chore(deps): bump ai-sdk-provider-claude-code from 3.4.2 to 3.4.4

[dependabot]

Bumps ai-sdk-provider-claude-code from 3.4.2 to 3.4.4.

Release notes

Sourced from ai-sdk-provider-claude-code's releases.

v3.4.4

Fixed

• Interrupted text blocks - Prevent duplicate text-end events when a text block is closed early by a tool-use block, reasoning block, or assistant tool message before its later content_block_stop arrives.

Contributors

Thanks to @​KaranJayakumar for this contribution!

v3.4.3

Added

• thinking setting - First-class ThinkingConfig support (adaptive, enabled with optional budgetTokens, disabled). Takes precedence over the deprecated maxThinkingTokens.
• effort setting - Control how much effort Claude puts into its response (low, medium, high, max).
• promptSuggestions setting - Enable prompt suggestions from the agent after each turn.
• Reasoning content parts - doGenerate now extracts thinking blocks as AI SDK reasoning content parts, and exposes thinkingTraces in providerMetadata for convenience access.
• Re-exported types - ThinkingConfig, ThinkingAdaptive, ThinkingEnabled, ThinkingDisabled are now exported from the package entry point.
• Thinking traces example - Added examples/thinking-traces.ts demonstrating reasoning access in both streaming and non-streaming modes.

Changed

• Bumped @anthropic-ai/claude-agent-sdk to ^0.2.63.
• DRY content-block extraction - Shared isContentBlock guard and filterContentBlocks utility replace duplicated filter logic in extractToolUses, extractToolResults, and extractToolErrors.

Fixed

• Empty-string content blocks - Fixed truthy bug where empty-string text or thinking content was silently dropped.
• Strict thinking schema validation - Thinking union variants now use .strict() to reject invalid extra keys (e.g. { type: 'adaptive', budgetTokens: 1000 }).

Contributors

Thanks to @​mirrir0 for this contribution!

Changelog

Sourced from ai-sdk-provider-claude-code's changelog.

[3.4.4] - 2026-03-10

Fixed

• Interrupted text blocks - Prevent duplicate text-end events when a text block is closed early by a tool-use block, reasoning block, or assistant tool message before its later content_block_stop arrives.

[3.4.3] - 2026-03-02

Added

• thinking setting - First-class ThinkingConfig support (adaptive, enabled with optional budgetTokens, disabled). Takes precedence over the deprecated maxThinkingTokens.
• effort setting - Control how much effort Claude puts into its response (low, medium, high, max).
• promptSuggestions setting - Enable prompt suggestions from the agent after each turn.
• Reasoning content parts - doGenerate now extracts thinking blocks as AI SDK reasoning content parts, and exposes thinkingTraces in providerMetadata for convenience access.
• Re-exported types - ThinkingConfig, ThinkingAdaptive, ThinkingEnabled, ThinkingDisabled are now exported from the package entry point.
• Thinking traces example - Added examples/thinking-traces.ts demonstrating reasoning access in both streaming and non-streaming modes.

Changed

• Bumped @anthropic-ai/claude-agent-sdk to ^0.2.63.
• DRY content-block extraction - Shared isContentBlock guard and filterContentBlocks utility replace duplicated filter logic in extractToolUses, extractToolResults, and extractToolErrors.

Fixed

• Empty-string content blocks - Fixed truthy bug where empty-string text or thinking content was silently dropped.
• Strict thinking schema validation - Thinking union variants now use .strict() to reject invalid extra keys (e.g. { type: 'adaptive', budgetTokens: 1000 }).

Commits

• c50bde3 chore: bump version to 3.4.4 and update changelog (#125)
• 61fb449 fix: avoid duplicate text-end events when text blocks are interrupted (#124)
• e9297e8 docs: update outdated Claude Code SDK references to Claude Agent SDK (#123)
• dd6a1f5 feat: add missing SDK options (effort, thinking, promptSuggestions) (#122)
• 97004c9 chore: fix prettier formatting in check-cli.ts
• bd5c6ac chore: update CLI install instructions and auth commands
• See full diff in compare view

[cubic-dev-ai]

No issues found across 3 files

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.

Architecture diagram

sequenceDiagram
    participant UI as App CLI / Dev
    participant Provider as ai-sdk-provider-claude-code v3.4.4
    participant SDK as AI SDK Core (ai)
    participant AgentSDK as @anthropic-ai/claude-agent-sdk
    participant Extractor as isContentBlock / filterContentBlocks
    participant EventStream as Content Block Event Stream
    
    Note over UI,EventStream: Runtime interaction after bump to 3.4.4
    
    UI->>Provider: doGenerate() with config (thinking, effort, promptSuggestions)
    Provider->>AgentSDK: sendMessage() with new SDK options
    AgentSDK-->>Provider: Response with thinking blocks + text/tool blocks
    
    alt Thinking enabled (adaptive, enabled, or with budgetTokens)
        Provider->>Provider: NEW: Extract thinking blocks as reasoning parts
        Provider->>Extractor: filterContentBlocks('thinking')
        Extractor-->>Provider: Thinking blocks
        Provider-->>UI: content with reasoning[] parts + thinkingTraces in providerMetadata
    else Thinking disabled
        Provider->>Provider: Skip thinking extraction
    end
    
    UI->>Provider: doStream()
    Provider->>AgentSDK: streamMessage()
    AgentSDK-->>EventStream: Content block events (text, tool_use, thinking, reasoning)
    
    Note over EventStream,Provider: CHANGED: Interrupted text block handling
    
    EventStream->>Provider: content_block_start (text)
    Provider->>UI: stream text-delta ...
    EventStream->>Provider: CHANGED: content_block_stop (text)  <-- arrives early? e.g. tool_use or reasoning block interrupts
    alt Interruption scenario (v3.4.4 fix)
        EventStream->>Provider: tool_use or reasoning block starts
        Provider->>UI: text-end event sent
        Note over Provider: Content block now "closed" locally
        EventStream->>Provider: content_block_stop (original text block) <-- delayed arrival
        Provider->>Provider: NEW: Detect that text-end already emitted → skip duplicate
        Provider->>UI: No duplicate text-end
    else Normal flow (no interruption)
        EventStream->>Provider: content_block_stop (text)
        Provider->>UI: text-end event
    end
    
    EventStream->>Provider: content_block_start (tool_use)
    Provider->>Extractor: isContentBlock('tool_use')
    Extractor-->>Provider: yes
    Provider->>UI: tool_use event
    
    loop For each content block
        Provider->>Extractor: filterContentBlocks(type)
        Extractor-->>Provider: Filtered block list
        Provider->>Provider: DRY: Use shared guard for text/thinking/tool_use extraction
    end
    
    UI->>Provider: Validate config (thinking schema)
    Provider->>Provider: NEW: Strict validation rejects { type: 'adaptive', budgetTokens: 1000 }
    alt Invalid thinking config
        Provider-->>UI: Schema validation error
    else Valid config
        Provider->>AgentSDK: Forward validated options
    end

Loading

_{Re-trigger cubic}

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[codacy-production]

Pull Request Overview

This PR updates the 'ai-sdk-provider-claude-code' dependency from version 3.4.2 to 3.4.4. The current Codacy analysis indicates the project is 'up to standards' with no new issues or complexity increases.

However, there is a gap in the provided PR metadata as no file changes were detected, which prevents verification of the dependency manifest update. Additionally, the upgrade path includes a version (3.4.3) that introduces breaking changes and deprecations that may require code migration to prevent runtime failures.

About this PR

• New strict schema validation for thinking configurations has been introduced. Any existing code passing unexpected keys in these configuration objects will now trigger runtime errors and should be audited.
• The code diff for this PR was not provided in the analysis, making it impossible to verify if the package manifest or lockfiles were correctly updated.
• The upgrade path deprecates 'maxThinkingTokens' in favor of a new 'thinking' configuration. Ensure that any usage of the deprecated field in the codebase is migrated to the new format.

Test suggestions

• Verify the dependency version bump in the package manifest file.

Prompt proposal for missing tests

Consider implementing these tests if applicable:
1. Verify the dependency version bump in the package manifest file.

_{TIP Improve review quality by adding custom instructions
TIP How was this review? Give us feedback}

[codacy-production]

Up to standards ✅

🟢 Issues 0 issues

Results:
0 new issues

View in Codacy

AI Reviewer: first review requested successfully. AI can make mistakes. Always validate suggestions.

_{TIP This summary will be updated as you push new changes.}