promise<void> {
const nextLine = `SCREENCI_SECRET=${secret}`
try {
const existing = await readFile(envFilePath, 'utf-8')
const lines = existing === '' ? [] : existing.split(/\r?\n/)
const firstSecretIndex = lines.findIndex((line) =>
line.startsWith('SCREENCI_SECRET=')
)
const linesWithoutSecret = lines.filter(
(line) => !line.startsWith('SCREENCI_SECRET=')
)
const finalLines =
firstSecretIndex >= 0
? [
...linesWithoutSecret.slice(0, firstSecretIndex),
nextLine,
...linesWithoutSecret.slice(firstSecretIndex),
]
: [...linesWithoutSecret, nextLine]
let nextContent = finalLines.join('\n')
if (!nextContent.endsWith('\n')) nextContent += '\n'
await writeFile(envFilePath, nextContent)
return
} catch (err) {
if (!isMissingFileError(err)) throw err
}
await writeFile(envFilePath, `${nextLine}\n`)
}
async function pollLinkSessionOnce(
spec: PersistedLinkSessionSpec
): Promise<{ status: LinkSessionStatus; secret?: string }> {
const response = await fetch(spec.pollUrl)
const body = (await response.json()) as {
status?: LinkSessionStatus
secret?: string
}
const status = body.status ?? 'invalid'
if (status === 'completed' && body.secret) {
return { status, secret: body.secret }
}
if (status === 'pending' && new Date().toISOString() >= spec.expiresAt) {
return { status: 'expired' }
}
return { status }
}
function getAuthPollTimeoutMs(defaultMs: number): number {
const raw = process.env.SCREENCI_POLL_AUTH_TIMEOUT_MS
if (raw === undefined) return defaultMs
const parsed = Number(raw)
return Number.isFinite(parsed) && parsed >= 0 ? parsed : defaultMs
}
async function pollLinkSession(
spec: PersistedLinkSessionSpec,
pollIntervalMs: number = SCREENCI_LINK_SESSION_POLL_INTERVAL_MS,
deadlineEpochMs?: number
): Promise<{ status: LinkSessionStatus | 'timed-out'; secret?: string }> {
for (;;) {
const result = await pollLinkSessionOnce(spec)
if (result.status === 'completed' && result.secret) {
return result
}
if (
result.status === 'expired' ||
result.status === 'consumed' ||
result.status === 'invalid'
) {
return result
}
// Stop before sleeping again once the optional deadline has passed so
// `--poll-auth` cannot block forever waiting for a human to sign in.
if (deadlineEpochMs !== undefined && Date.now() >= deadlineEpochMs) {
return { status: 'timed-out' }
}
await new Promise((resolveDelay) =>
setTimeout(resolveDelay, pollIntervalMs)
)
}
}
export async function ensureScreenciSecret(
resolvedConfigPath?: string,
opts: {
interactive?: boolean
pollAuth?: boolean
noPollAuth?: boolean
} = {}
): Promise<string | undefined> {
const interactive = opts.interactive ?? true
const pollAuth = opts.pollAuth ?? false
const noPollAuth = opts.noPollAuth ?? false
const existingSecret = process.env.SCREENCI_SECRET
if (existingSecret) return existingSecret
try {
const environment = getScreenCIEnvironment()
const apiUrl = getCliLinkSessionApiUrl()
const appUrl = getDevFrontendUrl()
const envFilePath = resolvedConfigPath
? await resolveProjectEnvFilePath(resolvedConfigPath)
: resolve(process.cwd(), '.env')
const projectDir = resolvedConfigPath
? dirname(resolvedConfigPath)
: process.cwd()
const specPath = getLinkSessionFilePath(projectDir)
const linkSessionContext = {
environment,
envFilePath,
...(resolvedConfigPath ? { resolvedConfigPath } : {}),
}
const ensureSpec = async (): Promise<PersistedLinkSessionSpec> => {
const storedSpec = await readPersistedLinkSessionSpec(specPath)
const spec =
storedSpec &&
isStoredLinkSessionReusable(storedSpec, linkSessionContext)
? storedSpec
: await createLinkSessionSpec({
apiUrl,
appUrl,
...linkSessionContext,
})
if (spec !== storedSpec) {
await writePersistedLinkSessionSpec(specPath, spec)
}
return spec
}
const saveCompletedSecret = async (secret: string): Promise<string> => {
process.env.SCREENCI_SECRET = secret
await persistScreenCISecret(envFilePath, secret)
deletePersistedLinkSessionSpec(specPath)
logger.info(`Successfully saved SCREENCI_SECRET to ${envFilePath}`)
return secret
}
// Decide whether this run waits for a browser sign-in or just prints the
// link and hands off. A human at a terminal always waits. A plain
// non-interactive run (a coding agent: no terminal, no CI) also waits by
// default, so `npx screenci record` records without needing a flag. CI and
// an explicit SCREENCI_NONINTERACTIVE never wait (CI must not hang and is
// expected to provide SCREENCI_SECRET), and `--no-poll-auth` opts back into
// that print-and-exit handoff. `--poll-auth` forces waiting and is kept for
// backwards compatibility (it is the default for agents now).
const isCI = process.env.CI === 'true'
const nonInteractiveForced = process.env.SCREENCI_NONINTERACTIVE === '1'
const waitForSignIn = noPollAuth
? false
: interactive || pollAuth || (!isCI && !nonInteractiveForced)
const pollIntervalMs = interactive
? SCREENCI_LINK_SESSION_POLL_INTERVAL_MS
: SCREENCI_LINK_SESSION_POLL_FLAG_INTERVAL_MS
const timeoutMs = getAuthPollTimeoutMs(
interactive
? SCREENCI_LINK_SESSION_INTERACTIVE_TIMEOUT_MS
: SCREENCI_LINK_SESSION_POLL_FLAG_TIMEOUT_MS
)
// Check the current/stored session once. If sign-in already completed (for
// example between runs) we pick up the secret immediately. Recreate a stale
// session and check once more before deciding to wait or hand off.
let spec = await ensureSpec()
let result = await pollLinkSessionOnce(spec)
if (
result.status === 'expired' ||
result.status === 'consumed' ||
result.status === 'invalid'
) {
deletePersistedLinkSessionSpec(specPath)
spec = await ensureSpec()
result = await pollLinkSessionOnce(spec)
}
if (result.status === 'completed' && result.secret) {
return await saveCompletedSecret(result.secret)
}
if (!waitForSignIn) {
// CI / SCREENCI_NONINTERACTIVE / --no-poll-auth: print the link and hand
// off without waiting. The caller exits cleanly (0); a pending sign-in
// here is an expected handoff, not a failure.
logger.info(
`Sign-in required to record. Open this link to sign in:\n${pc.cyan(spec.appUrl)}\n` +
`This session won't wait for sign-in. After signing in, rerun ${pc.cyan(getSuggestedScreenciCommand('record'))} to record. In CI, set SCREENCI_SECRET (from ${getScreenCISecretsUrl()}) to record without an interactive sign-in.`
)
return undefined
}
// Wait for sign-in: print the link, then poll until the human finishes and
// continue recording automatically once the secret lands. We re-print the
// link on every (re)created session so the latest valid link is always
// visible. We do not wait forever: after the timeout we throw so the caller
// exits non-zero (a timed-out sign-in means nothing was recorded). The link
// stays valid for a rerun.
const timeoutMinutes = Math.max(1, Math.round(timeoutMs / 60_000))
const intervalSeconds = Math.max(1, Math.round(pollIntervalMs / 1_000))
const deadlineEpochMs = Date.now() + timeoutMs
for (;;) {
logger.info(
`Sign-in required to record. Open this link to sign in:\n${pc.cyan(spec.appUrl)}\n` +
`Waiting for sign-in (checking every ${intervalSeconds} seconds, up to ${timeoutMinutes} minutes). Recording continues automatically once you finish.`
)
const polled = await pollLinkSession(
spec,
pollIntervalMs,
deadlineEpochMs
)
if (polled.status === 'completed' && polled.secret) {
return await saveCompletedSecret(polled.secret)
}
if (polled.status === 'timed-out' || Date.now() >= deadlineEpochMs) {
throw new SignInTimeoutError(
`Timed out after ${timeoutMinutes} minutes waiting for sign-in. The link is still valid:\n${pc.cyan(spec.appUrl)}\n` +
`After signing in, rerun ${pc.cyan(getSuggestedScreenciCommand('record'))} to record.`
)
}
// Session went stale (expired/consumed/invalid) before sign-in: recreate
// and keep waiting until the deadline.
deletePersistedLinkSessionSpec(specPath)
spec = await ensureSpec()
}
} catch (err) {
// A timed-out sign-in is a deliberate, already-formatted failure: let it
// propagate so the caller can exit non-zero rather than swallowing it as a
// generic auth error.
if (err instanceof SignInTimeoutError) throw err
const msg = err instanceof Error ? err.message : String(err)
logger.warn(`Authentication failed: ${msg}`)
logger.info(
`You can add SCREENCI_SECRET manually to ${resolvedConfigPath ? await resolveProjectEnvFilePath(resolvedConfigPath) : '.env'} later. Get it from ${getScreenCISecretsUrl()}.`
)
logScreenCISecretGuide()
return undefined
}
}
/**
* `screenci login`: get a sign-in link in front of a human as early as
* possible. It reuses the link `init` (or a prior run) persisted, or creates a
* fresh one, tries to open it in the browser best-effort, and always prints it
* as a fallback. It does NOT wait for sign-in to finish: the agent can run this,
* keep authoring and `test`ing, then `record` (which blocks for sign-in) once
* the script is green. If sign-in already completed, it saves the secret and
* exits.
*/
export async function runLogin(configPath?: string): Promise<void> {
const { resolvedConfigPath } = await loadScreenCIConfigAndEnv(configPath)
if (process.env.SCREENCI_SECRET) {
logger.info(
`Already signed in (SCREENCI_SECRET is set). Run ${pc.cyan(getSuggestedScreenciCommand('record'))} to record.`
)
return
}
const environment = getScreenCIEnvironment()
const apiUrl = getCliLinkSessionApiUrl()
const appUrl = getDevFrontendUrl()
const envFilePath = await resolveProjectEnvFilePath(resolvedConfigPath)
const projectDir = dirname(resolvedConfigPath)
const specPath = getLinkSessionFilePath(projectDir)
const linkSessionContext = { environment, envFilePath, resolvedConfigPath }
const createSpec = async (): Promise<PersistedLinkSessionSpec> => {
const created = await createLinkSessionSpec({
apiUrl,
appUrl,
...linkSessionContext,
})
await writePersistedLinkSessionSpec(specPath, created)
return created
}
try {
// Reuse a still-valid persisted session so login and a later record share
// the same link; otherwise create and persist a fresh one.
const stored = await readPersistedLinkSessionSpec(specPath)
let spec =
stored && isStoredLinkSessionReusable(stored, linkSessionContext)
? stored
: await createSpec()
// If sign-in already completed (e.g. between runs) save the secret and stop:
// there is nothing left to open. Recreate a stale session before opening so
// the link we hand off is valid.
const result = await pollLinkSessionOnce(spec)
if (result.status === 'completed' && result.secret) {
process.env.SCREENCI_SECRET = result.secret
await persistScreenCISecret(envFilePath, result.secret)
deletePersistedLinkSessionSpec(specPath)
logger.info(`Already signed in. Saved SCREENCI_SECRET to ${envFilePath}.`)
return
}
if (
result.status === 'expired' ||
result.status === 'consumed' ||
result.status === 'invalid'
) {
deletePersistedLinkSessionSpec(specPath)
spec = await createSpec()
}
const opened = await openUrlInBrowser(spec.appUrl)
if (opened.opened) {
logger.info(
`Opening the sign-in link in your browser:\n${pc.cyan(spec.appUrl)}`
)
} else {
logger.info(
`Could not open a browser automatically (${opened.reason}). Open this link to sign in:\n${pc.cyan(spec.appUrl)}`
)
}
logger.info(
`The link stays valid for 24 hours. You can keep authoring and run ${pc.cyan(getSuggestedScreenciCommand('test'))} while you sign in. Then run ${pc.cyan(getSuggestedScreenciCommand('record'))}, which waits for sign-in and records once you finish.`
)
} catch (err) {
// Login is a best-effort convenience: never fail the run on a network or
// browser hiccup. record can still create and wait on the link later.
const msg = err instanceof Error ? err.message : String(err)
logger.warn(`Could not start sign-in: ${msg}`)
logger.info(
`You can sign in later: ${pc.cyan(getSuggestedScreenciCommand('record'))} prints a link and waits for sign-in. In CI, set SCREENCI_SECRET (from ${getScreenCISecretsUrl()}).`
)
}
}
// Uploads the recordings already written under `.screenci` for the resolved
// config. Shared by `record` (after a Playwright run) and `retry` (which
// re-sends the existing recordings without re-running Playwright). A non-null
// `playwrightFailure` means the preceding record run had failures, which tunes
// the messaging and the upload policy; `retry` always passes null.
async function uploadRecordedVideosForConfig(
configPath: string | undefined,
playwrightFailure: Error | null,
verbose: boolean
): Promise<void> {
// After recording, upload results to API if configured. `run` already
// resolved the config (or exited), so this best-effort lookup only acts
// when a flat config is present in/under the current directory.
const resolution = findScreenCIConfig(configPath)
if (resolution.kind !== 'found') return
const resolvedConfigPath = resolution.path
try {
const screenciConfig =
await loadRecordConfigWithoutPlaywrightCollision(resolvedConfigPath)
loadEnvFile(
screenciConfig.envFile
? resolve(dirname(resolvedConfigPath), screenciConfig.envFile)
: resolve(dirname(resolvedConfigPath), '.env'),
true
)
const apiUrl = getDevBackendUrl()
const appUrl = getDevFrontendUrl()
const secret = process.env.SCREENCI_SECRET
const uploadPolicy = resolveRecordUploadPolicy(screenciConfig)
const configDir = dirname(resolvedConfigPath)
const screenciDir = resolve(configDir, '.screenci')
const completedRecordingCount = await countCompletedRecordings(screenciDir)
if (playwrightFailure !== null && completedRecordingCount === 0) {
logger.info('All recordings failed.')
} else if (!secret) {
logger.info(
`No SCREENCI_SECRET configured for uploads. Rerun ${getSuggestedScreenciCommand('record')} or add it to the project env file.`
)
} else if (
playwrightFailure !== null &&
uploadPolicy === 'all-or-nothing'
) {
logger.info(
'Some recordings failed, skipping upload because record.upload is "all-or-nothing".'
)
} else {
if (playwrightFailure !== null && uploadPolicy === 'passed-only') {
logger.warn('Some recordings failed, uploading successful videos only.')
}
let uploadResult: {
projectId: string | null
recordId: string | null
hadFailures: boolean
failedVideoNames: string[]
failedVideoMessages: Array<{ videoName: string; message: string }>
studioNotices: StudioUploadNotice[]
plan: OrgPlan | null
} = {
projectId: null,
recordId: null,
hadFailures: false,
failedVideoNames: [],
failedVideoMessages: [],
studioNotices: [],
plan: null,
}
try {
uploadResult = await uploadRecordings(
screenciDir,
screenciConfig.projectName,
apiUrl,
secret,
undefined,
verbose
)
} catch (err) {
if (isUploadCancelledError(err)) {
process.exit(130)
}
throw err
}
const {
projectId,
recordId,
hadFailures,
failedVideoNames,
failedVideoMessages,
studioNotices,
plan,
} = uploadResult
// Remember this run so `screenci info` can report exactly it.
if (recordId !== null) {
await saveLastRecordId(screenciDir, recordId)
}
// Emit upload-failure warnings (stderr) before the results block.
// logger.info writes to stdout, logger.warn to stderr; in non-TTY CI
// logs stdout is block-buffered while stderr flushes immediately, so
// warnings printed after the "Results available at:" line would split
// it from its URL. Reporting failures first keeps the URL directly
// under its message.
if (hadFailures) {
for (const failedVideo of failedVideoMessages) {
logger.warn(
formatFailedVideoMessage(failedVideo.videoName, failedVideo.message)
)
}
logger.warn(
`Not all recordings succeeded to upload. Failed videos: ${failedVideoNames.join(', ') || 'unknown'}. Some videos may be missing from the project.`
)
}
if (recordId !== null && projectId !== null) {
const recordUrl = `${appUrl}/record/${recordId}`
await writeGitHubProjectOutput(recordUrl)
logger.info('')
logger.info(
playwrightFailure !== null
? 'Recording partially succeeded, rendering in progress. Results available at:'
: 'Recording finished, rendering in progress. Results available at:'
)
logger.info(pc.cyan(recordUrl))
} else if (projectId !== null) {
const projectUrl = `${appUrl}/project/${projectId}`
await writeGitHubProjectOutput(projectUrl)
logger.info('')
logger.info(
playwrightFailure !== null
? 'Recording partially succeeded, rendering in progress. Results available at:'
: 'Recording finished, rendering in progress. Results available at:'
)
logger.info(pc.cyan(projectUrl))
}
if (projectId !== null && plan !== 'business') {
logger.info('')
logger.info(
'Upgrade for more renders, more active videos, and expressive narration:'
)
logger.info(pc.cyan(`${appUrl}/select-plan`))
}
for (const notice of studioNotices) {
if ('held' in notice.studio) {
logger.info('')
logger.info(
`Rendering for "${notice.videoName}" is on hold. Configure it in Studio:`
)
if (projectId !== null && notice.videoId !== null) {
logger.info(
pc.cyan(formatStudioUrl(appUrl, projectId, notice.videoId))
)
}
} else if (notice.studio.applied) {
logger.info('')
logger.info(`Studio configuration applied for "${notice.videoName}".`)
}
}
if (hadFailures && playwrightFailure === null) {
throw new PartialUploadError()
}
}
} catch (err) {
if (isPartialUploadError(err)) {
throw err
}
logger.warn('Failed to load config for upload:', err)
}
}
export async function main() {
if (process.argv.length <= 2) {
logger.error('Error: No command provided')
logger.error(
'Available commands: record, test, login, info, make-public, make-private, init'
)
process.exit(1)
}
const program = new Command()
const defaultPackageManager = determinePackageManager()
program.name('screenci')
program.exitOverride()
// record command — playwright args pass through as-is
program
.command('record [playwrightArgs...]')
.description('Record videos using Playwright')
.option('-v, --verbose', 'verbose output')
.option(
'--poll-auth',
'wait for sign-in before recording (now the default off-CI; kept for backwards compatibility)'
)
.option(
'--no-poll-auth',
'do not wait for sign-in: print the link and exit cleanly if not signed in (the default under CI)'
)
.option(
'--remote',
'trigger the GitHub Actions recording workflow for this project remotely instead of recording locally'
)
.option(
'--languages <langs>',
'record/render only these languages (comma-separated, e.g. fi,en)'
)
.allowUnknownOption(true)
.action(async () => {
const parsed = parseRecordCliArgs(getSubcommandArgv('record'))
// `--remote` is a pure dispatch: it fires the project's GitHub Actions
// recording workflow and exits, so there is no local Playwright run. A
// pass-through `--grep` becomes the remote recording filter, and
// `--languages` limits which language versions are recorded.
if (parsed.remote) {
await triggerRemoteRun(
parsed.configPath,
extractGrep(parsed.otherArgs),
parsed.languages
)
return
}
let playwrightFailure: Error | null = null
// UPLOAD_EXISTING re-sends the recordings already on disk under `.screenci`
// without re-running Playwright (resend the last local run when only the
// upload failed). We skip the recording run and fall straight through to
// the upload below, treating the on-disk recordings as the complete set.
const uploadExisting = isUploadExistingEnabled()
if (!uploadExisting) {
try {
await run(
'record',
parsed.otherArgs,
parsed.configPath,
parsed.verbose,
false,
parsed.pollAuth,
parsed.languages,
parsed.noPollAuth
)
} catch (error) {
if (!(error instanceof Error)) throw error
if (error.message.startsWith('Playwright exited with code ')) {
playwrightFailure = new RecordFailureHintError(error)
} else {
throw new RecordFailureHintError(error)
}
}
} else {
logger.info(
'UPLOAD_EXISTING set: skipping Playwright recording and re-uploading existing .screenci recordings.'
)
}
if (process.env.SCREENCI_RECORDING === 'true') return
await uploadRecordedVideosForConfig(
parsed.configPath,
playwrightFailure,
parsed.verbose
)
if (playwrightFailure !== null) {
throw playwrightFailure
}
})
program
.command('test [playwrightArgs...]')
.description('Run Playwright test with screenci.config.ts')
.option(
'--mock-record',
'keep recording-style cursor animation and sleeps during screenci test'
)
.option('-v, --verbose', 'verbose output')
.allowUnknownOption(true)
.action(async () => {
const parsed = parseConfigCliArgs(getSubcommandArgv('test'))
let configMockRecord = false
// Best-effort env preload before handing off to `run`, which performs the
// authoritative resolution (and emits the `cd screenci` guidance on miss).
const resolution = findScreenCIConfig(parsed.configPath)
if (resolution.kind === 'found') {
const resolvedConfigPath = resolution.path
try {
const screenciConfig =
await loadRecordConfigWithoutPlaywrightCollision(resolvedConfigPath)
configMockRecord = screenciConfig.test?.mockRecord ?? false
if (screenciConfig.envFile) {
const envFilePath = resolve(
dirname(resolvedConfigPath),
screenciConfig.envFile
)
loadEnvFile(envFilePath, true)
}
} catch (err) {
logger.warn('Failed to load config for test env:', err)
}
}
await run(
'test',
parsed.otherArgs,
parsed.configPath,
parsed.verbose,
parsed.mockRecord || configMockRecord
)
if (process.env.SCREENCI_RECORDING === 'true') return
const recordCommand = getSuggestedScreenciCommand('record')
logger.info(
`Tests passed. Run ${pc.cyan(recordCommand)} to render the videos.`
)
})
program
.command('login')
.description(
'Open the sign-in link in your browser (best-effort) so recording can finish later'
)
.option('-c, --config <path>', 'path to screenci.config.ts')
.action(async (options: Record<string, unknown>) => {
await runLogin(options['config'] as string | undefined)
})
program
.command('info')
.description(
"Print the latest record run's video URLs and render status as JSON"
)
.option('-c, --config <path>', 'path to screenci.config.ts')
.action(async (options: Record<string, unknown>) => {
await printInfo(options['config'] as string | undefined)
})
program
.command('make-public <id>')
.description(
'Enable public URLs for a video; get the id from screenci info'
)
.option('-c, --config <path>', 'path to screenci.config.ts')
.action(async (id: string, options: Record<string, unknown>) => {
await updateVideoVisibility(
id,
true,
options['config'] as string | undefined
)
})
program
.command('make-private <id>')
.description(
'Disable public URLs for a video; get the id from screenci info'
)
.option('-c, --config <path>', 'path to screenci.config.ts')
.action(async (id: string, options: Record<string, unknown>) => {
await updateVideoVisibility(
id,
false,
options['config'] as string | undefined
)
})
// init command
program
.command('init [name]')
.description('Initialize a new screenci project')
.option(
'--agent <name>',
'target agent for skills install, e.g. opencode. Supported agents: https://github.com/vercel-labs/skills#supported-agents'
)
.option(
'--package-manager <manager>',
`package manager to use: npm, pnpm, or yarn 2+ (default: ${defaultPackageManager})`
)
.option('-y, --yes', 'accept init defaults')
.option('-v, --verbose', 'verbose output')
.action(
async (name: string | undefined, options: Record<string, unknown>) => {
const agent = options['agent'] as string | undefined
await runInit(name, {
verbose: (options['verbose'] as boolean | undefined) ?? false,
yes: (options['yes'] as boolean | undefined) ?? false,
packageManager: parsePackageManager(
options['packageManager'] as string | undefined,
process.env['SCREENCI_INIT_CWD'] ?? process.cwd()
),
...(agent !== undefined ? { agent } : {}),
})
}
)
try {
await program.parseAsync(process.argv)
} catch (err) {
if (err instanceof CommanderError) {
if (err.code === 'commander.unknownCommand') {
const unknownCmd = process.argv[2] ?? ''
logger.error(`Unknown command: ${unknownCmd}`)
process.exit(1)
}
if (err.code === 'commander.optionMissingArgument') {
if (
err.message.includes('--config') ||
err.message.includes('-c, --config') ||
err.message.includes("'-c'")
) {
logger.error('Error: --config requires a path argument')
process.exit(1)
}
logger.error(`Error: ${err.message}`)
process.exit(1)
}
if (
err.code === 'commander.help' ||
err.code === 'commander.helpDisplayed'
) {
return
}
logger.error(`Error: ${err.message}`)
process.exit(1)
return
}
throw err
}
}
function getSubcommandArgv(command: string): string[] {
const argv = process.argv.slice(2)
const commandIndex = argv.indexOf(command)
return commandIndex === -1 ? [] : argv.slice(commandIndex + 1)
}
export function parseRecordCliArgs(args: string[]): {
configPath: string | undefined
verbose: boolean
pollAuth: boolean
noPollAuth: boolean
remote: boolean
languages: string | undefined
otherArgs: string[]
} {
let configPath: string | undefined
let verbose = false
let pollAuth = false
let noPollAuth = false
let remote = false
let languages: string | undefined
const otherArgs: string[] = []
for (let i = 0; i < args.length; i++) {
const arg = args[i]
if (arg === undefined) continue
if (arg === '--config' || arg === '-c') {
const nextArg = args[i + 1]
if (nextArg === undefined) {
logger.error('Error: --config requires a path argument')
process.exit(1)
}
configPath = nextArg
i++
} else if (arg === '--languages' || arg === '--language') {
// screenci-only flag: parsed out so it is not forwarded to Playwright.
const nextArg = args[i + 1]
if (nextArg === undefined) {
logger.error('Error: --languages requires a comma-separated value')
process.exit(1)
}
languages = nextArg
i++
} else if (
arg.startsWith('--languages=') ||
arg.startsWith('--language=')
) {
languages = arg.slice(arg.indexOf('=') + 1)
} else if (arg === '--verbose' || arg === '-v') {
verbose = true
} else if (arg === '--poll-auth') {
pollAuth = true
} else if (arg === '--no-poll-auth') {
noPollAuth = true
} else if (arg === '--remote') {
remote = true
} else {
otherArgs.push(arg)
}
}
return {
configPath,
verbose,
pollAuth,
noPollAuth,
remote,
languages,
otherArgs,
}
}
function parseConfigCliArgs(args: string[]): {
configPath: string | undefined
verbose: boolean
mockRecord: boolean
otherArgs: string[]
} {
let configPath: string | undefined
let verbose = false
let mockRecord = false
const otherArgs: string[] = []
for (let i = 0; i < args.length; i++) {
const arg = args[i]
if (arg === undefined) continue
if (arg === '--config' || arg === '-c') {
const nextArg = args[i + 1]
if (nextArg === undefined) {
logger.error('Error: --config requires a path argument')
process.exit(1)
}
configPath = nextArg
i++
} else if (arg === '--verbose' || arg === '-v') {
verbose = true
} else if (arg === '--mock-record') {
mockRecord = true
} else {
otherArgs.push(arg)
}
}
return { configPath, verbose, mockRecord, otherArgs }
}
function validateArgs(args: string[]): void {
const disallowedFlags = ['--retries']
for (const arg of args) {
if (arg === undefined) continue
// Check if it's a disallowed flag
if (disallowedFlags.includes(arg)) {
throw new Error(
`Flag "${arg}" is not supported by screenci. ` +
'screenci forces retries to 0 for proper video recording.'
)
}
// Check if it's a --retries=N format
if (arg.startsWith('--retries=')) {
throw new Error(
`Flag "${arg}" is not supported by screenci. ` +
'screenci forces retries to 0 for proper video recording.'
)
}
}
}
/**
* Fetch the project's current Studio text-field overrides so they can be
* injected as SCREENCI_VALUES_OVERRIDES before a recording. On-screen `values`
* fields render during the recording (they cannot be patched at render time),
* so any Studio edits must be present here; code-declared seeds are the
* fallback. Best-effort: any failure returns an empty env so the SDK uses the
* code seeds, and a recording is never blocked by this fetch.
*/
async function fetchTextOverridesEnv(
configPath: string,
languages: string | undefined,
verbose: boolean
): Promise<Record<string, string>> {
try {
const { screenciConfig, secret, apiUrl } = await requireScreenCISecret(
configPath,
{ interactive: false }
)
const params = new URLSearchParams({
projectName: screenciConfig.projectName,
})
if (languages) params.set('languages', languages)
const res = await fetch(
`${apiUrl}/cli/text-overrides?${params.toString()}`,
{ headers: { 'X-ScreenCI-Secret': secret } }
)
if (!res.ok) {
if (verbose) {
logger.warn(
`Could not fetch Studio text overrides (${res.status}); using code-declared values.`
)
}
return {}
}
const body = (await res.json()) as { overrides?: unknown }
const overrides = body.overrides
if (
overrides === undefined ||
overrides === null ||
typeof overrides !== 'object' ||
Object.keys(overrides as Record<string, unknown>).length === 0
) {
return {}
}
return { [SCREENCI_VALUES_OVERRIDES_ENV]: JSON.stringify(overrides) }
} catch (error) {
if (verbose) {
logger.warn(
`Could not fetch Studio text overrides; using code-declared values.${
error instanceof Error ? ` (${error.message})` : ''
}`
)
}
return {}
}
}
/**
* Fetch the project's current Studio record-option overrides so they can be
* injected as SCREENCI_RECORD_OPTIONS before a recording. Record options (aspect
* ratio, quality, fps) change the captured viewport/encode, so any Studio edits
* must be present here; code-declared values are the fallback. Best-effort: any
* failure returns an empty env so the SDK uses the code values, and a recording
* is never blocked by this fetch.
*/
async function fetchRecordOptionsEnv(
configPath: string,
verbose: boolean
): Promise<Record<string, string>> {
try {
const { screenciConfig, secret, apiUrl } = await requireScreenCISecret(
configPath,
{ interactive: false }
)
const params = new URLSearchParams({
projectName: screenciConfig.projectName,
})
const res = await fetch(
`${apiUrl}/cli/record-options?${params.toString()}`,
{ headers: { 'X-ScreenCI-Secret': secret } }
)
if (!res.ok) {
if (verbose) {
logger.warn(
`Could not fetch Studio record options (${res.status}); using code-declared values.`
)
}
return {}
}
const body = (await res.json()) as { recordOptions?: unknown }
const recordOptions = body.recordOptions
if (
recordOptions === undefined ||
recordOptions === null ||
typeof recordOptions !== 'object' ||
Object.keys(recordOptions as Record<string, unknown>).length === 0
) {
return {}
}
return { [SCREENCI_RECORD_OPTIONS_ENV]: JSON.stringify(recordOptions) }
} catch (error) {
if (verbose) {
logger.warn(
`Could not fetch Studio record options; using code-declared values.${
error instanceof Error ? ` (${error.message})` : ''
}`
)
}
return {}
}
}
async function run(
command: 'record' | 'test',
additionalArgs: string[],
customConfigPath?: string,
verbose = false,
mockRecord = false,
pollAuth = false,
languages?: string,
noPollAuth = false
) {
const configPath = resolveScreenCIConfigPathOrExit(customConfigPath)
if (command === 'test' || process.env.SCREENCI_RECORDING !== 'true') {
await loadEnvFileFromConfigSource(configPath, false)
}
// Only validate args for record command
if (command === 'record') {
if (!process.env.SCREENCI_SECRET) {
await requireScreenCISecret(configPath, {
interactive: detectInteractiveSession(),
pollAuth,
noPollAuth,
})
}
validateArgs(additionalArgs)
const screenciDir = resolve(dirname(configPath), '.screenci')
clearRecordingDirectories(screenciDir)
}
// Studio text-field overrides are injected for the record command only: they
// resolve in the SDK's `text` fixture before the recording runs. `test` (the
// preview run) keeps code-declared values.
const textOverridesEnv =
command === 'record'
? await fetchTextOverridesEnv(configPath, languages, verbose)
: {}
// Studio record-option overrides are likewise injected for record only: they
// resolve in the SDK's `recordOptions` fixture before the capture runs.
const recordOptionsEnv =
command === 'record' ? await fetchRecordOptionsEnv(configPath, verbose) : {}
const envForChild = { ...process.env }
await validateUniqueDiscoveredTestTitles(configPath, additionalArgs, {
...envForChild,
SCREENCI_CONFIG_DIR: dirname(configPath),
...(command === 'record' ? { SCREENCI_RECORDING: 'true' } : {}),
...(command === 'record' && languages
? { [SCREENCI_LANGUAGES_ENV]: languages }
: {}),
...textOverridesEnv,
...recordOptionsEnv,
...(command === 'test' && !mockRecord
? { [SCREENCI_DISABLE_RECORDING_TIMINGS_ENV]: 'true' }
: {}),
...(command === 'test' && mockRecord
? { [SCREENCI_MOCK_RECORD_ENV]: 'true' }
: {}),
})
if (verbose && process.env.SCREENCI_RECORDING !== 'true') {
logger.info(`Using config: ${configPath}`)
}
const playwrightArgs = ['test', '--config', configPath, ...additionalArgs]
const spawnSpec = resolvePlaywrightSpawnSpec(
playwrightArgs,
dirname(configPath)
)
const child = spawn(spawnSpec.command, spawnSpec.args, {
stdio: 'inherit',
...(process.platform !== 'win32' ? { detached: true } : {}),
...(spawnSpec.shell !== undefined ? { shell: spawnSpec.shell } : {}),
...(spawnSpec.windowsVerbatimArguments !== undefined
? {
windowsVerbatimArguments: spawnSpec.windowsVerbatimArguments,
}
: {}),
env: {
...envForChild,
SCREENCI_CONFIG_DIR: dirname(configPath),
// Enable recording only for record command
...(command === 'record' ? { SCREENCI_RECORDING: 'true' } : {}),
// Per-language filter: the builder records only these languages.
...(command === 'record' && languages
? { [SCREENCI_LANGUAGES_ENV]: languages }
: {}),
// Studio text-field overrides resolved from the backend (record only).
...textOverridesEnv,
// Studio record-option overrides resolved from the backend (record only).
...recordOptionsEnv,
...(command === 'test' && !mockRecord
? { [SCREENCI_DISABLE_RECORDING_TIMINGS_ENV]: 'true' }
: {}),
...(command === 'test' && mockRecord
? { [SCREENCI_MOCK_RECORD_ENV]: 'true' }
: {}),
},
})
const childSignals = forwardChildSignals(child, `screenci ${command}`, {
killTree: process.platform !== 'win32',
exitParentOnForward: true,
})
return new Promise<void>((resolve, reject) => {
child.on('close', (code, signal) => {
void (async () => {
const forwardedSignal = childSignals.getForwardedSignal()
childSignals.cleanup()
if (forwardedSignal) {
process.kill(process.pid, forwardedSignal)
return
}
if (signal) {
process.kill(process.pid, signal)
return
}
if (code === 0) {
resolve()
} else {
reject(new Error(`Playwright exited with code ${code}`))
}
})().catch(reject)
})
child.on('error', (err) => {
childSignals.cleanup()
reject(err)
})
})
}
function logRecordFailureHint(): void {
logger.info('')
logger.info(
`If ${pc.cyan('screenci test')} works but ${pc.cyan(
'screenci record'
)} fails, try ${pc.cyan('screenci test --mock-record')}.`
)
logger.info(`More info: ${pc.cyan(SCREENCI_MOCK_RECORD_DOCS_URL)}`)
}
export function logCliError(error: unknown): void {
if (isPartialUploadError(error)) {
return
}
const errorToLog = isRecordFailureHintError(error) ? error.cause : error
const message =
errorToLog instanceof Error ? errorToLog.message : String(errorToLog)
logger.error(message)
if (isRecordFailureHintError(error)) {
logRecordFailureHint()
}
}
// Only run if this file is being executed directly
// Check if this module is the main module (handles symlinks properly)
const currentFile = fileURLToPath(import.meta.url)
const mainFile = process.argv[1] ? realpathSync(process.argv[1]) : null
const currentRealFile = realpathSync(currentFile)
if (
mainFile &&
(currentFile === mainFile ||
currentRealFile === mainFile ||
currentFile === realpathSync(mainFile))
) {
main().catch((error) => {
logCliError(error)
process.exit(1)
})
}