[Bug] Kimi k2.5 API Error: "thinking is enabled but reasoning_content is missing" in tool calls

[spazbg]

I am encountering a blocking 400 error when using the Kimi k2.5 model (Moonshot AI) with Goose on macOS.

The error occurs specifically after the model attempts to use a tool. It appears that when "thinking/reasoning" is enabled on the model, the API requires the reasoning_content field to be preserved in the message history for assistant messages. If Goose strips this field or fails to send it back in the subsequent request, the API rejects the request.

---

To Reproduce
Steps to reproduce the behavior:

1. Configure Goose to use Kimi k2.5 (via Moonshot AI / OpenAI compatible provider).
2. Start a conversation that requires the agent to use a tool (e.g., creating a file or running a terminal command).
3. Allow the model to generate a response that includes both "thinking" (reasoning) and a tool call.
4. Observe the error when the tool output is submitted back to the model.

---

Expected behavior
Goose should preserve the reasoning_content (or equivalent thinking field) from the assistant's previous message when submitting tool outputs back to the API, preventing the validation error.

---

Screenshots
Screenshot

---

Please provide the following information

• OS & Arch: [Macos 15.7.3]
• Interface: [UI]
• Version: [v1.22.2]
• Extensions enabled: [e.g. Computer Controller, Figma]
• Provider & Model: [Moonshot – Kimi K2.5]

[stebbins]

@spazbg cheers, I believe #6962 will resolve

[stebbins]

@spazbg got the branch working -- mind checking it out?

I'm not sure how we can flag for maintainer review in the meantime... 😕