---
name: gui-task
description: >
  Start an AutoGLM GUI automation task. Use when the user wants to automate
  a browser workflow: filling forms, searching, extracting data, navigating
  sites, or completing multi-step web tasks. Invoke with /gui-task followed
  by the goal description.
---

# GUI Task Skill

You have been invoked to start an AutoGLM GUI automation task.

## What to do

1. Extract the user's goal from the invocation arguments
2. Confirm your understanding of the goal with one sentence
3. Immediately delegate to the `task-planner` subagent with the full goal

## Invocation

Pass the complete goal to task-planner, including:
- The target website or service (if specified)
- The desired outcome
- Any constraints the user mentioned (e.g. "don't spend money", "read-only")

## Example delegation prompt to task-planner

```
Automate the following task:
GOAL: [user's goal]
CONSTRAINTS: [any user-specified constraints, or "none"]
CONTEXT: [any relevant details from the conversation]

Decompose this into steps, validate each for safety, execute via web-navigator,
handle any blockers with human-checkpoint, and produce a final report.
```

## Before delegating, remind the user

"I'll start the automation. You can pause or stop at any time. I'll ask for
your input if I encounter a login, payment, or other sensitive step."