describe · wangzan101, tools — Claude Code prompt: what it does and how to install
The describe Claude Code prompt asks a vision model to explain the full screen. State your next task to get page details, control coordinates, obstacles, and task hints.
Apache-2.0 ★ 9,488 updated 2026-11-16
.claude/commands/describe.mdWhat describe does
The prompt has a vision model inspect the whole screen when the element list does not reveal what the page is or how its layout works. The response describes the page and its usable controls, gives each control’s centre coordinates on a 0–999 scale, and points out obstacles and task-related hints.
For a single field, icon, or control, use {{find_visually}} instead. A full-screen read is slow and often takes over ten seconds. In Claude Code, save the prompt as .claude/commands/describe.md.
What it helps with
- identify an unfamiliar page from its screen
- understand how controls are arranged
- get centre coordinates for usable controls
- spot obstacles before carrying out a task
Example prompts for the agent
- Describe the full screen and list control coordinates; I’m about to fill in the form.
- Explain this page and flag anything that could get in the way of submitting it.
- Read the screen so I can understand its layout before opening settings.
How to install describe
Manually: put the files in place
.claude/commands/describe.mdHow to use
Paste the text at the start of a chat or into project instructions
Questions about describe
How do I install describe in Claude Code?
Create .claude/commands/describe.md and put the prompt in that file.
What does describe do?
It asks a vision model to explain the screen, including the page, usable controls and their coordinates, obstacles, and hints for the task.
When should I use describe instead of {{find_visually}}?
Use describe to get a whole-screen explanation when the element list does not clarify the page or its layout. {{find_visually}} is for locating one field, icon, or control.
Source text
describe.md · Apache-2.0
Ask the vision model to read the whole screen for you, when you cannot tell from the element list what the page is or how it is laid out. Say in task what you are about to do. You get back what the page is, the controls you can use with their 0-999 centre points, anything odd or in the way, and hints for your task. It is slow (often over ten seconds): to find one control, icon or input box, use {{find_visually}} instead.
© wangzan101 · Apache-2.0 · wangzan101/Mika
Similar prompts
All prompts for Browser and scraping →describe by other authors comes first.