pine-computer-sdk
/pine-computer-sdk
Buduje produkt na bazie Pine Computer przez SDK w TypeScript, Go, Ruby lub Python: tworzy komputery, przekazuje zadania i pokazuje pulpit.
- Autor
- PineComputer ↗
- Repozytorium
- pine-computer ↗
- Licencja
- Apache-2.0
- Ostatnia zmiana repozytorium
- 13.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 3
Brak sygnałów. Repozytorium zmienione 13.11.2026, bez odwołań do wycofanych modeli i API.
Apache-2.0, otwarta: można używać i zmieniać, z zachowaniem licencji i informacji o zmianach.
Co robi i kiedy
Buduje produkt na bazie Pine Computer przez SDK w TypeScript, Go, Ruby lub Python: tworzy komputery, przekazuje zadania i pokazuje pulpit.
Oryginalny opis (EN)
Build Pine Computer into a product with its SDKs. Use when writing or changing code that creates a Pine Computer, hands its agent a job, shows its live desktop in a page, lets a person take the controls, or moves files in and out, in TypeScript, Go, Ruby or Python. Not for handing a job to a Computer from this chat; that is the pine-computer skill.
Jak działa
- Pick the way in
- Read the guide for the version in the project
- What holds in every version
- Before you say it works
Podgląd SKILL.md
---
name: pine-computer-sdk
description: Build Pine Computer into a product with its SDKs. Use when writing or changing code that creates a Pine Computer, hands its agent a job, shows its live desktop in a page, lets a person take the controls, or moves files in and out, in TypeScript, Go, Ruby or Python. Not for handing a job to a Computer from this chat; that is the pine-computer skill.
---
# Build with Pine Computer
A Pine Computer is a cloud computer with an agent built in. A product gives it a job in plain words, shows
its screen to the person, lets them step in, and gets the finished work back. The product keeps its own
model, logic and interface.
This skill tells you which way in fits and where the facts are. It does not restate the API: the API belongs
to the SDK version the project has installed, and the guide for that version ships inside the package.
## Pick the way in
| The job | Use |
| --- | --- |
| An agent or chat app should be able to hand work to a Computer | The tools or a bot from the `pine-computer` CLI (`pine-computer mcp`, `acp`, `serve`). No SDK code. See `docs/cli.md`. |
| A product of your own should run Computers for its users | A server SDK in the backend, and the web SDK in the page that shows the desktop. |
| A new product, from nothing | Copy `examples/starter` and change it. It already has sessions, the live work, the controls, files and a vault. |
`docs/cli.md` and `examples/starter` are in https://github.com/PineComputer/pine-computer. If that repository
isn't on disk, read them there.
## Read the guide for the version in the project
Do this before writing any SDK call. Don't write one from memory.
1. **Find the SDK the project uses**: the dependency in its manifest (`package.json`, `go.mod`, `Gemfile`,
`pyproject.toml`), and the version its lockfile resolves to.
2. **Read the guide that came with it.** In a JavaScript project, resolve the package from the app's own
folder (in a workspace it may sit in a parent `node_modules`) and read its `README.md` and `CHANGELOG.md`:
- `@pinesandbox/computer`: the server SDK. Computers, sessions, Pine's agent, files, artifacts, control,
skills, errors.
- `@pinesandbox/computer-web`: the browser SDK. The `<pine-desktop>` element and its React and Vue adapters.
For Go (`go.pinesandbox.io/computer`), Ruby (`pinesandbox`) and Python (`pinesandbox`), read the README
inside the installed module, gem or wheel. They have the same objects and verbs as the TypeScript SDK.
3. **Not installed yet?** Install the SDK first, then read what was installed: `npm install
@pinesandbox/computer` (Node.js 22 or later), and `@pinesandbox/computer-web` for a page that shows the
desktop. The project key and the endpoint come from the Console (https://console.pinecomputer.io) and go
in the server's environment; the guide names the variables. If you can't install, say so and use the
integration guide at https://console.pinecomputer.io/docs/integration, and tell the person the guide may
be for a newer version than the one they end up with.
4. **A type or method the guide doesn't cover**: read the package's type declarations. If it isn't there, it
isn't in that version.
## What holds in every version
- **The server SDK runs on the server only.** It holds the project key, the Computer's credentials and the
session tokens, and it refuses to load in a browser.
- **The browser gets one thing**: a desktop connection your backend issues for one session, for a signed-in
user, with `Cache-Control: no-store`. It lasts minutes, so the page asks for a fresh one each time it
connects. Never send the project key or a session token to a page.
- **Save three things, as secrets, per user or per workspace.** The Computer's credentials, before you create
it: they are the only way to start it again. Its binding, after every create and every start. A session's
name and token, if the session is to be continued later.
- **Hand over the job, not the clicks.** Give Pine's agent the goal and what done looks like; don't script the
browser step by step, even where the guide shows how.
- **Activity is for showing, waiting is for the outcome.** The stream of the agent's activity doesn't end by
itself when the job does: wait on the turn for its result, and end the stream then. A step arrives once as
it starts and again with its outcome. Where nothing shows the screen, turn the screen frames off.
- **Build the question and the take-over into the product.** The agent can stop and ask, and a person can take
the controls and hand them back. A product that only shows a spinner until the result has thrown away the
two things people need most.
- **Decide what needs the person's yes in your own code.** If your product must never spend, send or submit
without approval, say so in the goal and check it in your flow. Don't rely on the agent's habit alone.
- **Stop what you started.** A script or a test stops its Computer when it is done, also when it fails, and
treats a stop that isn't confirmed as not stopped. A product lets the idle timeout do it.
## Before you say it works
Run it against a real Computer with a real key. A type-check is not a run.
- **A backend**: create a Computer, hand it a small job, see the activity arrive and the result come back,
issue a desktop connection, stop the Computer.
- **A product with a page**: all of that, and open the desktop in the page, take the controls and hand them
back, and bring one file back.
Wersje
W bazie nie ma innych skilli o nazwie pine-computer-sdk.
Migawka bazy z 13.11.2026 · wersja paczki 8f55ef49f8ee (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 90pine-computer
Zleca pracę na prawdziwym komputerze w chmurze: logowanie na stronach, wypełnianie formularzy i gotowy plik, np. PDF lub arkusz. - 100pinescript-validation
Diagnozuje i naprawia błędy Pine Script v6 deterministycznie: czyta diagnostykę walidatora, stosuje znaną poprawkę i waliduje ponownie, także w istniejącej bazie zepsutych plików .pine. - 100mph-pine
Tworzy wskaźniki TradingView w Pine Script v6 w stylu autora: panel synthwave, wierna matematyka i zgodność z v6. - 100mph-pine
Tworzy wskaźniki TradingView w Pine Script v6 w stylu autora: panel synthwave, wierna matematyka i zgodność z v6. - 100pine-developer
Pisze produkcyjny kod Pine Script v6 zgodnie z wytycznymi TradingView: wskaźniki, strategie i inne funkcje. - 100pinescript
Opisuje wzorce Pine Script v6: zasady składni, wydajność, debugowanie, backtesting i wizualizację, przy pisaniu wskaźników i strategii dla TradingView.