Skip to content

task-and-automation.md

Latest commit

 

History

History
95 lines (65 loc) · 2.92 KB

task-and-automation.md

File metadata and controls

95 lines (65 loc) · 2.92 KB

Task And Automation Examples

Browser CLI currently ships with three automation-packaged reference and real-site tasks:

All three also include automation.toml examples for import/export review.

The repository also includes a real-site task example that remains task-first:

That task is useful for direct browser-cli task run ... validation, but it does not yet serve as the canonical automation publish example.

If you are using an installed Browser CLI environment rather than this repo checkout, start by listing the shipped examples:

browser-cli task examples

To bootstrap your own task bundle from the canonical template:

browser-cli task template --output tasks/my_task

Direct Task Runtime Pattern

Use the thin Python runtime inside task.py:

from browser_cli.task_runtime.flow import Flow


def run(flow: Flow, inputs: dict) -> dict:
    flow.open(inputs["url"])
    snapshot = flow.snapshot()
    ref = snapshot.find_ref(role="button", name="Reveal Message")
    flow.click(ref)
    flow.wait_text("Revealed", timeout=5)
    return {"html": flow.html()}

Run it directly:

browser-cli task validate tasks/interactive_reveal_capture
browser-cli task run tasks/interactive_reveal_capture --set url=https://example.com

Real-site examples can be run the same way:

browser-cli task validate tasks/douyin_video_download
browser-cli task run tasks/douyin_video_download --set url=https://v.douyin.com/6Zy2Ip3kk-g

browser-cli task validate tasks/karpathy_nitter_latest_five
browser-cli task run tasks/karpathy_nitter_latest_five

One-Shot Read Pattern

Use Flow.read(...) when a task wants the same one-shot behavior as CLI read:

from browser_cli.task_runtime.flow import Flow


def run(flow: Flow, inputs: dict) -> dict:
    result = flow.read(inputs["url"], output_mode="snapshot", scroll_bottom=True)
    return {"snapshot": result.body}

Automation Pattern

For durable recurring automation, publish or import an automation:

browser-cli automation publish tasks/interactive_reveal_capture
browser-cli automation publish tasks/douyin_video_download
browser-cli automation inspect douyin_video_download
browser-cli automation ui
browser-cli automation status

The automation service:

  • persists automation definitions and run history in the Browser CLI home
  • schedules repeated execution for supported automation schedule modes
  • exposes a local Web UI for editing automation parameters and schedule settings
  • still executes the underlying task through task.py + browser_cli.task_runtime + browser daemon