FileMaker Second Opinion — AI skill for Claude Code

[](https://github.com/andykear/FileMaker-second-opinion) [](https://creativecommons.org/licenses/by/4.0/)

FileMaker Second Opinion

A reasoning skill for AI-assisted FileMaker development. Built from production experience. Released as v1 after months of refining the logic against real work, now at v1.1.

It contains no techniques and no knowledge. It corrects one bias: an AI's tendency to mistake the common FileMaker solution for the one that holds up in production.

The problem

FileMaker's accessibility is its great strength. It is also the source of a quiet problem for anyone using AI to build in it.

A huge, capable, largely self-taught community means the platform's common patterns are everywhere in the material a model learns from. The deepest patterns mostly are not. The techniques that matter at scale live in client work and production systems that never got written up in public.

So the training data is weighted toward the common solution. And for a specific class of problems, the common solution and the professional one are not the same thing. A model gives you the common one, confidently, because that is what the weight of the data says.

This would not matter if wrong answers announced themselves. Many do. A bad calculation returns the wrong value, a mistyped field name throws an error, you fix it and move on.

The dangerous ones are different. The common answer works in the demo and degrades silently later. Record volume. Concurrent users. WAN latency. A growing found set.

A filtered portal that is fine on two hundred rows and unusable on two hundred thousand. A uniqueness constraint that holds on manual entry and does nothing on a scripted import. An unstored calculation that is instant when you test it alone and a bottleneck once real users are on the hosted file.

The schema looks right