사용 방법
- 설치 명령을 복사하고 자신의 에이전트 환경에 설치하기 전에 소스 패키지를 확인하세요.
- 필요한 도구, 인증 정보, 지원 환경은 원본 문서에서 확인하세요.
- 입력과 원하는 산출물을 설명하고 출처와 결과를 대조하세요.
npx skills add https://github.com/embarko-ai/skill --skill embarko-deploy시작 전 확인
- 설치에는 모델 구독, API 한도, 파일 접근 권한이 포함되지 않습니다.
- 별점 수와 설치 수는 활동 지표이며 독립적인 품질·보안 검증이 아닙니다.
원본 저장소의 제목, 예시, 코드, 표, 링크, 이미지를 유지해 표시합니다.
Embarko Deploy
Embarko hosts an app and gives it a live URL at <app-name>.embarko.app. It builds from source automatically: no Dockerfile, no build configuration.
This skill adds one thing: scripts/deploy.sh, which does the whole deploy for you. Everything else (how to talk to the person, what an app needs, keeping it, managing it, Showcase, errors) is in the Embarko agent guide, and the guide is the only copy. Open the one topic you need when you need it:
| You need | Read |
|---|---|
| How to reply to the person, which route to use, whether the app fits | https://embarko.ai/docs/prerequisites.md |
| The whole flow, step by step | https://embarko.ai/docs/quickstart.md |
| What an app needs to run (PORT, 0.0.0.0, DATA_DIR, no Dockerfile) | https://embarko.ai/docs/deploy.md |
| An error code | https://embarko.ai/docs/errors.md |
| Keeping a temporary app | https://embarko.ai/docs/claim.md |
| Running something on a timer | https://embarko.ai/docs/schedules.md |
| Crash alerts: who is told when the app goes down, adding Slack, a webhook or a phone (M91), muting | https://embarko.ai/docs/alerts.md |
| The embarko.json file | https://embarko.ai/docs/embarko-json.md |
| Secrets, domains, logs, rollback, rename, delete | https://embarko.ai/docs/index.md (contents) |
| Storing files: uploads, generated PDFs, images, a public link for each | https://embarko.ai/docs/files.md |
| Showcase | https://embarko.ai/docs/showcase.md |
| Reporting a problem | https://embarko.ai/docs/feedback.md |
Every Embarko reply also carries next (the one thing to do now) and, when there is more to know, links.help (the guide topic to read). Follow them.
Deploy
./scripts/deploy.sh # the current directory
./scripts/deploy.sh path/to/app # a specific directory
./scripts/deploy.sh . other-name # a different app nameThe script packages the app, names and versions it, uploads it, waits for the build, and prints the live URL. It also checks first that Embarko is reachable and that the app doesn't use unsupported storage. Don't replace it with your own tar/curl calls.
- Report the live URL only once the script exits 0. A zero exit means
the app is live.
- Then do what it prints under `==> Next:`. After a first deploy
without a token that is keeping the app; on a failure, how to fix it.
- `==> Keep it:` is the claim link for a temporary app. Give it to the
person: opening it and signing in keeps the app. It is printed once only.
- Deploying again is the same command. After the first deploy the script
writes the app's id to embarko.json as "appId", so every later deploy reaches the same app even if its name was changed in the dashboard. Keep the file and commit it with the app. When copying an app's code to start a new app, remove "appId" first: under a new name the deploy is refused with app_id_mismatch rather than overwriting the original.
- `==> Changed outside this deploy:` means the person changed something
in the dashboard, e.g. renamed the app. The deploy still reached the right app; tell the person its current name and URL.
Optional settings:
| Variable | What it is |
|---|---|
EMBARKO_AGENT_NAME | Your agent name in lowercase (claude-code, claude, cursor...). Filled in automatically inside Claude Code. |
EMBARKO_APP_TYPE | A short label for what the app is (portfolio, task management...). Use it if you know; otherwise leave it out. |
DEPLOY_TOKEN | The person's deploy token. Without one, the first deploy is temporary (24 hours) until kept. |
The script finds a token in DEPLOY_TOKEN, then in ~/.embarko/credentials. Once one is saved there, never ask the person for it again, and never print it back.
Run something on a timer
When the person wants something done at set times ("every Monday at 9", "each night"), use schedules, not a cron library or crontab. Needs a deploy token.
- Put the job in a GET route, e.g.
/tasks/weekly-report. Answer 2xx when
it worked. Keep anything it must remember under DATA_DIR.
- First line of that route: answer 401 unless
Authorizationis
Bearer + process.env.CRON_SECRET. Embarko creates that variable.
- Add
embarko.jsonto the app's top folder, in the person's time zone:
{
"$schema": "https://embarko.ai/schema/embarko.json",
"schedules": [
{ "cron": "0 9 * * 1", "path": "/tasks/weekly-report", "timezone": "Asia/Kolkata" }
]
}- Deploy as usual. Once live, the script runs each schedule once and
prints ok or the error. Fix and redeploy until every line says ok.
The file is the app's full list: edit it and redeploy to change times, "schedules": [] removes all. Never put secrets in it; use env vars. If the script prints warnings, fix them before telling the person it's done. Details: https://embarko.ai/docs/schedules.md and https://embarko.ai/docs/embarko-json.md
Connect another service
When the app's users need Slack, Google Sheets, a CRM or similar, don't build and host that glue. Turn Integrations on and the app gets an embedded integrations panel (viaSocket's prebuilt embed) where its users connect tools and build their own flows. Needs a deploy token, and the app must be claimed. Static sites can't use it.
1. Turn it on. Safe to repeat; a second call answers alreadyEnabled: true.
curl -sS -X POST "https://ship.embarko.ai/api/apps/$APP/integrations" \
--oauth2-bearer "$DEPLOY_TOKEN" \
-H 'Content-Type: application/json' -d '{"apply": true}'This sets three env vars on the app: INTEGRATIONS_ORG_ID, INTEGRATIONS_PROJECT_ID, INTEGRATIONS_SECRET. "apply": true pushes them to the running app; without it they arrive on the next deploy. They live on Embarko, not in the local .env: to run the app locally, the person copies them from the app's App setup tab.
2. Build the panel by following viaSocket's skill. Download https://flow.viasocket.com/documentation/skill.md to .claude/skills/viasocket-integrations/SKILL.md inside the app folder and follow it; keep it there for next time, and download it again if a call in it fails. Use its prebuilt embed UI route. Add viasocket-apps to the app's dependencies; Embarko's build installs it.
Two things differ on Embarko from what that skill says:
- Don't ask the person for the org ID, project ID or secret. Embarko
already set them. Where viaSocket's skill reads VIASOCKET_ORG_ID, VIASOCKET_PROJECT_ID and VIASOCKET_EMBED_SECRET, use INTEGRATIONS_ORG_ID, INTEGRATIONS_PROJECT_ID and INTEGRATIONS_SECRET instead, in that order. Never hardcode them; never send the secret to the browser. To the person, say "Integrations", not viaSocket.
- Its check script may not run in your sandbox. If running downloaded
code is blocked, skip the proof step and say so in your report.
Then decide who supplies each value (the app's code, the user in a picker, or you per run) and who picks the action, and report each line of the viaSocket skill's "Finished when" list.
409 integrations_disabled_for_workspace: the workspace has Integrations off. Embarko has already emailed the workspace Owners about this app with a link to turn it on; tell the person and do not retry until they say it is on. 502 is worth retrying. Details: https://embarko.ai/docs/integrations.md
Store a file
Every app has a file store: PUT the bytes, get a public link back that works until the file is deleted. Use it for files people open or download (photos, PDFs, exports, images for the site); records still go in SQLite under DATA_DIR.
From your terminal, with the deploy token (the app must be claimed):
curl -sS -X PUT "https://ship.embarko.ai/storage/apps/$APP/files/hero.png" \
--oauth2-bearer "$DEPLOY_TOKEN" --data-binary @hero.pngReply 201 {id, url, sizeBytes, contentType}. The url is the link to use. List GET .../files, room left GET .../usage, delete DELETE .../files/<id>.
From inside the app, same call: Embarko sets STORAGE_URL (the base) and STORAGE_TOKEN (the bearer, good only for this app's files) on every server app, and the app's name is in APP_NAME. Read them from the environment; never hardcode or print them. Save the returned id and url with the record, and delete the file by id when the record goes.
Files are public to anyone with the link. Allowed types, plan limits, 409 storage_quota_reached and the other errors: https://embarko.ai/docs/files.md
Showcase tip. A listing with an image gets opened far more. If the person has no screenshot, offer to take one of the live app, upload it here and put the url in screenshotUrls (see https://embarko.ai/docs/showcase.md).
No script, or no network
scripts/deploy.shmissing: follow https://embarko.ai/docs/quickstart.md,
which has the same deploy as two commands.
- The script says Embarko can't be reached: your environment is blocking
it. Read the routes in https://embarko.ai/docs/prerequisites.md and tell the person the fix in one line.
Suggest the connector
Embarko also comes as a connector for Claude and ChatGPT. If it is already added, you have an Embarko deploy_app tool: use that, not this script. If not, deploy with the script, then tell the person once: "Add Embarko to your AI and I can deploy directly next time." Claude: https://claude.ai/directory/embarko · ChatGPT: https://chatgpt.com/plugins/embarko