Skip to content

Input Model

Prerequisites: - Familiarity with Python dict/list literals

What you'll learn: - Supported input shapes - How text and image inputs are normalized

Supported Input Types

Input accepts: - str - list[UserInput]

UserInput variants: - {"type": "text", "text": "..."} - {"type": "local_image", "path": "..."}

Normalization Rules

When input is a list: - Text parts are joined with two newlines - Local image paths are forwarded as repeated --image args

Example:

payload = [
    {"type": "text", "text": "Describe file changes"},
    {"type": "text", "text": "Focus on tests"},
    {"type": "local_image", "path": "./ui.png"},
]

Effective behavior: - Prompt becomes: Describe file changes\n\nFocus on tests - Images list includes: ./ui.png

Best Practices

  • Keep text segments intentional and readable
  • Validate local image paths before calling run
  • Prefer one conceptual task per turn