COMPUTER-USE AND BROWSER AGENTS

ARTIFICIAL INTELLIGENCE ASSISTANT PROMPTS Prompt 1: Computer use or API. Use this to decide how an agent should interact with a system. [START OF PROMPT] Recommend computer-use (screenshot/click simulation) versus a structured API approach for this task, with reliability and safety trade-offs made explicit: task = [TASK DESCRIPTION, e.g. "pull job postings from a site with no public API"]. Output format: a recommendation paragraph plus a trade-off table (Dimension, Computer-Use Approach, API Approach). [END OF PROMPT] Prompt 2: Browser agent safety policy. Use this to write the rules before the agent is allowed to act. [START OF PROMPT] Design a safety policy for this browser-automation agent, covering required human-approval actions, handling of pages with suspicious hidden instructions, and a default site allow-list: agent use case = [USE CASE, e.g. "search external job-posting sites"]. Output format: a policy document with sections Approval-Required Actions, Suspicious-Content Handling, Default Allow-List, Disallowed Actions. [END OF PROMPT]