Whatsapp scraper - Beginning

Straight to the purpose: we moved to Spain with zero knowledge of Español, and the building we moved in has a residents chat in Spanish, so I need to get news and some important info in time and in convenient way. So I thought I need to scan it somehow and get the info, then summarize and tranlsate it via LLM.

I started from ChatGPT Atlas, since it has built-in ChatGPT right on your page and kind of should be capable of retrieving anything from the page and act autonomously. Unfortunately, it can't scroll the page and scanned just a part of loaded DOM.

Then I tried Claude Chrome extension which actually can see the page, see the DOM, even see the Console output, with Javascript tool. I asked it to scan the chat, scroll from start of the month to the very end, collect all the info and summarize for me in English. I reached limits 3 times and gave away.

Then I was suggested to ask not to consume the whole chat history, but write a script to do it for me. but for some reason it couldn't produce working script, with Sonnet 4.6 and even couple runs of Opus 4.6.

I was surprised that after getting back to Atlas, the embedded ChatGPT actually managed to write a first working version of the script. Then I made several checks and corrections in that way and started laying out the project architecture (if we can call it that way).

On that stage I need to integrate this "tool" into scheduled agentic flow. I want to to schedule a task once a week, so, for instance, Claude uses Chrome tool to open web Whatsapp page, insert the script and run it with specified dates, then download the json, then ingest it into the chat or claude code/codex session via CLI and render the resulting summary in markdown format.

That's the overall vision, will see how it goes.

To be continued...