Day 6 done! My loop checked the comments on my post every 20 minutes. Working with scheduled tasks turned out to be much more interesting and challenging. Initially, I created a scheduled task and discussed what I needed it to do in the chat. Only an hour later did I realize that I had done it in Cowork, while what I actually needed was Routines in Code. I just didn't know I had to look for them. Luckily, I was working in the same project, and Claude was able to read status.md and the scheduled task description, so I didn’t waste any more time. I used the skill I created on Day 3. I gave Claude my work plans, and now every Monday morning it will check what has already been done, see whether there are any overdue tasks, review my work plan for the current week, and start searching for the keywords I need for work. I’m tired but content. 🙂 Just one more day left. Let’s get it done.

I'm finally done with the website! Somehow this task turned out to be more soul-draining than all the previous ones 😅 But still, the site is ready https://idonea-macrame.vercel.app/ There's a lot to do. For now, it’s just a showcase. Payment methods haven’t been set up yet, there’s no shopping cart, and so on. I'm just not ready to actually sell my trinkets and turn my hobby into something more serious. But it’s already a solid starting point for a future fully functional website. I used all the hacks except #4. Hard to choose which one made the biggest difference, but I'd say the screenshot loop.

Another task done! My automation script parsed and saved competitors' sitemaps to the DB. From now on, every single Monday, it’s going to compare the latest sitemap with the previous version and drop a report in my messenger with the number and list of new URLs. Zero crashes, all because Claude and I spent a ton of time planning what to do and how. Then we built and tested it in chunks, so the final launch was flawless. There were only minor struggles, mostly with setting up the bot, navigating the interfaces, and wiring everything together.

Day 3 task is done! Claude and I built a custom skill that will save me hours, if not days, of work. Here is what it does: it takes either a product description or a list of search queries that already drive traffic to the product. Then, the AI uses API to analyze organic snippets, image results and related searches to find out how users actually name this item. Finally, it gathers keyword masks and negative words for further scraping. After some consideration I added 2 modes, Quick mask pull (A, default) and Deep core research (B). The whole thing is somehow just 126 lines. I also created a separate file with company info so the AI knows which queries are relevant and which should be filtered out. After 3 or 4 runs and some fine-tuning, I can now say it’s a fully working internal tool)

Another task complete! I've collected (meaning Claude's collected) а list of preferable apartments for a vacation. Still a lot to go with how the results look, but I'm impressed with the convenience. Claude even gave me a piece of advice which place to choose. I'm supposed to come up with an idea how to use the scraper for work but all I'm thinking now is that I can make Claude extract and analyze all the reviews (4K of them in total 🙈) and give me a short summary on pros and cons for every apartment option