@aadamsx is attempting to deploy a commit to the Sim Team on Vercel.

A member of the Team first needs to authorize it.

Greptile Summary

This PR adds Anthropic's native structured outputs (November 2025 feature) to the Anthropic provider, replacing unreliable prompt-based approaches with constrained decoding for Claude 4.x models. The implementation automatically detects model capabilities and falls back to prompt-based instructions for older Claude 3.x models.

Key Changes:

• Adds STRUCTURED_OUTPUT_MODELS constant for Claude 4.x model detection
• Implements supportsNativeStructuredOutput() for model capability checking
• Adds ensureAdditionalPropertiesFalse() to prepare schemas for the API requirement
• Creates createMessage() and createStreamingMessage() helpers that route requests to beta API or standard API
• Integrates native structured outputs into all request paths (streaming, non-streaming, with/without tools)
• Maintains automatic fallback to prompt-based approach for unsupported models

Issues Found:

1. Model matching logic is overly permissive - uses both .includes() and .startsWith() which could cause false positives (e.g., "claude-3-5-sonnet" would match "claude-sonnet-4-5")
2. Schema recursion doesn't handle oneOf, anyOf, allOf - common JSON Schema keywords won't be processed, risking API validation failures
3. Logger parameter is unnecessarily passed as any instead of using module-scope logger
4. TypeScript safety bypassed with as any cast on beta API calls, preventing compile-time error detection

Confidence Score: 3/5

• This PR is safe to merge with attention to 4 issues: model matching false positives, incomplete schema recursion, and TypeScript safety concerns. The changes are backward compatible but the issues should be addressed to ensure reliability.
• Score of 3/5 reflects a solid implementation with good fallback design, but notable issues that could cause problems in production. The model matching logic could misidentify old models as new ones, and schema recursion gaps mean complex schemas may fail validation. The TypeScript any casts are concerning for long-term maintenance. These are not blocking issues but should be fixed before full confidence.
• The single file apps/sim/providers/anthropic/index.ts requires attention to the 4 identified issues before merging. No other files were changed in this PR.

Important Files Changed

Sequence Diagram

sequenceDiagram
    actor Client
    participant Provider as Anthropic Provider
    participant DetectSupport as supportsNativeStructuredOutput()
    participant CreateMsg as createMessage()
    participant APIStandard as Standard API
    participant APIBeta as Beta API (Structured Output)
    participant Fallback as Fallback Prompt-Based

    Client->>Provider: executeRequest(responseFormat, model)
    Provider->>DetectSupport: supportsNativeStructuredOutput(model)
    
    alt Model supports native structured output
        DetectSupport-->>Provider: true
        Provider->>Provider: ensureAdditionalPropertiesFalse(schema)
        Provider->>CreateMsg: useNativeStructuredOutput=true
        CreateMsg->>APIBeta: create({output_format: json_schema, betas: [structured-outputs-2025-11-13]})
        APIBeta-->>CreateMsg: Anthropic.Message (constrained JSON)
        CreateMsg-->>Provider: response
    else Model does not support native structured output
        DetectSupport-->>Provider: false
        Provider->>Fallback: Build JSON template instructions
        Provider->>CreateMsg: useNativeStructuredOutput=false
        CreateMsg->>APIStandard: create({system: "...JSON schema instructions..."})
        APIStandard-->>CreateMsg: Anthropic.Message (prompt-based)
        CreateMsg-->>Provider: response
    end
    
    Provider-->>Client: ProviderResponse | StreamingExecution

resolved in #2515