In multi-turn conversations with kubectl-ai, when confirming execution with replies like "sure" or "yes", the assistant often responds with:

"Of course. I'll start by..."
[and correctly identifies a function to execute, e.g., oc4 logs]

However, the function/tool is not executed, even though the logs show the tool call was generated properly. The conversation ends prematurely with:

No function calls were made, so most likely the task is completed
What happened?
The assistant appears to send a response having both:

A partially completed text block ("Of course. I'll start by checking the logs...")

A valid function call, such as:

{"name":"oc4","arguments":{"command":"logs -n d7regserv enrich-credit-pre-9-lp5qn", "modifies_resource":"no"}}
But due to how kubectl-ai parses the response, it logs:

Warning: there are non-text parts FunctionCall in the response, returning concatenation of all text parts.
The tool is then never executed, even though the LLM fully constructed the right action.

Flow ends with:

text
No function calls were made, so most likely the task is completed

⏱️ Steps to Reproduce
Use kubectl-ai and a tool like oc4 (in tools.yaml).

Ask a query that requires action, e.g.:

Check logs for pod enrich-credit-pre-9-lp5qn in namespace d7regserv
Wait for the assistant to reply:

Would you like me to proceed with any of these steps?
Respond with:

sure
Observe that the assistant begins with "I'll start by..." AND generates a tool call — but never executes it.

🔍 Logs (Redacted):

Would you like me to proceed with any of these steps

**iteration=0
response: Of course. I'll start by
response: checking the logs for the enrich-credit-pre-9-lp5qn pod to investigate why its readiness probe is failing
response: .
function calls=[{"name":"oc4","arguments":{"command":"logs -n d7regserv enrich-credit-pre-9-lp5qn","modifies_resource":"no"}}]

Warning: there are non-text parts FunctionCall in the response, returning concatenation of all text parts.
No function calls were made, so most likely the task is completed, so we're done.**

💡 Expected Behavior:

kubectl-ai should recognize the presence of a valid function_call even when combined with text parts, and execute it accordingly.

Even if the LLM generates both response types simultaneously, the system should:

Execute the tool,

Then append the assistant's natural language reasoning (ex: "I'm checking the logs now...").

Version Info:

kubectl-ai version:
version: 0.0.15
commit: 77936f0
date: 2025-07-02T22:57:51Z

Model: Gemini 2.5 Pro

Custom tool: oc4 defined in tools.yaml

Host Cluster: OpenShift