Creating Workflows
Creating complex multi-step workflows
Building workflows
Workflows represent chaining multiple blocks together. Imagine calling multiple tasks in a row, doing conditional logic, extracting data to a CSV, etc. All of these ideas will be supported within our workflows feature
All of our workflows are defined in YAML format, and allow chaining multiple components together to generate some defined output
Today, we’re building the workflows for most of our customers as we iterate on the specification. This is a cumbersome experience — rest assured that we are improving our web application that will offer a significantly enhanced user experience for this process.
Request - Create Workflow (YAML)
POST api.skyvern.com/api/v1/workflows
Use this API to create a workflow. The response of this API is a workflow_permanent_id, which can be used to run workflows below
| Parameter | Type | Required? | Sample Value | Description |
|---|---|---|---|---|
| title | String | yes | Calculate a product price diff % | A title for a workflow |
| description | String | yes | Compare two products’ price diff % on alibaba vs newlabelwholesale | A description for a workflow |
Workflow Parameters
Workflow parameters are specific parameters you’re going to be passing into the workflows to allow execution
| Parameter | Type | Required? | Sample Value | Description |
|---|---|---|---|---|
| key | String | yes | alibaba_url | unique key corresponding to a specific parameter |
| parameter_type | Enum | yes | workflow | The type of parameter for the workflow. Meant to indicate whether this parameter is being passed in via the run workflow endpoint (workflow), or whether a parameter is the output of a different workflow step (output). Can be workflow , context, aws_secret, or output |
| workflow_parameter_type | Enum | no? | string | The actual type of the parameter, meant to be used for type-safety reasons. Supported types: STRING = “string” INTEGER = “integer” FLOAT = “float” BOOLEAN = “boolean” JSON = “json” FILE_URL = “file_url” |
| description | string | yes | Alibaba product URL for checking the price of the product | Description of the parameter |
Blocks
Blocks are the building block (pun intended) of Skyvern’s workflows. Each block is one discrete task you want to occur. Multiple blocks may be chained together, with outputs from one block being fed as inputs to the next block.
| Parameter | Type | Required? | Sample Value | Description |
|---|---|---|---|---|
| block_type | Enum | yes | Task | Signifying the type of block this is in the workflow |
| label | String | yes | get_alibaba_price | The unique identifier for this block within this workflow |
| parameter_keys | Array | yes | parameter_keys: - alibaba_price - NLW_price | The list of parameters this block depends on for execution |
| output_parameter_key | string | yes | output_parameter_key: price_diff_percentage | The optional output of the block, so that it may be used by other blocks |
| {{ block specific parameters }} | ?? | yes | Other parameters, specific to the block_type specified above. These are covered below |
Building blocks supported today:
- TaskBlock: The magic block. Skyvern navigates through the websites to take actions and/or extract information.
- ForLoopBlock
- CodeBlock
- TextPromptBlock
- DownloadToS3Block
- UploadToS3Block
- SendEmailBlock
- FileParserBlock
To read about specific blocks, check out the block documentation
Managing Credentials and Sensitive Information
This is something the Skyvern team will need to set you up with today. If you’re interested, please book a call with Suchintan
Common concepts
continue_on_failure
continue_on_failure flag indicates whether a failed block execution should block subsequent blocks or not
error_code_mapping
Maps errors to specific error codes so you can have deterministic outputs
persist_browser_session
The persist_browser_session flag indicates whether the browser session should be retained between different workflow runs. When enabled, it uses the same user_data_dir for each run and updates it at the end of each run. This is useful for maintaining the browser state, such as login sessions and cookies, across multiple runs of the same workflow, leading to more efficient and seamless execution.
Note: This flag is set at the workflow level, not the block level, meaning it applies to the entire workflow’s session persistence rather than individual blocks.
output_parameter_key (autogenerated)
Specifies the output parameter of a specific block so it can be re-used in a subsequent block
Its format is always: {label}_output
ie the output parameter for a block like this (which can be referenced in subsequent blocks) would be: login_output
- block_type: task
label: login
parameter_keys:
- credentials
url: website_url
navigation_goal: >-
If you're not on the login page, navigate to login page and login using
the credentials given. First, take actions on promotional popups or cookie prompts that could prevent taking other action on the web page. If you fail to login to find the login page or can't login after several trials, terminate. If login is
completed, you're successful.
data_extraction_goal: >-
Extract anything for the sake of this demo
error_code_mapping:
stuck_with_popups: terminate and return this error if you can't close popups after several tries and can't take the necessary actions on the website because there is a blocking popup on the page
failed_to_login: terminate and return this error if you fail logging in to the page
Example workflow
title: Invoice Downloading Demo (Jun 13)
description: >-
Login to the website, download all the invoices after a date, email the
invoices
workflow_definition:
parameters:
- key: website_url
parameter_type: workflow
workflow_parameter_type: string
- key: credentials
parameter_type: bitwarden_login_credential
bitwarden_client_id_aws_secret_key: SECRET
bitwarden_client_secret_aws_secret_key: SECRET
bitwarden_master_password_aws_secret_key: SECRET
bitwarden_collection_id: SECRET
url_parameter_key: website_url
- key: invoice_retrieval_start_date
parameter_type: workflow
workflow_parameter_type: string
- key: smtp_host
parameter_type: aws_secret
aws_key: SKYVERN_SMTP_HOST_AWS_SES
- key: smtp_port
parameter_type: aws_secret
aws_key: SKYVERN_SMTP_PORT_AWS_SES
- key: smtp_username
parameter_type: aws_secret
aws_key: SKYVERN_SMTP_USERNAME_SES
- key: smtp_password
parameter_type: aws_secret
aws_key: SKYVERN_SMTP_PASSWORD_SES
- parameter_type: context
key: order_history_url
source_parameter_key: get_order_history_page_url_and_qualifying_order_ids_output
- parameter_type: context
key: order_ids
source_parameter_key: get_order_history_page_url_and_qualifying_order_ids_output
- parameter_type: context
key: order_id
source_parameter_key: order_ids
blocks:
- block_type: task
label: login
parameter_keys:
- credentials
url: website_url
navigation_goal: >-
If you're not on the login page, navigate to login page and login using the credentials given, and then navigate to the personal account page. First, take actions on promotional popups or cookie prompts that could prevent taking other action on the web page. Then, try to login and navigate to the personal account page. If you fail to login to find the login page or can't login after several trials, terminate. If you're on the personal account page, consider the goal is completed.
error_code_mapping:
stuck_with_popups: terminate and return this error if you can't close popups after several tries and can't take the necessary actions on the website because there is a blocking popup on the page
failed_to_login: terminate and return this error if you fail logging in to the page
- block_type: task
label: get_order_history_page_url_and_qualifying_order_ids
parameter_keys:
- invoice_retrieval_start_date
navigation_goal: Find the order history page. If there is no orders after given start date, terminate.
data_extraction_goal: >-
You need to extract the order history page url by looking at the current
page you're on. You need to extract contact emails you see on the page. You also need to extract the order ids for orders that
happened on or after invoice_retrieval_start_date. Make sure to filter
only the orders that happened on or after invoice_retrieval_start_date. You need to compare each order's date with the invoice_download_start_date. You can only include an order in the output if the order's date is after or the same as the invoice_download_start_date.
While comparing dates, first compare year, then month, then day. invoice_retrieval_start_date
is in YYYY-MM-DD format. The dates on the websites may be in different formats, compare accordingly and compare year, date, and month.
error_code_mapping:
failed_to_find_order_history_page: return this error if you can't find the order history page on the website
no_orders_found_after_start_date: return this error if there are no orders after the specified invoice_download_start_date
data_schema:
type: object
properties:
order_history_url:
type: url
description: >-
The exact URL of the order history page. Do not make any
assumptions. Return the URL that's passed along in this context.
contact_emails:
type: array
items:
type: string
description: Contact email for the ecommerce website you're on. If you can't find any return null
date_comparison_scratchpad:
type: string
description: >-
You are supposed to filter the orders that happened on or after the invoice_download_start_date. Think through how you will approach this task step-by-step here. Consider these before starting the comparison:
- What format is the order date in? How can you parse it into a structured format?
- What is the correct way to compare two dates?
- How will you compare the order dates to the invoice_download_start_date?
Write out your thought process before filling out the order_ids field below. Remember, the original date may be in any format, so parse it carefully! The invoice_download_start_date will be an exact date you can directly compare against in the format YYYY-MM-DD.
order_ids:
type: array
items:
type: object
properties:
order_date:
type: iso-8601-date-string
order_id:
type: string
description: >-
Return a list of order id strings. Do not return order ids of
orders that happened before the specified
invoice_retrieval_start_date
- block_type: for_loop
label: iterate_over_order_ids
loop_over_parameter_key: order_ids
continue_on_failure: true
loop_blocks:
- block_type: task
label: download_invoice_for_order
complete_on_download: true
continue_on_failure: true
parameter_keys:
- order_id
url: order_history_url
navigation_goal: Download the invoice of the order with the given order ID. Make sure to download the invoice for the given order id. If the element tree doesn't have a matching order id, check the screenshots. Complete if you have successfully downloaded the invoice according to action history, if you were able to download it, you'll see download_triggered=True for the last step. If you don't see a way to download an invoice, navigate to the order page if possible. If there's no way to download an invoice terminate. If the text suggests printing, you can assume you can download it. Return click action with download=True if you want to trigger a download.
error_code_mapping:
not_possible_to_download_invoice: return this error if the website doesn't allow downloading/viewing invoices
cant_solve_captcha: return this error if captcha isn't solved after multiple retries
- block_type: upload_to_s3
label: upload_downloaded_files_to_s3
path: SKYVERN_DOWNLOAD_DIRECTORY
- block_type: send_email
label: send_email
smtp_host_secret_parameter_key: smtp_host
smtp_port_secret_parameter_key: smtp_port
smtp_username_secret_parameter_key: smtp_username
smtp_password_secret_parameter_key: smtp_password
sender: hello@skyvern.com
recipients:
- founders@skyvern.com
subject: Skyvern - Downloaded Invoices Demo
body: website_url
file_attachments:
- SKYVERN_DOWNLOAD_DIRECTORY