[FEATURE]

[syedaun-techsur]

Feature Area

Core functionality

Is your feature request related to a an existing bug? Please link it here.

N/A

Describe the solution you'd like

Enable CrewAI agents to use Claude's extended thinking feature seamlessly with tool use. Currently, when thinking is enabled on an agent's LLM, subsequent tool-result messages fail because CrewAI's message handling doesn't preserve thinking blocks as required by the Anthropic API.

Desired behavior:

1. When an agent receives a response with thinking blocks
   enabled, extract and preserve those thinking blocks
2. When sending tool results back to the LLM in follow-up
   messages, re-include the thinking blocks from the previous
   assistant message
3. Maintain thinking block continuity across the entire
   tool-use loop until the agent reaches a final answer
4. Ensure the thinking parameter is managed consistently -
   if thinking is enabled for an agent, all messages in a
   conversation turn should include thinking blocks

Expected workflow:

Turn 1: User message → Agent LLM call with thinking=enabled
Response: [thinking block] + [tool_use block]
↓
Turn 2: Include thinking block from Turn 1 + tool result →
Agent LLM call with thinking=enabled
Response: [text block] (final answer)

Implementation should follow Anthropic's specification:
According to https://docs.anthropic.com/claude/docs/build-wi
th-claude/extended-thinking, when using extended thinking
with tool use:

• Tool use loops are part of a single assistant turn
• Thinking blocks must be preserved and passed back to the
  API
• The complete unmodified thinking block must be included in
  follow-up messages to maintain reasoning continuity

This would enable:

• Agents to use Claude's extended thinking with any tools
• Deeper reasoning during tool selection and result
  processing
• Better quality outputs for complex code generation
  tasks
• Full compatibility with Anthropic's native thinking
  capabilities

Describe alternatives you've considered

No response

Additional context

Willingness to Contribute

Yes, I'd be happy to submit a pull request