Hi, just restarted Tinderbox, in workflow with Devonthink, Gemini and Scrivener, for a book project. Can only establish a one way communication from Germini to Tinderbox. Seems Tinderbox requests via ai, system or AppleScript are blocked by MacOS. Doesn’t show up in Setup/Automation.
Any hint on how I can establish a two way communication? Cheers, Eberhard
It sounds like an OS-level permissions issue. But which setting, I can’t guess. I don’t believe there is a listing and the current mac System Preferences app is a real mess of a design. A known general problem is if you miss a permissions approval prompt once, you may never get asked again.
Can you give us a concrete example of what you’re trying to do? I’m having a little trouble following.
Thanks for getting back to me. Here’s an example, a stamp intended for testing:
$MyAnnotation = AppleScript(‘return do shell script “whoami”’);
AppleScript (like ai and system) are marked while in Tinderbox and action doesn’t execute, which seem to mean not recognised (or not applicable). Setup open for Tinderbox Disk Access, don’t know how to add/modify the “Automation” list. Hurts my concept.
Would a clean actual MacOS help? Only permission interrogation in the context of Tinderbox was “Trusted Folder”. Cheers, Eberhard
I think you want to use something like:
$MyAnnotation = runCommand("osascript ...")
And make sure that any spaces in the osascript command line are properly escaped and quoted.
This can’t work as Tinderbox action code has not ‘AppleScript()’ operator. But runCommand() is what you would use for this task.
It may be the reason you are getting nothing in $MyAnnotation is simply that you are not using valid action code. Before testing further, I would encode the task using runCommand() and test that in a small stand-alone file. The latter separates the test from other issue in you main file and proves, or disproves, whether your command line call is returning a value.
Where, if I might ask did you get the suggestion of an AppleScript() command in Action Code. It wasn’t by chance an AI suggestion? Anyway, the most exhaustive list of allowed Action code operators is here: Full Operator List.
What do you mean by ‘marked’? If the point is that the stamp doesn’t return a value, see above as the cause might be as simple as using invalid action code that returns a silent fail.
Before digging into permissions, I’d confirm the task at the centre aof things actually works, as the example given will not!
HTH 
I’m really pleased with the timely response, great support, and the positive result! Was working with Gemini, which was not aware about using “runCommand” 
Thanks a lot and enjoy your day, Eberhard
The current Tinderbox Readings (in /Hints/AI/Gemini/Readings) give only the barest bones of the action language. (And of course you might not have these, or maybe Gemini didn’t read them; that’s worth knowing.)
We don’t tell Gemini a lot about actions because, for plenty of applications, it doesn’t need to know very much. In part, this is because Gemini and Claude have read a lot of software documentation, and they often can guess pretty well. It’s like explaining Tinderbox to a research colleague: they don’t yet know anything about the action language, but from some small examples they might be able to say “oh, sure, this is a little LISP in Smalltalk dress.”
I’m planning to add an optional Readings note on how to use actions. Last weekend, I worked on new note-taking instructions that tell the AI to keep its notes organized so that future AI sessions will know what exists without filling the context with too much unwanted detail. It’s a lot like teaching a student.
Thanks Mark, in the midst of gathering experience. Writing social fiction (implication induced by stuff like weather change and technology) and looking mainly for guidance on writing, as my first try. I’ll explore others in the Eastgate domain to gain a better understanding. Cheers, Eberhard