feat(deepseek): implement interleaved thinking mode for deepseek-reasoner

[hannesrudolph]

Summary

Implements proper interleaved thinking support for DeepSeek V3's thinking mode (deepseek-reasoner model) following the DeepSeek API documentation at https://api-docs.deepseek.com/guides/thinking_mode

Changes

• Add thinking: { type: "enabled" } parameter for deepseek-reasoner model
• Handle streaming reasoning_content from DeepSeek API that yields reasoning chunks
• Add tool call conversion (Anthropic tool_use → OpenAI tool_calls format) for thinking mode
• Add tool result conversion (Anthropic tool_result → OpenAI tool messages)
• Extract reasoning from content blocks for API continuations (Task stores reasoning as content blocks)
• Add getReasoningContent() method to accumulate and expose reasoning content
• Add comprehensive tests for interleaved thinking mode (6 new tests in deepseek.spec.ts)
• Add tests for tool call support (7 new tests in r1-format.spec.ts)

How It Works

Per DeepSeek's API documentation:

1. Thinking mode enabled via thinking: { type: "enabled" } for deepseek-reasoner model
2. During streaming: reasoning_content is yielded as reasoning chunks and accumulated
3. Tool call flow: The model can think → call tools → receive results → continue thinking
4. Within a turn: reasoning_content is preserved and passed back to the API for continuation
5. Between turns: Reasoning content is cleared (new user question starts fresh)

---

Important

Implements interleaved thinking mode for DeepSeek reasoner model, handling reasoning content and tool calls, with comprehensive tests added.

• Behavior:
  • Implements interleaved thinking mode for deepseek-reasoner model with thinking: { type: "enabled" } parameter.
  • Handles streaming reasoning_content and tool calls in DeepSeekHandler.
  • Resets reasoning content between turns and accumulates it within a turn.
• Functions:
  • Adds getReasoningContent() in DeepSeekHandler to expose accumulated reasoning content.
  • Updates convertToR1Format() to handle reasoning_content and tool calls.
• Tests:
  • Adds 6 tests for interleaved thinking mode in deepseek.spec.ts.
  • Adds 7 tests for tool call support in r1-format.spec.ts.

^{This description was created by for f03a5b6. You can customize this summary. It will automatically update as commits are pushed.}

[roomote]

Oroocle   Follow along on Roo Cloud

Re-review complete. No new issues were identified for the latest commit.

• No review issues identified in this pass

Previous reviews

\n> - 09cdadf: [Review #1](https://app.roocode.com/cloud-jobs/tEr8Cpoh?utm_source=cloud-job-url&utm_medium=link&utm_campaign=previous-reviews) - f03a5b6: [Review #2](https://app.roocode.com/cloud-jobs/PjSjBCct?utm_source=cloud-job-url&utm_medium=link&utm_campaign=previous-reviews) - 08a03fe: [Review #3](https://app.roocode.com/cloud-jobs/nQZ63af4?utm_source=cloud-job-url&utm_medium=link&utm_campaign=previous-reviews) - cdad3e7: [Review #4](https://app.roocode.com/cloud-jobs/CEgPN06m?utm_source=cloud-job-url&utm_medium=link&utm_campaign=previous-reviews)

_{Mention @roomote in a comment to request specific changes to this pull request or fix all unresolved issues.}

[bozoweed]

so needed too for kimi-k2-thinking

here is some more details on that tech

# Your tool implementation
def get_weather(city: str) -> dict:
    return {"weather": "Sunny"}

# Tool schema definition
tools = [{
    "type": "function",
    "function": {
        "name": "get_weather",
        "description": "Retrieve current weather information. Call this when the user asks about the weather.",
        "parameters": {
            "type": "object",
            "required": ["city"],
            "properties": {
                "city": {
                    "type": "string",
                    "description": "Name of the city"
                }
            }
        }
    }
}]

# Map tool names to their implementations
tool_map = {
    "get_weather": get_weather
}

def tool_call_with_client(client: OpenAI, model_name: str):
    messages = [
        {"role": "system", "content": "You are Kimi, an AI assistant created by Moonshot AI."},
        {"role": "user", "content": "What's the weather like in Beijing today? Use the tool to check."}
    ]
    finish_reason = None
    while finish_reason is None or finish_reason == "tool_calls":
        completion = client.chat.completions.create(
            model=model_name,
            messages=messages,
            temperature=1.0,
            tools=tools,          # tool list defined above
            tool_choice="auto"
        )
        choice = completion.choices[0]
        finish_reason = choice.finish_reason
        if finish_reason == "tool_calls":
            messages.append(choice.message)
            for tool_call in choice.message.tool_calls:
                tool_call_name = tool_call.function.name
                tool_call_arguments = json.loads(tool_call.function.arguments)
                tool_function = tool_map[tool_call_name]
                tool_result = tool_function(**tool_call_arguments)
                print("tool_result:", tool_result)
                
                messages.append({
                    "role": "tool",
                    "tool_call_id": tool_call.id,
                    "name": tool_call_name,
                    "content": json.dumps(tool_result)
                })
    print("-" * 100)
    print(choice.message.content)