Add Selenium plugin for browser automation

Add Selenium plugin for browser automation

Summary

Add a Selenium plugin so Kestra flows can drive a real browser: navigate pages, fill and
submit forms, click elements, wait for content, extract text, run JavaScript, capture
screenshots, and download files. It connects to a remote Selenium Grid over WebDriver, so
the browser runs in a container and the flow stays declarative.

This reaches data and actions that only exist inside a browser (sites with no API, content
rendered by JavaScript, multi-step web workflows), which no existing Kestra plugin can do.

Motivation

Today, anything that requires a browser falls outside Kestra. Users work around it with
hand-written Python/Selenium scripts in a Python task, external RPA tools, or manual work.
Common needs that have no API and no plugin:

• Scraping data from pages that render content with JavaScript.
• Filling and submitting forms in legacy systems or vendor portals that offer no API.
• End-to-end / synthetic monitoring of a login or checkout path.
• Downloading invoices, statements, and reports from portals behind a login.

A first-class task keeps these in the same flow as the upstream preparation, retries,
branching, and notifications, instead of a black-box script.

Plugin Structure

• Repository: plugin-selenium (OSS, public)
• Namespace: io.kestra.plugin.selenium
• Category: TOOL
• Reference model: scaffolded from plugin-apify conventions.

Scope (this issue)

One core task, Browse, that opens a single WebDriver session against a Selenium Grid,
runs an ordered list of actions in that session, then closes it. A session cannot span
multiple Kestra tasks (a live WebDriver is not serializable and tasks run independently),
so a single task with an actions list is the correct model for multi-step automation.

io.kestra.plugin.selenium.Browse

Connection properties (shared, on an AbstractSeleniumTask base):

• remoteUrl (required): WebDriver / Grid endpoint, e.g. http://localhost:4444.
• browser: CHROME (default), FIREFOX, or EDGE.
• headless: default true.
• pageLoadTimeout: default PT30S.
• capabilities: extra browser capabilities merged into the options.

Actions (the action field on each list item):

1. NAVIGATE — go to a URL.
2. CLICK — click an element by CSS selector.
3. TYPE — type text into an element.
4. WAIT_FOR — wait until an element is present (waitTimeout, default PT10S).
5. EXTRACT_TEXT — read text from one element, or all matches when multiple: true.
6. SCREENSHOT — capture the viewport, stored in Kestra internal storage.
7. EXECUTE_SCRIPT — run JavaScript and capture the return value.
8. DOWNLOAD — trigger and capture a file download. Uses Selenium 4 managed downloads
   (se:downloadsEnabled + HasDownloads), so files that land on the grid node are
   retrieved and stored in Kestra.

Outputs:

• extracted (Map): texts from EXTRACT_TEXT, keyed by action id or extract_<index>.
• screenshots (Map): internal storage URIs, keyed by filename.
• scriptResults (Map): return values from EXECUTE_SCRIPT.
• downloads (Map): internal storage URIs of downloaded files, keyed by filename.

Gradle Dependencies

// Selenium WebDriver Java client
implementation "org.seleniumhq.selenium:selenium-java:4.27.0"

Browser runtime for development and tests

Integration tests connect to a Selenium Grid over remoteUrl. On Apple Silicon the
multi-arch Chromium image is required (selenium/standalone-chromium, 4.21.0+). A
docker-compose-ci.yml starts the grid on port 4444 (and noVNC on 7900), with
SE_NODE_ENABLE_MANAGED_DOWNLOADS=true so the DOWNLOAD action works.

YAML Examples

Example 1 — Scrape headlines and capture a screenshot

id: selenium_scrape
namespace: company.web

tasks:
  - id: browse
    type: io.kestra.plugin.selenium.Browse
    remoteUrl: "{{ secret('SELENIUM_GRID_URL') }}"
    browser: CHROME
    headless: true
    actions:
      - action: NAVIGATE
        url: "https://news.ycombinator.com"
      - action: WAIT_FOR
        selector: ".titleline"
      - action: EXTRACT_TEXT
        id: headlines
        selector: ".titleline > a"
        multiple: true
      - action: SCREENSHOT
        name: frontpage.png

  - id: log
    type: io.kestra.plugin.core.log.Log
    message: "Scraped {{ outputs.browse.extracted.headlines | length }} headlines"

Example 2 — Fill and submit a login form, read the result

id: form_login
namespace: company.web

inputs:
  - id: username
    type: STRING

tasks:
  - id: login
    type: io.kestra.plugin.selenium.Browse
    remoteUrl: "{{ secret('SELENIUM_GRID_URL') }}"
    actions:
      - action: NAVIGATE
        url: "https://the-internet.herokuapp.com/login"
      - action: TYPE
        selector: "#username"
        value: "{{ inputs.username }}"
      - action: TYPE
        selector: "#password"
        value: "{{ secret('APP_PASSWORD') }}"
      - action: CLICK
        selector: "button[type='submit']"
      - action: WAIT_FOR
        selector: "#flash"
      - action: EXTRACT_TEXT
        id: flash
        selector: "#flash"

  - id: log
    type: io.kestra.plugin.core.log.Log
    message: "{{ outputs.login.extracted.flash }}"

Example 3 — Download a file from a page and store it

id: selenium_download
namespace: company.ops

tasks:
  - id: fetch
    type: io.kestra.plugin.selenium.Browse
    remoteUrl: "{{ secret('SELENIUM_GRID_URL') }}"
    actions:
      - action: NAVIGATE
        url: "https://the-internet.herokuapp.com/download"
      - action: WAIT_FOR
        selector: ".example a"
      - action: DOWNLOAD
        selector: ".example a:first-of-type"

  - id: log
    type: io.kestra.plugin.core.log.Log
    message: "Stored files: {{ outputs.fetch.downloads }}"