浏览器自动化 Agent 任务规划与错误恢复提示词

You are a browser automation architect specializing in AI-driven web agents (e.g., browser-use, Stagehand, Playwright MCP).

I need you to create a robust task execution plan for a browser automation agent.

Task Description

[DESCRIBE YOUR BROWSER AUTOMATION TASK HERE]

Planning Requirements

1. Step-by-Step Action Plan

For each step, specify:

• Action: click / type / navigate / scroll / wait / extract
• Target: CSS selector + ARIA fallback + visual description
• Precondition: What must be true before this step
• Postcondition: How to verify the step succeeded
• Timeout: Maximum wait time (default 10s)

2. Element Location Strategy (Priority Order)

1. data-testid or aria-label (most stable)
2. Semantic role + text content
3. CSS selector with structural context
4. Visual/screenshot-based fallback

3. Error Recovery Matrix

• Element not found: Re-scan DOM, try alt selector (3 retries)
• Page navigation failed: Retry navigation, clear cache (2 retries)
• CAPTCHA detected: Pause + notify human (1 retry)
• Rate limited (HTTP 429): Exponential backoff (5 retries)
• Session expired: Re-authenticate (1 retry)

4. Human-in-the-Loop Checkpoints

Insert pause points for:

• Sensitive actions (payments, deletions, form submissions)
• Ambiguous UI states requiring human judgment
• CAPTCHA or 2FA challenges

5. Output

• Generate the plan as structured JSON
• Include a Mermaid flowchart of the happy path
• List all assumptions and potential failure modes

Generate the complete automation plan now.