From a conversation to the Script Workspace.

How does AI-assisted FileMaker development work?

Every build starts as a Claude Code project inside my Obsidian vault, with a FileMaker-WebDev skill that holds my script and web standards. Claude turns my requirements into a web app and FileMaker scripts, Python tests check them, and I review everything before converting the scripts to fmxmlsnippet XML, pasting them into FileMaker and verifying the result live.

1. 01
   Claude Code

   Start a Claude Code project

   Every build starts as a new Claude Code project. Its folder becomes the one place where requirements, scripts, web code and tests live.

   Watch this step ▸
2. 02
   Vault

   Give it context through Obsidian

   The project folder sits inside my Obsidian vault. The requirements and schema notes I write in Obsidian are the same files Claude reads, so nothing gets copied between tools.

   Watch this step ▸
3. 03
   Skill

   Load the FileMaker-WebDev skill

   My FileMaker-WebDev skill rides along with every session. It holds the best practices and script step rules I expect, so everything Claude writes follows them.

   Watch this step ▸
4. 04
   Requirements

   Describe my requirements to Claude

   I describe what I need in plain language and point Claude at the requirements in the vault. Claude reads them with the schema notes and lays out a plan before writing anything.

   Watch this step ▸
5. 05
   Generate

   Claude generates the web app and FileMaker scripts

   Claude writes both halves: the HTML, CSS and JavaScript for the web app, and the FileMaker scripts that work with it. Every file lands in the project folder, inside the vault.

   Watch this step ▸
6. 06
   Python tests

   Verify with Python tests

   Right after generation, Claude writes a Python test suite for the new scripts and web code. It runs from the project folder, and any failure goes straight back to Claude as a fix.

   Watch this step ▸
7. 07
   Markdown

   Save the scripts as Markdown

   Once the tests pass, each FileMaker script is saved as plain-text steps in a Markdown file, one step per line, the way the Script Workspace prints them. The files go into the project’s scripts folder in the vault.

   Watch this step ▸
8. 08
   Obsidian

   Review it in Obsidian

   Because the script lands in the vault, Obsidian renders it right away. I read and check it there as a document before it goes anywhere near FileMaker.

   Watch this step ▸
9. 09
   VS Code

   Review the web code in VS Code

   With the scripts reviewed, I open the web app in VS Code on macOS and read its HTML, CSS and JavaScript straight from the project’s web folder.

   Watch this step ▸
10. 10
    Converter

    Convert it to fmxmlsnippet

    I paste the Markdown into my own converter page. It turns each plain-text step into the fmxmlsnippet XML that FileMaker reads, ready to copy to the clipboard.

    Watch this step ▸
11. 11
    FileMaker

    Paste scripts and save the web app in FileMaker

    One paste in the Script Workspace turns the clipboard into real script steps. The web app’s HTML, CSS and JavaScript go into text fields on a tagged record, and the web viewer builds its source from those three by querying the tag.

    Watch this step ▸
12. 12
    Live tests

    Run live tests for verification

    Last, I run everything for real in FileMaker: the script against live data, the web viewer form, and the server schedule. Anything that misses becomes the next request.

    Watch this step ▸

The tools and their order are my real process. The project, script, file, test and result names in the animation are examples. How client data is handled