feat(agents-extensions): #628 add Anthropic extended thinking support

[AryanBagade]

Summary

Fixes #628
This PR adds support for capturing Claude's extended thinking (reasoning) blocks when using the AI SDK adapter with Anthropic models.

Changes

• Non-streaming: Extract reasoning parts from AI SDK responses and output them as ReasoningItem objects
• Streaming: Handle reasoning-start, reasoning-delta, reasoning-end events to accumulate thinking content
• Signature preservation: Store Anthropic's thinking signature in providerData for multi-turn conversation support
• Ordering: Reasoning items are output FIRST, before tool calls (required by Anthropic API)

Test plan

• Added 6 comprehensive unit tests covering all scenarios
• All 671 tests pass
• Verified with real Anthropic API call - signature captured successfully. Tested with my own APIs

References for this Issue

• openai/openai-agents-python#1744
• Anthropic's documentation
• vercel/ai/issues/7729

[changeset-bot]

🦋 Changeset detected

Latest commit: d82623f

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
@openai/agents-extensions	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Preserve reasoning metadata even when no text streamed

In the streamed path we only emit a reasoning item when reasoningBlock.text is truthy. For streams where Anthropic returns a reasoning signature but redacts the thinking text (no deltas before the reasoning-end event), this branch is skipped and the reasoning item is omitted, so the signature in providerMetadata is dropped and cannot be echoed back on the next turn despite the PR’s goal of preserving it. Consider emitting a reasoning item even when the text is empty so providerData survives redacted thinking streams.

Useful? React with 👍 / 👎.

[AryanBagade]

let me update a code to preserve the signature even with empty text.

[seratch]

@codex review again

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. What shall we delve into next?

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[seratch]

I've confirmed this change enables ai-sdk layer to work with the thinking mode. Example model settings:

  const modelSettings = {
    providerData: {
      providerOptions: {
        anthropic: {
          thinking: { type: 'enabled', budgetTokens: 32000 },
        },
      },
    },
  };