1.5: Fix tool results — use role:tool with tool_call_id

Previously tool results were injected as role:user messages which
confuses some models. Now the live agent loop uses proper OpenAI
function-calling format:
- Assistant messages include tool_calls array with IDs
- Tool results use role:tool with matching tool_call_id

History replay in router.py is unchanged (no tool_call_ids in DB).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

cheddahbot/agent.py

@@ -185,12 +185,23 @@ class Agent:
 
             # Execute tools
             if self._tools:
-                messages.append(
+                # Build OpenAI-format assistant message with tool_calls
+                openai_tool_calls = [
                     {
-                        "role": "assistant",
-                        "content": full_response or "I'll use some tools to help with that.",
+                        "id": tc.get("id", f"call_{tc['name']}_{i}"),
+                        "type": "function",
+                        "function": {
+                            "name": tc["name"],
+                            "arguments": json.dumps(tc.get("input", {})),
+                        },
                     }
-                )
+                    for i, tc in enumerate(unique_tool_calls)
+                ]
+                messages.append({
+                    "role": "assistant",
+                    "content": full_response or None,
+                    "tool_calls": openai_tool_calls,
+                })
 
                 for tc in unique_tool_calls:
                     yield f"\n\n**Using tool: {tc['name']}**\n"
@@ -201,9 +212,11 @@ class Agent:
                     yield f"```\n{result[:2000]}\n```\n\n"
 
                     self.db.add_message(conv_id, "tool", result, tool_result=tc["name"])
-                    messages.append(
-                        {"role": "user", "content": f'[Tool "{tc["name"]}" result]\n{result}'}
-                    )
+                    messages.append({
+                        "role": "tool",
+                        "tool_call_id": tc.get("id", f"call_{tc['name']}"),
+                        "content": result,
+                    })
             else:
                 # No tool system configured - just mention tool was requested
                 if full_response: