[BUG] Tool calls during model thinking are not displayed to users

[clareliguori]

Checks

• I have updated to the lastest minor and patch version of Strands
• I have checked the documentation and this is not expected behavior
• I have searched ./issues and there are no duplicates of my issue

Strands Version

1.23.0

Python Version

3.12.3

Operating System

Ubuntu

Installation Method

pip

Description

When using models with extended thinking/reasoning capabilities (I'm using qwen.qwen3-next-80b-a3b), some tool calls are executed during the model's internal reasoning phase before being announced in the streaming response. These "silent" tool calls are tracked in the agent's message history but are never displayed to users via the
PrintingCallbackHandler, creating confusion about which tools were actually executed.

Expected Behavior

All tool executions should be visible to users through the callback handler, either:

1. By emitting ToolUseStreamEvent with current_tool_use for all tool calls, OR
2. By providing a configuration option to show/hide reasoning-phase tool calls, OR
3. By emitting a different event type for reasoning-phase tools that can be optionally displayed

Actual Behavior

Tools executed during the model's reasoning phase:

• ✅ Are executed successfully
• ✅ Appear in agent.messages
• ❌ Never trigger callbacks with current_tool_use
• ❌ Are never printed as Tool #X: tool_name
• ❌ Are invisible to users watching the console output

Reproduction

Setup:

from strands import Agent
from strands.models import BedrockModel

@tool
def get_user_info() -> dict:
    """Get user information."""
    return {"name": "Alice", "card": "LIB-456789"}

@tool
def get_book_status(book_id: str) -> dict:
    """Get book status."""
    return {"book_id": book_id, "status": "ACTIVE"}

agent = Agent(
    model=BedrockModel(model_id="qwen.qwen3-next-80b-a3b"),
    tools=[get_user_info, get_book_status],
    system_prompt="You are a helpful assistant."
)

result = agent("Please check the status of book BOOK-123")

Observed Output:

Tool #1: get_book_status

Actual Tool Execution (from logs):

2026-01-22 15:55:45 - TOOL EXECUTION START: get_user_info
2026-01-22 15:55:46 - Callback with tool: get_book_status
Tool #1: get_book_status
2026-01-22 15:55:46 - TOOL EXECUTION START: get_book_status

Timeline:

1. get_user_info executes (no callback, not printed)
2. Model streams "I'm using get_book_status" (callback fires, printed as Tool #1)
3. get_book_status executes

Root Cause

From investigation in strands/tools/executors/_executor.py and strands/handlers/callback_handler.py:

1. The callback handler only prints tools when it receives current_tool_use in the event
2. current_tool_use is populated from ContentBlockStartEvent during model streaming (in strands/event_loop/streaming.py)
3. Tools executed during reasoning phase don't generate ContentBlockStartEvent because the model hasn't announced them in its streaming response yet
4. All tools return ToolResultEvent directly, but only tools announced in the model's streaming response get wrapped with current_tool_use context

Suggested Solutions

Option 1: Always emit tool use events
Before executing any tool, emit a ToolUseStreamEvent or similar event that triggers the callback with current_tool_use, regardless of whether the model announced it in streaming.

Option 2: Configuration flag
Add a flag to PrintingCallbackHandler or Agent:

agent = Agent(
    ...,
    callback_handler=PrintingCallbackHandler(
        verbose_tool_use=True,
        show_reasoning_tools=True  # NEW FLAG
    )
)

Option 3: Different event type
Emit a different event type (e.g., ReasoningToolUseEvent) for tools executed during reasoning, allowing developers to handle them differently.