Ok so just shipped a batch of Claude prompts to someone on here, small thing, like 20 bucks, but it actually sold and I got the notification and I was genuinely happy about it for like an hour. Still am, actually. But then I started thinking about what comes next and I got nervous about something specific.
- The person who bought it is going to use these prompts in production probably, or at least share them with their team, and I realized I never stress-tested them against different Claude versions or different context window sizes.
- I wrote them assuming someone has Claude 4.7 and decent system prompt support, but what if they're running Sonnet 4.6 or they're in some environment where you can't do multi-turn setup? The prompts still work, I think, but they might degrade.
- I did zero documentation on what these prompts are actually good at versus what they're not. Like they work great for database schema generation but they're kind of mediocre at refactoring legacy code, and I didn't say that anywhere.
- The kicker is my clients at the agency do this all the time. They'll ask me to "make an AI tool" in two weeks and then get surprised when prompt engineering takes half that time, and they want to ship before I've even thought about version compatibility or edge cases.
- But here's the thing, when it's my own work I suddenly get it. I cut corners I shouldn't have. The prompts are good but they're not production-ready in the way I'd expect if someone paid me for this at the agency.
I'm going to add documentation and test against a few model versions before I sell another batch. Small thing but it matters.
Token count on the full prompt suite is 8,240.