Assistant response text does not match actual write_file tool call content from open-terminal

[amit4422]

Summary

When asking the model to "add 2 random tasks" to a task list, the content written to the file via the write_file tool differs from what the assistant displays in its response. The tool call payload contains one set of tasks, while the rendered chat output shows completely different tasks.

Environment

• App: [Open Terminal](https://github.com/open-webui/open-terminal) (Desktop — "My local terminal")
• Model: Claude opus-4.6 (Anthropic)
• OS: Windows
• App version: latest (as of 2026-03-16)

Steps to Reproduce

1. Have an existing current-tasks.md file with a few checklist items.
2. Prompt: "add 2 random tasks to the list"
3. Observe the server:1/write_file tool call parameters vs. the assistant's displayed response.

Expected Behavior

The tasks shown in the assistant's response text should exactly match the tasks written to the file via the write_file tool call.

Actual Behavior

Tool call payload (write_file parameters → content field):

- [ ] Organize the bookshelf by genre
- [ ] Water the indoor plants

Assistant's displayed response (items 5 & 6, marked "newly added"):

5. ☐ Design a new landing page for the portfolio website (newly added)
6. ☐ Set up automated unit tests for the API endpoints (newly added)

The two sets of tasks are completely different. The file on disk contains the tool-call version; the user sees the response-text version — creating a data mismatch.

[tjbck]

You should be using native function calling mode with open terminal.

[amit4422]

Thanks a lot @tjbck .
I set the native function calling and its working as expected. However, I am still facing some issues. Maybe you could guide.
When I ask to delete an entry it confirms that the entry was deleted, however it is actually not from the file.