Step Types

Complete reference for all 46 step types available in Kadmo workflows (43 implemented, 3 chat types pending).

Browser Steps

browser.navigate

Navigate to a URL.

- type: browser.navigate
  url: "https://example.com"
  new_tab: false  # Optional: open in new tab

Parameter	Type	Required	Description
url	string	Yes	URL to navigate to
new_tab	boolean	No	Open in new tab (default: false)

browser.click

Click an element.

- type: browser.click
  selector: "button.submit"
  new_tab: false  # Optional: Ctrl+click to open in new tab

Parameter	Type	Required	Description
selector	string	Yes	CSS selector or text selector
new_tab	boolean	No	Ctrl+click for new tab

Selector formats:

• CSS: button.submit, #login-btn
• Text: button:has-text('Submit')
• Aria: [aria-label="Close"]

browser.type

Type text into an input.

- type: browser.type
  selector: "input[name='email']"
  text: "user@example.com"

Parameter	Type	Required	Description
selector	string	Yes	CSS selector for input
text	string	Yes	Text to type

browser.scroll

Scroll the page.

- type: browser.scroll
  direction: down
  amount: 500  # pixels

Parameter	Type	Required	Description
direction	string	Yes	up, down, left, right
amount	number	No	Pixels to scroll (default: 300)

Scroll in containers

Use browser.hover first to position the cursor inside a scrollable container.

browser.hover

Hover over an element.

- type: browser.hover
  selector: ".dropdown-menu"

Parameter	Type	Required	Description
selector	string	Yes	CSS selector

browser.wait

Wait for element or delay.

# Wait for element
- type: browser.wait
  selector: ".loading-complete"
  timeout: 10000

# Fixed delay
- type: browser.wait
  ms: 2000

# Named delay constant (resolves to base value +/-10% jitter)
- type: browser.wait
  ms: "mid"

Parameter	Type	Required	Description
selector	string	No	Wait for element to appear
ms	number | DelayConstant	No	Fixed delay in milliseconds or named constant
timeout	number	No	Max wait time (default: 30000)

browser.extract

Extract text from an element.

- type: browser.extract
  selector: "h1"
  as: page_title

Parameter	Type	Required	Description
selector	string	Yes	CSS selector
as	string	Yes	Variable name to store result

browser.extractAll

Extract text from multiple elements.

- type: browser.extractAll
  selector: ".list-item"
  as: items
  separator: ", "  # Optional

Parameter	Type	Required	Description
selector	string	Yes	CSS selector
as	string	Yes	Variable name
separator	string	No	Join separator (default: ", ")

browser.extractMap

Extract structured data from repeated elements, mapping child selectors to named fields.

- type: browser.extractMap
  selector: ".user-row"
  fields:
    name:
      selector: ".user-name"
    email:
      selector: ".user-email"
    role:
      selector: ".user-role"
      attribute: "data-role"
  as: user_data

Parameter	Type	Required	Description
selector	string	Yes	CSS selector for container elements
fields	object	Yes	Map of field names to { selector, attribute? }
as	string	Yes	Variable name to store result
separator	string	No	Row separator (default: newline)

browser.exists

Check if element exists.

- type: browser.exists
  selector: ".error-message"
  as: has_error
  timeout: 1000

Parameter	Type	Required	Description
selector	string	Yes	CSS selector
as	string	Yes	Variable name (boolean)
timeout	number	No	Wait timeout (default: 1000)

browser.key

Send keyboard key.

- type: browser.key
  key: Enter
  selector: "input.search"  # Optional: focus first

Parameter	Type	Required	Description
key	string	Yes	Key name
selector	string	No	Element to focus first

Supported keys: Escape, Enter, Tab, ArrowUp, ArrowDown, ArrowLeft, ArrowRight, Backspace, Delete, Space

browser.snapshot

Capture an accessibility snapshot of the page. Returns a structured YAML representation of the page's accessibility tree, useful for LLM-driven analysis.

- type: browser.snapshot
  as: page_snapshot
  includeContent: true

Parameter	Type	Required	Description
as	string	Yes	Variable name to store snapshot
includeContent	boolean	No	Include visible text content (default: false)

browser.screenshot

Capture the page — or a crop of it — and write a PNG to a file on the machine running the workflow. The image bytes never enter the agent's context; the step returns only the path it wrote.

- type: browser.screenshot
  selector: "#app"
  path: "~/shots/01-dashboard.png"
  as: shot_path

Parameter	Type	Required	Description
path	string	Yes	Where to write the PNG. ~/…, absolute, or relative to ~/workspace
selector	string	No	CSS selector to crop to. Empty or omitted = full viewport
ref	number	No	Element reference from a prior browser.snapshot
region	object	No	Manual crop: {x, y, width, height} in CSS pixels
as	string	No	Variable name to store the written path

Parent directories are created as needed, the file is written 0600, and an existing file at the same path is overwritten — so a retried step lands on the same file rather than leaving a numbered trail.

Paths are contained to the home directory

path must resolve inside the home directory of the user running the agent runtime; .. traversal and symlinks that escape it are rejected. This matches the rule publish_artifact enforces, so anything this step writes can be published directly.

Redact before you capture

Pair it with browser.injectCSS to blur sensitive elements before the shot, so no unredacted image is ever written:

- type: browser.navigate
  url: "{{url}}"
- type: browser.injectCSS
  id: sb-redact
  css: ".user-email, .invoice-total { filter: blur(6px); }"
- type: browser.wait
  ms: 400
- type: browser.screenshot
  selector: "#app"
  path: "~/shots/01-dashboard.png"

This is exactly the bundled docs_screenshot workflow.

browser.injectCSS

Inject CSS styles into the page.

- type: browser.injectCSS
  css: |
    .highlight { background: yellow; }
  id: "my-styles"

Parameter	Type	Required	Description
css	string	Yes	CSS rules to inject
id	string	No	Style element ID for replacement/idempotency

Idempotent injection

When id is provided and a style element with that ID already exists, its content is replaced instead of creating a duplicate.

browser.injectJS

Inject JavaScript into the page.

- type: browser.injectJS
  js: |
    document.querySelector('.dropdown').click();
  id: "my-script"

Parameter	Type	Required	Description
js	string	Yes	JavaScript code to inject
id	string	No	Script element ID for replacement/idempotency

Idempotent injection

When id is provided and a script element with that ID already exists, its content is replaced instead of creating a duplicate.

browser.select_option

Select an option from a <select> dropdown. Matches by value or visible label text.

- type: browser.select_option
  selector: "select#country"
  label: "United States"

Parameter	Type	Required	Description
selector	string	Yes	CSS selector for <select> element
value	string	No	Option value to select
label	string	No	Option visible text to select

browser.scrollIntoView

Scroll an element into the visible viewport.

- type: browser.scrollIntoView
  selector: "#footer-section"
  block: center

Parameter	Type	Required	Description
selector	string	Yes	CSS selector
block	string	No	Scroll alignment: start, center (default), end, nearest

browser.fill

Set an input value directly. Works with React controlled inputs by using native value setters and dispatching synthetic events.

- type: browser.fill
  selector: "input[name='date']"
  value: "2026-03-01"

Parameter	Type	Required	Description
selector	string	Yes	CSS selector for input
value	string	Yes	Value to set

When to use fill vs type

Use browser.fill instead of browser.type for date inputs, React-controlled inputs, or when you need to set a value without triggering the native input UI (e.g., date pickers).

Shell Steps

shell.run

Execute a shell command.

- type: shell.run
  cmd: "echo 'Hello World'"
  cwd: "/path/to/dir"  # Optional
  as: output          # Optional

Parameter	Type	Required	Description
cmd	string	Yes	Command to run
cwd	string	No	Working directory
as	string	No	Variable for stdout

terminal.open

Open a visible terminal window (macOS).

- type: terminal.open
  title: "My Terminal"
  cwd: "/path/to/project"

Parameter	Type	Required	Description
title	string	No	Terminal window title
cwd	string	No	Starting directory

terminal.run

Run command in open terminal.

- type: terminal.run
  cmd: "npm run dev"

Parameter	Type	Required	Description
cmd	string	Yes	Command to run

LLM Steps

Requires API Key

Set OPENAI_API_KEY environment variable or configure in Settings.

llm.classify

Classify text into categories.

- type: llm.classify
  input: "{{message}}"
  categories:
    - urgent
    - normal
    - spam
  as: category

Parameter	Type	Required	Description
input	string	Yes	Text to classify
categories	array	Yes	List of categories
as	string	Yes	Variable for result

llm.generate

Generate text with AI.

- type: llm.generate
  prompt: "Summarize this article:\n{{content}}"
  as: summary

Parameter	Type	Required	Description
prompt	string	Yes	Prompt for the LLM
as	string	Yes	Variable for result

llm.decide

Present a situation and a list of possible actions to the LLM, and let it pick the best one based on its role context.

- type: llm.decide
  input: "{{situation}}"
  actions:
    - id: escalate
      description: "Escalate to a human"
    - id: respond
      description: "Send an automated response"
    - id: ignore
      description: "No action needed"
  as: chosen_action

Parameter	Type	Required	Description
input	string	Yes	Situation description
actions	array	Yes	List of { id, description } actions
as	string	Yes	Variable for chosen action ID

Control Flow Steps

control.if

Conditional branching.

- type: control.if
  condition: "{{status}} == 'ready'"
  then:
    - type: browser.click
      selector: ".start-btn"
  else_steps:
    - type: control.stop
      message: "Not ready"

Parameter	Type	Required	Description
condition	string	Yes	Condition to evaluate
then	array	Yes	Steps if true
else_steps	array	No	Steps if false

Operators: ==, !=

control.retry

Retry steps on failure.

- type: control.retry
  max_attempts: 3
  delay_ms: 1000
  steps:
    - type: browser.click
      selector: ".flaky-button"

Parameter	Type	Required	Description
steps	array	Yes	Steps to retry
max_attempts	number	No	Max retries (default: 3)
delay_ms	number | DelayConstant	No	Delay between retries (default: 1000)

control.foreach

Iterate over a list of items.

- type: control.foreach
  items: "{{user_list}}"
  as: user
  separator: ","
  delay_ms: "small"
  steps:
    - type: browser.click
      selector: "button[data-user='{{user}}']"

Parameter	Type	Required	Description
items	string	Yes	Interpolated string to split into items
as	string	Yes	Variable name for current item
steps	array	Yes	Steps to execute per item
separator	string	No	Item separator (default: ,)
index_as	string	No	Variable name for current index
max_items	number	No	Max items to process (default: 1000)
delay_ms	number | DelayConstant	No	Delay between iterations (default: 0)

control.stop

Stop workflow execution.

- type: control.stop
  message: "Completed successfully"

Parameter	Type	Required	Description
message	string	No	Message to log

Workflow Steps

workflow.call

Call another workflow.

- type: workflow.call
  workflow: extract_data
  params:
    url: "{{target_url}}"
  as: data  # Namespace for child variables

Parameter	Type	Required	Description
workflow	string	Yes	Workflow ID to call
params	object	No	Parameters to pass
as	string	No	Namespace for variables
timeout	number	No	Max execution time in ms

Data Steps

data.first

Get first item from a list.

- type: data.first
  input: "{{items}}"
  as: first_item
  separator: ", "

Parameter	Type	Required	Description
input	string	Yes	Comma-separated list
as	string	Yes	Variable for first item
separator	string	No	List separator (default: ", ")

data.get

Get an item from a list by index.

- type: data.get
  input: "{{items}}"
  index: "2"
  as: third_item
  separator: ", "

Parameter	Type	Required	Description
input	string	Yes	Delimited list
index	string	Yes	Zero-based index (supports interpolation)
as	string	Yes	Variable for result
separator	string	No	List separator (default: ", ")

variable.set

Set a variable value directly.

- type: variable.set
  name: base_url
  value: "https://example.com"

Parameter	Type	Required	Description
name	string	Yes	Variable name to set
value	string	Yes	Value to assign (supports interpolation)

Issues Steps

Abstract issue tracker steps. Provider is auto-detected from environment variables (Jira, GitHub Issues, etc.).

Requires Provider

Configure an issues provider (e.g., set JIRA_HOST, JIRA_EMAIL, JIRA_API_TOKEN for Jira).

issues.create

Create a new issue.

- type: issues.create
  project: "PROJ"
  summary: "Bug: login fails on mobile"
  description: "Steps to reproduce..."
  issue_type: "Bug"
  labels:
    - mobile
    - critical
  as: issue_key

Parameter	Type	Required	Description
project	string	Yes	Project key
summary	string	Yes	Issue title
description	string	No	Issue body
issue_type	string	No	Issue type (e.g., Bug, Story)
labels	array	No	Labels to apply
provider	string	No	Force specific provider
site	string	No	Provider site/host
as	string	No	Variable for issue key

issues.get

Get issue details.

- type: issues.get
  issue_key: "PROJ-123"
  fields: "summary,status,assignee"
  as: issue

Parameter	Type	Required	Description
issue_key	string	Yes	Issue identifier
fields	string	No	Comma-separated fields to return
provider	string	No	Force specific provider
site	string	No	Provider site/host
as	string	No	Variable for result

issues.search

Search for issues.

- type: issues.search
  query: "project = PROJ AND status = 'To Do'"
  max_results: 10
  as: results

Parameter	Type	Required	Description
query	string	Yes	Search query (JQL for Jira)
max_results	number	No	Max results to return
fields	string	No	Comma-separated fields to return
provider	string	No	Force specific provider
site	string	No	Provider site/host
as	string	No	Variable for results

issues.attach

Attach files to an issue.

- type: issues.attach
  issue_key: "PROJ-123"
  files:
    - filename: "screenshot.png"
      data: "{{screenshot_base64}}"
      content_type: "image/png"
  as: result

Parameter	Type	Required	Description
issue_key	string	Yes	Issue identifier
files	array	Yes	Files to attach: { filename, data, content_type? }
provider	string	No	Force specific provider
site	string	No	Provider site/host
as	string	No	Variable for result

issues.transition

Transition an issue to a new status.

- type: issues.transition
  issue_key: "PROJ-123"
  status: "In Progress"
  as: result

Parameter	Type	Required	Description
issue_key	string	Yes	Issue identifier
status	string	Yes	Target status name
provider	string	No	Force specific provider
site	string	No	Provider site/host
as	string	No	Variable for result

issues.comment

Add a comment to an issue.

- type: issues.comment
  issue_key: "PROJ-123"
  body: "Automated test passed. Ready for review."
  as: comment_id

Parameter	Type	Required	Description
issue_key	string	Yes	Issue identifier
body	string	Yes	Comment text
provider	string	No	Force specific provider
site	string	No	Provider site/host
as	string	No	Variable for comment ID

Git Steps

Abstract git platform steps. Provider is auto-detected from environment (GitHub, etc.).

Requires Provider

Configure a git provider (e.g., set GITHUB_TOKEN for GitHub).

git.listPRs

List pull requests.

- type: git.listPRs
  repo: "org/repo"
  state: "open"
  limit: 10
  as: pull_requests

Parameter	Type	Required	Description
repo	string	No	Repository (default: detected from env)
state	string	No	Filter by state: open, closed, all
limit	number	No	Max results
provider	string	No	Force specific provider
as	string	No	Variable for results

git.getPR

Get pull request details.

- type: git.getPR
  repo: "org/repo"
  number: 42
  as: pr

Parameter	Type	Required	Description
repo	string	No	Repository
number	number	Yes	PR number
provider	string	No	Force specific provider
as	string	No	Variable for result

git.createPR

Create a pull request.

- type: git.createPR
  repo: "org/repo"
  title: "feat: add login flow"
  body: "Implements the new login flow"
  head: "feat/login"
  base: "main"
  as: pr_url

Parameter	Type	Required	Description
repo	string	No	Repository
title	string	Yes	PR title
body	string	No	PR description
head	string	Yes	Source branch
base	string	No	Target branch (default: main)
provider	string	No	Force specific provider
as	string	No	Variable for PR URL

git.listIssues

List repository issues.

- type: git.listIssues
  repo: "org/repo"
  state: "open"
  labels: "bug,critical"
  limit: 20
  as: issues

Parameter	Type	Required	Description
repo	string	No	Repository
state	string	No	Filter by state: open, closed, all
labels	string	No	Comma-separated label filter
limit	number	No	Max results
provider	string	No	Force specific provider
as	string	No	Variable for results

git.getIssue

Get issue details.

- type: git.getIssue
  repo: "org/repo"
  number: 15
  as: issue

Parameter	Type	Required	Description
repo	string	No	Repository
number	number	Yes	Issue number
provider	string	No	Force specific provider
as	string	No	Variable for result

Chat Steps

Not Yet Implemented

Chat steps are registered but not yet functional. They will be available when a chat provider (e.g., Slack) is integrated.

chat.listChannels

List available channels.

- type: chat.listChannels
  limit: 50
  as: channels

Parameter	Type	Required	Description
types	string	No	Channel type filter
limit	number	No	Max results
provider	string	No	Force specific provider
as	string	No	Variable for results

chat.readChannel

Read messages from a channel.

- type: chat.readChannel
  channel: "#general"
  limit: 20
  as: messages

Parameter	Type	Required	Description
channel	string	Yes	Channel name or ID
limit	number	No	Max messages
max_days	number	No	Limit to recent days
provider	string	No	Force specific provider
as	string	No	Variable for messages

chat.readThread

Read a message thread.

- type: chat.readThread
  channel: "#general"
  thread_ts: "1234567890.123456"
  as: thread

Parameter	Type	Required	Description
channel	string	Yes	Channel name or ID
thread_ts	string	Yes	Thread timestamp ID
max_days	number	No	Limit to recent days
provider	string	No	Force specific provider
as	string	No	Variable for messages

Delay Constants

Any delay_ms or ms parameter that accepts a number also accepts a named delay constant. Named constants resolve to a base millisecond value with +/-10% random jitter on each invocation, making automation timing less predictable and more human-like.

Constant	Base	Range
"small"	500ms	450-550ms
"mid"	1000ms	900-1100ms
"large"	5000ms	4500-5500ms

Raw numeric values are used as-is without jitter.

Applies to: browser.wait → ms, control.retry → delay_ms, control.foreach → delay_ms