Operations like `pods_list` are too verbose and blow the model context window

[dbsanfte]

My clusters have a few hundred microservices and pods_list on such a cluster quickly blows the entire context window. For instance, listing one of my namespaces I end up with 700,000 tokens because this operation seems to be describing every single pod. This much detail is bad for scaling.

pods_list should literally just list the pods, or provide a way to just list the pods, with filters available for extra granularity.

[manusa]

I'm currently working on adding a few options that go in this line.

To summarize:

• --profile flag:
  Adds the option to limit the number of tools depending on the selected profile
  • full: all of the tools are enabled
  • slm: (Small Language Model), only essential tools are enabled (e.g. no Helm)
• Provide a way to configure the output.
  I guess something like what kubectl -o --output does.
  yaml, name, ...
  Exactly what you are requesting.

Please, let me know your thoughts about this.
Will probably have something ready by the end of today ;)

[dbsanfte]

Yes a way to filter the output seems to be exactly what we're missing. I can add filter instructions to my model's instruction prompt to make sure it always uses the filters to avoid a big dump of spam :)

[eranco74]

+1 for this issue.
I belive #87 should help filter the pods.
Regarding the output formatting the standard way of doing it in kubectl is: -o custom-columns or -o jsonpath
Although this server doesn't use kubectl it shouldn't be to complicated to filter the results with a provided jsonpath

[manusa]

The behavior has changed to list resources as a plain-text, human-readable table by default.
The previous behavior can be restored by passing the appropriate configuration (--list-output yaml)

This change is effective starting on v0.0.40

Please, share any feedback you may have, it is much appreciated 🙏