Would it be possible to upload a compressed DDR to the tool?

[jasonnmark]

I have a script which takes a 150MB DDR and extras the usable data from the XML file and turns it into a collection of 10 text files totaling 1.2MB. This lets me import my entire DDR into Claude so it can help me track down complex problems. Of course I have to re-export my XML file every so often, but it's very powerful and would be great if something like this could be integrated into agentic-fm.

Here's the code:

https://github.com/jasonnmark/FilemakerDDRCompressor

Full Disclaimer, I don't know Python, Claude wrote this.

Also note, I'm dyslexic, so I use a lot of Emoji's when naming things to make it easier to trace code. Since XML often breaks Emojis I have a library that's specific to me that tries to fix it which probably wouldn't apply to anyone else in the world... :)

Please let me know if you have questions, but it would be awesome if something like this was built right into agentic-fm so that there was a button to upload your XML. I'm happy to help any way I can.

[petrowsky]

Sounds great. Let's cover what the intent and goals are. Are you trying to track down issues? Get a wider view of the scope of a process? Gain overall insight of the whole solution?

Right now agentic-fm is working off the Save As XML (SaXML) not the DDR. The DDR, being the older format, can't be automated like the SaXML. Therefore, we likely adapt your DDR version to work with the SaXML version. We just need to have a good outline and then we can create both a skill and any pre/post processing scripts in order to make it happen.

I'll do a preliminary pass given just your repository, but if you can use natural language to explain what you're trying to ultimately accomplish I can use that as part of the prompt to add it into agentic-fm.

[jasonnmark]

Oh wow. Yeah I should be using SaXML. I remember trying it when it was first introduced and there was some bug that blocked me so I went back to DDR and forgot to revisit, thanks for the nudge. That seems to handle my emoji's fine, so if for no other reason.

Oh great questions:

Are you trying to track down issues? Get a wider view of the scope of a process? Gain overall insight of the whole solution?

Yes!

Right now the main things I use it for are:

• Is field ABC actually still use anywhere (excluding commented out code)?
• Is Script ABC called from any layouts or other scripts (again excluding commented out code)?
• Are there any SQL statements that call missing tables or fields?
• I want to rename table XYZ and field ABC, what SQL statements will this break?

But I would really LIKE to have it help me track down issues, such as:

• This script hangs sometimes at overnight, can you recommend some debugging checkpoints to put in, or do you see anything that could cause problems
• I want to do away with the duplicate tables ABC and XYZ and just use XYZ since ABC is from a legacy import. Can you please tell me all plans that ABC is references, and what relationships need to be converted to XYZ?
• Loading layout ABC or script XYZ is very slow... can you please look at fields and scripts and layout objects and help find likely spots that slow it down?

With Claude I just upload the condensed 1.2MB files for context and it rocks right through it... it would be great if this was embedded right in the agent if possible.

[petrowsky]

Ok, I just added a new trace skill. You can read the details about it here.

https://github.com/petrowsky/agentic-fm/blob/main/.cursor/skills/trace/SKILL.md

I have not tested it at all. It would be best if you can start to test it since you are trying to solve this problem.

It does a deterministic extraction from a parse SaXML file first and then uses an agent with heuristics in order to find any ambiguous connections (or at least it is supposed to).

Use Claude to test and enhance the skill and let me know how it works out.

Do a git pull on the repository to pull down all the latest stuff.

[jasonnmark]

This is awesome. It might take me a couple of days to get to this, but PLEASE don't take that as a sign of lack of interest or appreciation... :)

[jasonnmark]

OK I appreciate this but it broke me. I think I got it partially working but I just don't understand what I'm supposed to do with it, or if it's even working. I ran the compressor and it generated the files here: /Users//GitHub/agentic-fm/agent/context/HPS/04_LAYOUTS.txt Most of them seem to work although Custom Functions are blank. I tried running the trace scripts and got errors:

(base) jason@M3 agentic-fm % python3 agent/scripts/trace.py build  -s "HPS"    
==> Building xref.index for: HPS
  Fields: 0, TOs: 0, Scripts: 0, CFs: 0
  Parsing field calculations...
    0 references found
  Parsing relationships...
    0 references found
  Parsing scripts...
    0 references found
  Parsing layout summaries...
    0 references found
  Parsing custom functions...
    0 references found
  Parsing value lists...
    0 references found

==> Done! 0 total references
  Output: /Users/jason/GitHub/agentic-fm/agent/context/HPS/xref.index

I'm also unclear how I'm supposed to use it. I have very limited experience in terminal and mostly just work in Filemaker directly when I'm coding so I'm unclear if trace sets it up so that the Agent in Filemaker "knows more stuff" or if it has to get reloaded to know it, or if I need to use terminal to ask questions, or what's supposed to happen.

I'd love to dig into it further but I don't want to be a drain on your time/energy if my needs are too ... pedestrian (if that's the right word) .... to be worth your time.

[petrowsky]

I need to take it for a test run myself. I pushed it to the repo quickly so you could both test with it and evolve or improve it if you were familiar with doing pull requests. If you've not spent a lot of time on the command line then it's no worry. I'll test it out here in a day or so. I've been working on other improvements for speeding things up.

One thing I do notice, and it's not a big thing, but your command line prompt is showing (base) which means you have miniconda installed. Not likely an issue given that the version of python being used is probably higher than the one mac OS has installed at /usr/bin/python3.

I'll report back after I play with the skill.

[petrowsky]

Oh, and just to let you know, you should not have to call any of the tools directly. The agent will do this for you. So calling python3 agent/scripts/trace.py build is only ever needed for debugging and testing purposes. I'll shoot a short video about how I've been testing it.

[jasonnmark]

Yeah a video would be super helpful to give me context. I'm not sure if I'm trying to do things I should be trying to do... lol.

[jasonnmark]

And I might try to tinker with it and push and pull things... I'm not familiar with it, but hey 53 years ago I wasn't familiar with brushing y own teeth, but I learned somewhere along the way... 🤣