Dane i analityka · Integracja z narzędziem
deploy-connector
/deploy-connector
Pakuje i wdraża konektor Fivetran w serwisie Fivetran, gdy użytkownik chce go opublikować lub wdrożyć.
- Autor
- fivetran ↗
- Repozytorium
- connector_sdk_tools ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 12.11.2026
Ocena i bezpieczeństwo
100/ 100Klasa: dobry
Format25 / 25
Frontmatter z nazwą i opisem, poprawna nazwa, opis do 1024 znaków, nazwa zgodna z folderem.
Odkrywalność25 / 25
Opis mówi, kiedy użyć skilla, ma sensowną długość i jest w 3. osobie.
Treść35 / 35
Co najmniej 150 słów i do 500 linii, nagłówki, listy kroków, przykłady albo kod.
Higiena15 / 15
Bez zaślepek typu TODO, bez ścieżek Windows, z licencją.
Ryzykobrak sygnałów w skanie
- Pliki w paczce
- 2
- Skrypty
- 0
- Komentarze HTML
- 1
- Adresy URL w SKILL.md
- 4
Przestarzałość
Brak sygnałów. Repozytorium zmienione 12.11.2026, bez odwołań do wycofanych modeli i API.
Licencja
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co sprawdzić przed instalacją
- SKILL.md zawiera komentarze HTML (1). Nie widać ich w podglądzie Markdown, a model je czyta.
Co robi i kiedy
Pakuje i wdraża konektor Fivetran w serwisie Fivetran, gdy użytkownik chce go opublikować lub wdrożyć.
Oryginalny opis (EN)
Package and deploy a Fivetran connector to Fivetran. Use when the user wants to deploy or ship their connector.
Jak działa
- Step 1: Pre-Deployment Validation
- Step 2: Run Final Test
- Step 3: Deploy
- Step 4: Offer to Start the Initial Sync (new connections)
- Redeploying (updating an existing connection)
- Reusable Packages (1:N)
- Alternative: Manual Packaging
Sekcje z nagłówków SKILL.md, w oryginale · 254 linii, 2 078 słów
Podgląd SKILL.md
---
name: deploy-connector
description: Package and deploy a Fivetran connector to Fivetran. Use when the user wants to deploy or ship their connector.
---
<!--
GENERATED FILE — DO NOT EDIT.
Canonical source: canonical/skills/deploy-connector/SKILL.md
Regenerate with: bash scripts/sync-plugins.sh
-->
> **Context**: This plugin is for the Fivetran Connector SDK (CSDK). "CSDK" is shorthand for "Connector SDK".
# Deploy Fivetran Connector
**FIRST**: Read `sdk-reference.md` from the plugin directory to load SDK rules and patterns.
Package and deploy the connector in the current directory.
## Step 1: Pre-Deployment Validation
If the user wants to bind or rebind a connection to an existing package (see
**Reusable Packages (1:N)** below) via `--package-id`, no local code is uploaded, so skip the
code-quality checks below and go straight to the binding/rebinding command in that section.
This does **not** skip configuration: the wrapper still reads and submits any local
`configuration.json` unless `--no-configuration` is passed, so a package rebind can silently
overwrite an existing connection's production configuration with unvalidated local values —
review and confirm that file with the user first, or pass `--no-configuration` if the stored
configuration should be left untouched.
Otherwise, verify the connector is ready:
1. **Files exist**: `connector.py`, `requirements.txt`, `README.md`; `configuration.json` when supplying configuration.
2. **Code quality**: Read `connector.py` and check for:
- Both `schema()` and `update()` functions present
- `connector = Connector(update=update, schema=schema)` in global scope
- `if __name__ == "__main__": connector.debug()` entry point
- No forbidden patterns (`Dict[str, Any]`, `Generator[op.Operation, ...]`, `op.Operation` in type hints)
3. **Configuration**: `configuration.json` is JSON. Plaintext string values are supported; do not require custom encryption. Do not read, print, copy, or deploy plaintext configuration values in chat.
## Step 2: Run Final Test
Use the secure runner:
```bash
python "<plugin>/tools/run_connector.py" "<connector_directory>" --timeout-seconds 600
```
The runner defaults to 120 seconds and accepts up to 600. Use 600 for debug
runs because the first run downloads and starts the Java tester. Set the
harness command timeout to 600 seconds as well.
If the test fails, classify the error (INFRA / FIRST_RUN / CODE) and — for CODE errors — apply the fixer workflow (see `workflows/fixer.md` in the plugin, or — in plugins that support subagents — invoke the `connector-fixer` subagent).
## Step 3: Deploy
Check the SDK version with the same executable the wrapper uses: the connector's own
`.venv/bin/fivetran version` (`.venv\Scripts\fivetran.exe version` on Windows), or bare
`fivetran version` only if the connector has no `.venv`. On SDK 3.0.0+ the CLI is stateless. For a
**new** connection, the Python version and agent ID are optional: the user can provide them,
otherwise the SDK default and the destination's default agent are used. A **redeploy** needs the
Python version from the `--python-version` flag, `FIVETRAN_PYTHON_VERSION`, an options file
profile, or `[tool.package-metadata].python_version` in `pyproject.toml`. On a Hybrid Deployment
destination it also needs `--hybrid-deployment-agent-id`, even when the connection uses the
destination's default agent (see **Stateless CLI & Options File** in `sdk-reference.md`). The first
deploy after upgrading auto-selects the migrated profile for the connection, so it may need
neither. Do not ask up front: try the deploy, and if it fails with a missing-field error, ask the
user for the Python version (3.10–3.14) and agent ID and re-run with them. Before 3.0.0, omit them.
For an existing connection, use its ID so the tool reuses its current name and
destination without listing destinations or prompting:
```bash
python "<plugin>/tools/deploy_connector.py" "<connector_directory>" --connection-id "<id>"
```
For a new connection, confirm the connection name with the user before deploying — do not
silently derive it from the directory name and deploy. Tell the user what name will be used
(the directory name, if that's the default you're about to pass) and let them override it.
Supply the destination (group) name and the confirmed connection name:
```bash
python "<plugin>/tools/deploy_connector.py" "<connector_directory>" --destination "<name>" --connection "<name>"
```
If the harness has no way to ask (no input channel), state the name you're about to use and
give the user a chance to stop you before the deploy call runs.
The tool reads `FIVETRAN_API_KEY`, passes local configuration through a named
pipe when present, and invokes `fivetran deploy --destination <name> --connection <name> --force`.
It also forwards these optional flags unchanged: `--python-version`, `--options-file`, `--profile`,
`--proxy-id`, `--proxy-host-config-key`, `--hybrid-deployment-agent-id`. Environment variables such
as `FIVETRAN_PYTHON_VERSION` and `FIVETRAN_OPTIONS_FILE` are inherited.
For an existing connection it reads connection details and then that connection's
group details. Do not combine `--connection-id` with name or destination overrides.
If no destination is supplied for a new connection, a single available destination
is selected automatically. Multiple destinations require a selection on stdin or an explicit argument.
Harnesses without an input channel should pass the target explicitly. Closed
input ends the command; do not retry without providing the target.
A missing local configuration file is allowed; the SDK may still use environment
configuration. Preserve production settings during code repairs and verify the
installed SDK's configuration behavior before deploying. Do not confuse local
test inputs with the production configuration.
For required configuration or unusable encrypted values, follow **Configuration
entry** in `sdk-reference.md`. Plaintext values are supported without a key; do
not require re-entry or encryption of user-supplied values.
### Prerequisite: `FIVETRAN_API_KEY`
If the user hasn't set the env var, the tool exits with a clear message. Direct the user to:
1. Create a Fivetran API key at https://fivetran.com/dashboard/user/api-config. It must be the base64-encoded `{key}:{secret}` string, with permission to manage connections and read destinations (so destination lookup, deploy, and unpause all work).
2. Add it to their shell config.
macOS/Linux:
```bash
export FIVETRAN_API_KEY=...
```
Windows PowerShell:
```powershell
setx FIVETRAN_API_KEY "..."
```
3. Reload their shell and re-run the deploy command.
### If no destinations exist
If the user has zero destinations, the tool exits with a link to the destinations page. Direct the user to create one in the dashboard (requires warehouse credentials) and re-run deploy.
Reference: https://fivetran.com/docs/connector-sdk/working-with-connector-sdk#deploytheconnector
## Step 4: Offer to Start the Initial Sync (new connections)
A newly deployed connection is created **paused**. Deploying does not start a sync.
After a successful deploy, surface the Connection ID and dashboard link the tool printed, then **ask the user** whether to start the initial sync now. State plainly that starting the sync begins consuming [MAR](https://fivetran.com/docs/core-concepts/usage-based-pricing#monthlyactiverows). Do not start it automatically.
Only if the user explicitly confirms, unpause the connection:
```bash
python "<plugin>/tools/deploy_connector.py" "<connector_directory>" --start-sync --connection-id "<id>"
```
This calls `PATCH /v1/connections/{id}` with `{"paused": false}`; Fivetran then begins the initial sync. If the user declines, tell them they can start it anytime from the dashboard link or by re-running the command above.
## Redeploying (updating an existing connection)
To update a deployed connection, use `--connection-id <id>`. This preserves its
name and destination even when the recovered project directory has a different
name. Redeployment replaces code and supplied configuration; it does not itself
unpause the connection. On SDK 3.0.0+ other settings (Python version, proxy, hybrid agent) are not
carried over from the existing connection: supply them via flags, env vars, the options file, or (Python version only) `pyproject.toml`. The
first deploy after upgrading uses the migrated profile instead. Otherwise a redeploy fails without
the Python version, and without the agent ID on a Hybrid Deployment destination.
### Troubleshooting (SDK 3.0.0+)
- `--python-version is required` (redeploy) — pass `--python-version <ver>`, set `FIVETRAN_PYTHON_VERSION`,
add it to the options file profile, or set `python_version` under `[tool.package-metadata]` in `pyproject.toml`.
- `--hybrid-deployment-agent-id is required` (redeploy on a Hybrid Deployment destination) — pass
`--hybrid-deployment-agent-id <id>`, set `FIVETRAN_HD_AGENT_ID`, or add `hybrid_deployment_agent_id`
to the options file profile.
- `migration failed: 404 ... connector-sdk/migration/connections` — non-fatal. The one-time import of
existing connection settings into `~/.fivetran/connector_sdk/options.toml` is all-or-nothing: no
file is written and it is retried on the next deploy; supply the values explicitly meanwhile.
If the connection's current code package is shared with other connections (see **Reusable
Packages (1:N)** below), a normal redeploy is refused. A plain redeploy always uploads a
brand-new package and points the connection at it (it never edits the old package's code in
place), which would silently detach this connection from the shared group — so the tool blocks
it instead. Rebind it explicitly with `--package-id <id>` to a *different* package if that's
genuinely what's wanted, or confirm with the user whether they actually want a one-off divergent
copy (in which case, deploy to a different connection rather than forcing this one off its shared
package).
## Reusable Packages (1:N)
A reusable package lets the *same* uploaded code run identically across many connections (e.g.
one per customer/tenant), instead of every connection holding its own independent copy. This is a
Private Preview CLI feature — the flags exist and work but are hidden from `fivetran --help`.
**Creating a package without a connection** (direct CLI call, not through the wrapper — these
subcommands have no configuration to protect, so the wrapper's encryption/pipe handling isn't
needed):
```bash
fivetran package create "<connector_directory>" --yes
```
This uploads the project and prints a server-assigned `package id: <id>`. Share that ID with the
user; it's what every bound connection will reference.
**Updating an existing package's code** (pushes new code to every connection using it, on their
next sync):
```bash
fivetran package update "<package-id>" "<connector_directory>" --yes
```
**Always pass `--yes`** on `package create`/`package update`: like `deploy`, both run the same
`requirements.txt` dependency check, which otherwise prompts interactively (and would block an
agent run) when it detects missing/mismatched dependencies. `--yes` auto-accepts and lets the tool
fix `requirements.txt` for you; prefer it over `--force`, which skips the dependency check
entirely instead of just auto-answering its prompt.
**Listing packages in the account** (package ID and how many connections use each; when the API exposes `package_name`/`visible`, `PACKAGE NAME` and `VISIBLE` columns are also shown):
```bash
fivetran package list
```
- **Visible packages (connection creation via setup form):** set `[tool.package-metadata].visible = true` (a native TOML boolean) in `pyproject.toml` to make a reusable package appear in the Fivetran dashboard so users can create connections from it via its setup form. A visible package **must** implement [`configuration_form()`](https://fivetran.com/docs/connector-sdk/technical-reference/connector-sdk-methods#configurationform) — `package create`/`package update` are rejected server-side (HTTP 4xx, `A visible package must implement configuration_form()`) otherwise, so confirm the connector defines one before marking it visible. A plain redeploy (no `--package-id`) of a connection whose current package is `visible` is refused with exit 1, the same as a shared package. A visible package can only be redeployed with a new package, so use the rebind form below with `--package-id <id>` to deploy a new package version.
- **Setup-form metadata validation** (`package create`/`package update`, only when the response carries `package_name`/`visible`): the CLI reads the local `pyproject.toml` and validates package metadata. It never edits the file.
- `[project].name`: a valid name is at least 3 characters, uses only ASCII letters, digits, `.`, `_` or `-`, and starts and ends with a letter or digit. If **missing**, the platform falls back to the package ID. If **invalid** (does not meet the rules), **deployment fails**.
- `[tool.package-metadata].visible`: must be a native TOML boolean (`true` or `false`), not a string (e.g. not the quoted string `"true"`). If **non-boolean**, **deployment fails**.
- `[tool.package-metadata].python_version`: must be a supported Python version for the connector SDK. If **unsupported**, **deployment fails**.
- `[tool.package-metadata].proxy_host_config_key` (Proxy Agent packages): names the `configuration` key that holds the proxy `host:port` endpoint(s), and that key must be defined in the connector's `configuration_form()`. If it is omitted from `pyproject.toml`, the setup form falls back to the default `host`/`hosts` keys, so one of those default keys must instead be defined in `configuration_form()`
**Binding a new connection to an existing package** (no local code is uploaded — confirm the
connection name with the user first, same as a normal new-connection deploy in Step 3):
```bash
python "<plugin>/tools/deploy_connector.py" "<connector_directory>" --destination "<name>" --connection "<name>" --package-id "<package-id>"
```
**Rebinding an existing connection to a (possibly different) package**:
```bash
python "<plugin>/tools/deploy_connector.py" "<connector_directory>" --connection-id "<id>" --package-id "<package-id>"
```
- If the connection already uses that package, this is a no-op.
- Otherwise it replaces the connection's code (and configuration, if supplied) with the target
package's. The wrapper always passes `--force`, so this happens without an interactive prompt —
confirm the rebind with the user before running it, the same way you'd confirm any other
destructive redeploy.
- If the connection's previous package was shared with other connections, the tool reports that
the connection left that package group; the other connections are unaffected and keep using it.
`<connector_directory>` only needs to exist as a directory for this flow — `connector.py` is not
required locally since no code is uploaded.
## Alternative: Manual Packaging
If the user prefers manual deployment (e.g., wants to inspect the package before upload):
1. Build the deployable archive:
```bash
fivetran package
```
This produces a ZIP containing `connector.py`, `configuration.json`, `requirements.txt` (or `pyproject.toml`), `README.md`, and any additional source files, respecting `.gitignore`.
2. Upload via the Fivetran dashboard.
Pliki w paczce (2)
- SKILL.md 15 KB
- agents/openai.yaml 1 KB
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · fivetran/connector_sdk_tools | 100 | MIT |
| fivetran/connector_sdk_tools | 100 | MIT |
| fivetran/connector_sdk_tools | 100 | MIT |
Migawka bazy z 13.11.2026 · wersja paczki 5a90be75ff01 (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
- 100build-connector
Buduje nowy konektor Fivetran na podstawie opisu źródła danych API i tworzy jego szkielet. - 100build-connector
Tworzy nowy konektor Fivetran z opisu: generuje szkielet konektora dla API lub źródła danych. - 100fivetran-webhooks
Odbiera i weryfikuje webhooki Fivetran, w tym podpis X-Fivetran-Signature-256 i zdarzenia potoków danych, takie jak sync_start, sync_end i connection_failure. - 100fivetran
Zarządza potokami danych Fivetran przez serwer MCP: listuje, tworzy, zmienia, synchronizuje i monitoruje połączenia, miejsca docelowe, transformacje, webhooki i grupy. - 95migrating-from-fivetran
Przenosi konektory Fivetran do źródeł hurtowni danych PostHog: raportuje, które połączenia przechodzą bez zmian, a które wymagają poprawek. - 100fivetran-dagster-to-orchestra
Przekształca integrację Dagster z Fivetran (FivetranResource, FivetranWorkspace, build_fivetran_assets) w równoważne zadanie potoku Orchestra.