genaiscript is still new to me but I still find it hard to grasp a bit

[alyahmedaly]

I will update this discussion with stuff I struggle so maybe it's miss understanding from me or something to improve in docs

examples of lack of understanding

• using run runPrompt returned tool call
  • I expected the tool call will be executed but it didn't this might be it's inline prompt and I will make use of the output my self but maybe we can add flag to handle tool calls or an example of how can I pass the output to be handled.
• while having a script with only runPrompt and nothing else I got this error "No LLM provider configured (see https://microsoft.github.io/genaiscript/getting-started/configuration/)" even while providing a model to runPrompt config
  • I had to add empty script above with only model config there
• so confused between agent-fs and fs tools
• good to know that if I used fs in tools that will ref all other tools start with fs
• fallback tools property under script not available but if I used command line script it works
  • not working script({ fallbackTools: true })
  • --fallback-tools works
• installation and run
  • I understand why you need to use npx genaiscript run but most of the time it's so slow also not sure if it will install latest if it's available anyway what I found works better for me is just install the package and use npx --offline genaiscript run
• API feedback
  • I'm typescript user it's super nice to have the typescript declarations to look at anytime but I think it will be great if you have namespace to group all the exported functions so I don't need to go to docs more often and use typescript auto complete more, something like ai.$ ai.script

[pelikhan]

• fallback tools fallbackTools not configuration from script #1147

installation and run
I understand why you need to use npx genaiscript run but most of the time it's so slow also not sure if it will install latest if it's available anyway what I found works better for me is just install the package and use npx --offline genaiscript run

Yes you're right. We overuse npx and we should teach users to add genaiscript to their package.json in order to run locally (running global tools on windows is somewhat not always working). I'll patch the docs.

good to know that if I used fs in tools that will ref all other tools start with fs

Need to add this to the docs. Might be a terrible idea but it allows to organize tools as a tree and easily match them.

while having a script with only runPrompt and nothing else I got this error "No LLM provider configured (see https://microsoft.github.io/genaiscript/getting-started/configuration/)" even while providing a model to runPrompt config
I had to add empty script above with only model config there

Do you still have this example by any chance?

[pelikhan]

• nage so many repos for my compony, and I wanted to have like a tool box for me is there now I can put this shared scripts somewhere and run it from terminal on any project, which is great but maybe I can use vscode integration to run it which will enable me to use copilot model, better that my ollama:qwen 😄

This might help you. https://microsoft.github.io/genaiscript/reference/configuration-files/

[pelikhan]

There is no dry run but you could use any local LLM to simulate.

local llm can be slow but I got your point :)

adding not valid model name can act like a dry run it show the system message and user message but not tools but good anyway

#1153

[pelikhan]

Try the "echo" model in 104.6

[alyahmedaly]

Thanks so much the new echo mode DX is better that what I expected, you are smart 😄
I thought you will add new cli flag and the flag will just output early, but using the preview is so smart and now with echo model I can combine real model with echo to debug

[pelikhan]

I should have had this echo mode earlier. Now thinking of "none" as well for script that don't expect to run top level LLM calls.

[pelikhan]

#1154

[pelikhan]

using run runPrompt returned tool call
I expected the tool call will be executed but it didn't this might be it's inline prompt and I will make use of the output my self but maybe we can add flag to handle tool calls or an example of how can I pass the output to be handled.

There was a bug in the fallback_tools. Please try again.

[craigdallimore]

I agree with the need for typescript definitions - it would be great to have typedefs for $ and so on.

[craigdallimore]

I write the scripts in a different text editor, and execute them using npx genaiscript run myscript.

I presume vscode with the extension sorts out the types? :)

[craigdallimore]

Oh, thank you for mentioning genaiscript scripts fix
That works perfectly; my editor is now providing completions:

[craigdallimore]

Ah, it is documented, I just didn't find it. My apologies

https://microsoft.github.io/genaiscript/reference/cli/commands/#scripts-fix

[pelikhan]

It's not well documented since I mostly assumed vscode usage. The cli "run" command should probably check for this. Will add docs.

[pelikhan]

Added more discoverable docs.

https://microsoft.github.io/genaiscript/getting-started/running-scripts/#command-line