RITSU v0.1 — Agentic Workflow Instruction Syntax
Every Copelf workflow is stored in the RITSU v0.1 format — a structured YAML manifest that describes inputs, steps, and execution behavior. RITSU files are auto-generated from your video recording and fully editable in the visual workflow editor.
ritsu: '0.1'
name: 'Workflow Name'
inputs:
variableName:
type: string
label: 'Display Label'
description: 'Helper text'
required: true
steps:
- id: step-1
name: 'Open page'
action: navigate
with:
url: 'https://example.com'
- id: step-2
name: 'Enter email'
action: fill
mode: auto
with:
target:
description: 'Email input field'
vision:
hint: "Text field with placeholder 'Enter email'"
value: '${email}'Step Types
Copelf supports five action types. Each step in a workflow uses exactly one.
navigate
Opens a URL in the browser.
- id: step-1
action: navigate
name: 'Open registration form'
with:
url: 'https://example.com/form'
waitUntil: load # load | domcontentloaded | networkidleclick
Clicks a UI element identified by description and vision hint.
- id: step-2
action: click
name: 'Click submit button'
mode: auto
approval: false
with:
target:
description: 'Submit button at the bottom of the form'
vision:
hint: "Blue button labeled 'Submit'"fill
Types text into an input field. Supports template variables.
- id: step-3
action: fill
name: 'Enter email'
mode: auto
with:
target:
description: 'Email input field'
vision:
hint: "Text field with placeholder 'Enter email'"
value: '${email}'select
Chooses an option from a dropdown.
- id: step-4
action: select
name: 'Select country'
mode: auto
with:
target:
description: 'Country dropdown'
vision:
hint: "Dropdown labeled 'Country'"
value: '${country}'agent
AI-driven autonomous multi-step execution. The browser agent receives a natural language prompt and operates the browser independently for up to maxSteps iterations.
- id: step-5
action: agent
name: 'Fill complex form'
with:
prompt: 'Fill in all required fields on this registration form'
maxSteps: 10 # 1–100The agent action does not support mode or approval settings — it always
runs autonomously using a more powerful AI model.
Execution Modes
Each step (except navigate and agent) can be configured with a mode:
| Mode | Behavior |
|---|---|
| auto | AI operates the browser using element detection and vision. Fully autonomous. |
| manual | The user operates the browser directly. Copelf waits for you to complete the step. |
| approval | Same as auto, but pauses for human confirmation before the AI proceeds. Set approval: true on an auto-mode step. |
- id: step-6
action: click
name: 'Confirm payment'
mode: auto
approval: true # AI will pause and wait for your approval
with:
target:
description: 'Confirm payment button'
vision:
hint: "Green button labeled 'Pay Now'"Conditional Steps
Any step can include an if expression to run conditionally:
- id: step-7
action: click
name: 'Accept optional terms'
mode: auto
if: 'acceptTerms == true'
with:
target:
description: 'Terms checkbox'
vision:
hint: "Checkbox labeled 'I accept the terms'"Workflow Editor
The visual workflow editor provides a full-featured interface for building and refining workflows.
Editor Features
- Drag-and-drop reordering — Rearrange steps by dragging (powered by dnd-kit)
- Step cards — Each step is a card showing action type, name, and configuration
- Expand/collapse — Click a step card to view or hide its full configuration
- Add and delete steps — Insert new steps or remove existing ones
- Watch original recording — Replay the source video alongside your workflow
- Download YAML — Export the workflow as a RITSU YAML file
- Validation — The editor validates your workflow before saving, catching missing fields and configuration errors
- Execution history — View past runs directly from the editor
- Unsaved changes indicator — A visual marker shows when you have uncommitted edits
After AI generates a workflow from your video, always review the steps in the editor. Adjust target descriptions and vision hints if the AI's interpretation needs refinement.