kingabzpro/Multi-Agent-Research-Assistant — reverse-engineered prompt

Build me a simple browser app for doing research reports. I want to type in a research question, click a button, and watch live progress as a few AI agents work through it.

The app should use a manager agent to coordinate everything, a judge agent to decide if the evidence is strong enough, and an analyst agent to write the final report. It should start with a quick answer from Olostep, then if the judge thinks it needs more proof, it should search and scrape more pages, pick the best sources, and use those to make a polished Markdown report with citations or source links.

Show the report nicely in the browser with headings, bullets, tables, code blocks, and clickable links. Add buttons to stop, reset, and download the finished report as a formatted PDF. Use Python, Reflex, OpenAI Agents SDK, and Olostep. Include setup with environment variables for the OpenAI and Olostep keys.