mirror of
https://github.com/actionbook/actionbook.git
synced 2026-09-19 03:28:06 +08:00
sen
132 lines
4.5 KiB
Plaintext
132 lines
4.5 KiB
Plaintext
---
|
|
title: 'Dify Plugin'
|
|
description: 'Use Actionbook inside Dify workflows and agents'
|
|
---
|
|
|
|
<Note>
|
|
The recommended way to get started is the [CLI](/guides/installation). The
|
|
Dify plugin is an **optional, advanced** integration for teams building
|
|
workflows and agents inside Dify.
|
|
</Note>
|
|
|
|
Use this option if you want verified selectors from Actionbook and cloud browser
|
|
automation directly in Dify.
|
|
|
|
## Why Use Actionbook in Dify
|
|
|
|
- Start from verified selectors and action manuals instead of guessing from raw
|
|
HTML.
|
|
- Combine Actionbook search/get tools with live browser automation in one Dify
|
|
workflow.
|
|
- Recover from selector drift by taking a fresh accessibility `snapshot` and
|
|
continuing the run.
|
|
|
|
## Included Tools
|
|
|
|
| Tool | What it does | Best for |
|
|
| --- | --- | --- |
|
|
| `search_actions` | Search Actionbook for verified page areas and selectors | Finding the right element on a site |
|
|
| `get_action_by_area_id` | Fetch the full action manual for a selected area ID | Getting precise CSS, XPath, and ARIA selectors |
|
|
| `browser_create_session` | Start a managed cloud browser session | Opening a browser before multi-step automation |
|
|
| `browser_operator` | Navigate, click, fill, snapshot, and inspect pages | Executing browser steps in a workflow |
|
|
| `browser_stop_session` | Stop the cloud browser session and release resources | Cleanup at the end of a workflow |
|
|
|
|
## Installation
|
|
|
|
Install the plugin directly from the official Dify Marketplace:
|
|
|
|
1. Open the [Actionbook plugin page](https://marketplace.dify.ai/plugin/actionbook/actionbook).
|
|
2. Click `Install`.
|
|
3. Enable the plugin in your Dify workspace.
|
|
4. Configure the plugin credentials before running browser workflows.
|
|
|
|
<Tip>
|
|
The Marketplace page always provides the latest version, so you no longer
|
|
need to download a GitHub release package or build a `.difypkg` manually.
|
|
</Tip>
|
|
|
|
## Credentials
|
|
|
|
### Actionbook API Key
|
|
|
|
- Optional for `search_actions` and `get_action_by_area_id`.
|
|
- Leave it empty to use the free tier, or add a key for higher quotas.
|
|
- Get a key from [actionbook.dev](https://actionbook.dev).
|
|
|
|
### Hyperbrowser API Key
|
|
|
|
- Required for `browser_create_session`, `browser_operator`, and
|
|
`browser_stop_session`.
|
|
- Get a key from [Hyperbrowser](https://app.hyperbrowser.ai/).
|
|
|
|
## Recommended Workflow
|
|
|
|
The most reliable Dify flow is:
|
|
|
|
1. `search_actions` to find a relevant page area.
|
|
2. `get_action_by_area_id` to retrieve the full selector manual.
|
|
3. `browser_create_session` to start a cloud browser session.
|
|
4. `browser_operator` for navigation and page interaction.
|
|
5. `browser_stop_session` to release the session when the workflow finishes.
|
|
|
|
Example sequence:
|
|
|
|
```text
|
|
search_actions(query="github login", domain="github.com")
|
|
get_action_by_area_id(area_id="github.com:login:email-input")
|
|
|
|
browser_create_session()
|
|
|
|
browser_operator(
|
|
session_id="<session_id>",
|
|
cdp_url="<ws_endpoint>",
|
|
action="navigate",
|
|
url="https://github.com/login"
|
|
)
|
|
|
|
browser_operator(
|
|
session_id="<session_id>",
|
|
cdp_url="<ws_endpoint>",
|
|
action="snapshot"
|
|
)
|
|
|
|
browser_stop_session(session_id="<session_id>")
|
|
```
|
|
|
|
<Tip>
|
|
Configure the Hyperbrowser API key in the plugin credentials first.
|
|
`search_actions` and `get_action_by_area_id` are recommended for getting
|
|
verified selectors first, but they are not a hard prerequisite for
|
|
`browser_operator`.
|
|
</Tip>
|
|
|
|
How this maps to the current plugin behavior:
|
|
|
|
- `browser_operator` accepts either `session_id` or `cdp_url`; for multi-step
|
|
workflows, pass both for the best recovery behavior.
|
|
- `browser_create_session` returns `ws_endpoint`, which should be passed to
|
|
`browser_operator` as `cdp_url`.
|
|
- `snapshot` is a fallback step when selectors fail or page state changes, not
|
|
a mandatory success-path step.
|
|
- `browser_stop_session` should be called after each browser workflow to release
|
|
the remote session cleanly.
|
|
|
|
## Browser Operator Tips
|
|
|
|
- Use `fill` for form inputs when possible. It clears and writes the value in a
|
|
single step.
|
|
- Use `snapshot` after navigation, after major page changes, or whenever a
|
|
selector fails.
|
|
- Snapshot responses include accessibility refs like `[ref=e3]`; these are often
|
|
the most reliable selectors for follow-up `click` and `fill` calls.
|
|
- Always call `browser_stop_session` at the end of the workflow to avoid
|
|
unnecessary browser billing.
|
|
|
|
## Example Use Cases
|
|
|
|
- Search or filter content on dynamic web apps.
|
|
- Combine Actionbook search/get tools with browser automation for agentic task
|
|
execution.
|
|
- Recover from selector drift by taking a fresh `snapshot` and continuing the
|
|
workflow.
|