hermes - 💡(How to fix) Fix [Bug] Mid-thought tool execution splits reasoning into separate "Thinking" blocks

Problem

When the model pauses its reasoning to make a tool call mid-thought, the desktop app splits what was one continuous reasoning sentence into two separate "Thinking" blocks. Visually jarring — you see:

Thinking… (first half)
Thinking… (second half)

instead of one seamless block.

Root cause

appendReasoningPart() in apps/desktop/src/lib/chat-messages.ts only appends to the last reasoning part if it's the absolute last element in the parts array (next.at(-1)?.type === 'reasoning'). Mid-thought tool execution inserts a tool-call part at the end, so the reasoning delta creates a new part instead of extending the existing one.

Fix

Scan backwards from the end of the array to find the last reasoning part regardless of what part types sit between. Tool-call parts between reasoning fragments no longer split the block.

 export function appendReasoningPart(parts: ChatMessagePart[], delta: string): ChatMessagePart[] {
   const next = [...parts]
-  const last = next.at(-1)
-  if (last?.type === 'reasoning') {
-    next[next.length - 1] = { ...last, text: `${last.text}${delta}` }
-    return next
+  for (let i = next.length - 1; i >= 0; i--) {
+    const entry = next[i]
+    if (entry?.type === 'reasoning') {
+      next[i] = { ...entry, text: `${entry.text}${delta}` }
+      return next
+    }
   }
   next.push(reasoningPart(delta))

Workaround

None — this is a UI presentation bug. Reasoning content is preserved in the underlying message, it's just displayed as a fragmented block.