How to disable extended thinking on some adapter models?

[ests]

F.i. adpater: anthropic, model: claude-sonet-4

Even if model has capabilities for reasoning, that does not mean I need to use it every time. Currently it goes through reasoning phase even for simple tasks. End when combined with MCP extensions I get this error:

{
	"type": "error",
	"error": {
		"type": "invalid_request_error",
		"message": "messages.1.content.0.type: Expected `thinking` or `redacted_thinking`, but found `text`. When `thinking` is enabled, a final `assistant` message must star t with a thinking block (preceeding the lastmost set of `tool_use` and `tool_result` blocks). We recommend you include thinking blocks from previous turns. To avoid this requirement, disable `thinking`. Please consult our documentat ion at https://docs.anthropic.com/en/docs/build-with-claude/extended-thinking"
	}
}

messages.1.content.0.type: Expected `thinking` or
`redacted_thinking`, but found `text`. When `thinking` is enabled, a final
`assistant` message must star t with a thinking block (preceeding the
lastmost set of `tool_use` and `tool_result` blocks). We recommend you
include thinking blocks from previous turns. To avoid this requirement,
disable `thinking`. Please consult our documentat ion at
https://docs.anthropic.com/en/docs/build-with-claude/extended-thinking

I've looked into https://github.com/olimorris/codecompanion.nvim/blob/6195175810c92d4644149de7b6b1c0758ac6d99c/lua/codecompanion/adapters/anthropic.lua
And tried different options, but no success so far.

[olimorris]

I've raised this as an issue. I encountered it earlier today too. It's due to the refactor in f298bd9. Anthropic are very particular about how the message stack is structured.

Edit: You'll need to explore the schema on the adapter to see how it's configured. This is covered in the docs.

[andtankian]

Any update on this topic?

[ests]

The bug seems to be resolved. In addition, it looks like extended thinking/reasoning (for anthropic specificly) can be disabled this way:

    adapters = {
      anthropic = function()
        return require('codecompanion.adapters').extend('anthropic', {
          schema = {
            extended_thinking = {
              default = false,
            },
          },
        })
      end,
}

But, I haven't found any way that would make this setting easily togglable, and most generic AI clients seem to offer that.

I think it would be really great to have slash command or variable that would turn that on (and it should be off by default).

What is the meaning of live, universe and all that thing?

Please use @{reasoning} to answer.

[olimorris]

You can do gd and manually do it. I'm sure it's not a common activity so shouldn't be too bothersome

[ests]

You can do gd and manually do it. I'm sure it's not a common activity so shouldn't be too bothersome

That indeed worked, thank you. A wee bit cumbersome, but I agree that this is something that would be used on occasion only, and so it is acceptable.