FileMaker AI Vocabulary — AI skill for Claude Code

FileMaker AI Vocabulary

[](https://creativecommons.org/licenses/by/4.0/)

**Every FileMaker 26 function and script step, in about 7,000 tokens.**

Small enough to load into context as part of the skill, so the model has the whole vocabulary in front of it while it writes, instead of guessing from training data and inventing functions that don't exist. The raw Claris help pages for the functions alone run to roughly 122,000 tokens. This is the same knowledge at roughly 1/30th the token footprint.

Developed by Andrew Kear of Clockwork Creative Technology and shared openly with the FileMaker/Claris community.

The problem it solves

FileMaker's namespace is large and full of near-identical names: `Right`, `RightWords`, `RightValues`; `GetField`, `GetFieldName`. An AI generating a calculation guesses from training data and confidently invents a function that sounds right, or transposes a parameter, or assumes the wrong return type. The fix is to hand it the real vocabulary as ground truth, but a reference only helps if it's loaded at the moment of generation, not fetched afterward. Most references are too big to keep resident. This one isn't. That's the whole idea.

What's inside

Two flat files, one line per entry, no prose to wade through:

• 368 functions — exact name, parameter order, optional-parameter shape, return type.
• 216 script steps — exact name, internal ID, and platform support across all seven FileMaker clients and hosts.

Case ( test1 ; result1 {; test2 ; result2 ; ... ; defaultResult} ) → text, number, date, time, timestamp, container
Substitute ( text ; [search1 ; replace1] {; [search2 ; replace2] ; ...} ) → text
Set Variable   #141
Perform Script On Server   #164   ⚠ Go:No, WebD:No

Return types are carried because models infer them wrongly when chaining calculations, not just function names. It's a constrained vocabulary, not a manual: enough to verify a na