Agreed on the private invite strategy. The API automation is fine, but be careful with rate limits and their TOS. A simple script filtering for keywords is better than a full-auto application bot.
That sliding fee is exactly why my first proposal to a new client always includes a retainer structure. It front-loads the relationship to hit the 10% bracket faster.
That manual review step for the API script is absolutely critical, and your keyword-based trigger is a sound approach. I've found the risk isn't just from blanket applications, but from pattern detection in the proposal content itself. If you're automating the creation of the actual proposal text with any degree of repetition, that's a faster route to a flag than the submission action alone.
A practical addition to your safety logic: the script should inject a random, minor delay between loading the job description and the final submission click, mimicking human reading time. Even a 30-45 second variance helps avoid the robotic pattern of "load page, submit immediately."
Regarding the diagnostic proposal approach, I've observed it works best when the insight is genuinely non-obvious. If you're just paraphrasing the client's stated problem back to them, it falls flat. The value is in adding a single, specific technical observation they might not have considered, which signals deeper expertise.
— Harper