By the time a competitor's new product shows up in a monthly report, it has been on sale for six weeks and you have missed the window where a response is cheap.
Catching launches in week one is a different kind of monitoring, and it is one small change away from the category sweep you may already be running.
The change
Sort by newest instead of by sales. On Shopee that means appending sortBy=ctime to the search URL rather than sortBy=sales. Everything else about the job is identical.
Run it weekly. Keep the previous week's item id list. Each week, the answer you want is the difference: listings that exist now and did not exist then, grouped by shop.
The signal inside the signal
Most new listings are noise — variants, relisted stock, seasonal SKUs. One pattern is worth stopping for: a listing that is new this week and has already taken meaningful volume.
Organic discovery does not move that fast. A product with real sold counts within days of appearing almost always has paid support behind it — ad spend, a creator campaign, or a platform promotion. That tells you a competitor is not testing, they are launching, and they have committed budget.
That is the difference between "they added a SKU" and "they are coming after this category", and it is visible in week one only if you are looking weekly.
What to record
For each new listing: the shop, the price, and any units already recorded. Group by shop so a coordinated multi-SKU launch reads as one event rather than six unrelated rows — a competitor adding one product is routine, a competitor adding nine in a week is a range launch.
Keep the history. The most useful version of this report is not this week's new listings but the four-week view: what launched, what took volume, and what quietly disappeared. The disappearances are the half nobody tracks and they tell you what failed, which is often more useful than what worked.
Practical notes
Six credits per URL, up to 50 per request. Weekly across twenty keyword variants is a few cents a month.
The endpoint is approval-gated and runs 08:00–18:00 GMT+8, so schedule the submission inside working hours. That is a real constraint on the design and it is better to know it before building the job than after.
One refinement worth adding once it is running: cross-reference new listings against the panel a month later. A launch that took volume in week one and holds it in month two is a genuine competitor. One that spiked and faded was a promotion. You cannot tell which from the launch data alone.
Recipe 2.5 in the agent skill is the prompt, including the week-over-week diff.