The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
docs	Skipped		Feb 6, 2026 8:02am

@cursor review

Greptile Overview

Greptile Summary

Fixed variable resolution in deactivated workflow branches by exposing output schemas for evaluator metrics and agent response formats, even when blocks haven't executed. This enables proper validation and autocomplete for variable references like <agent1.field> throughout the workflow, regardless of execution state. Also added childWorkflowId to workflow block outputs for better traceability.

Key Changes:

• Added getEvaluatorMetricsSchema() to dynamically generate schemas from evaluator block metric configurations
• Added getResponseFormatSchema() to parse and expose agent response format schemas
• Both functions are called in getBlockSchema() before checking execution state, ensuring schemas are available for all blocks
• Added childWorkflowId output field to both workflow and workflow_input blocks

Technical Context:
The collectBlockData() function builds blockOutputSchemas for ALL workflow blocks, not just executed ones. This is critical for the variable resolver to validate references to blocks in deactivated conditional branches. Previously, only inputFormat-based schemas were exposed regardless of execution state. This fix extends that behavior to evaluator metrics and agent response formats.

Confidence Score: 5/5

• This PR is safe to merge with minimal risk
• The changes are well-structured and follow existing patterns in the codebase. The new helper functions mirror the existing getInputFormatFields() approach, with proper validation and error handling. The evaluator metrics logic correctly filters valid metrics and normalizes names to lowercase. The response format parsing uses existing utility functions (parseResponseFormatSafely, extractFieldsFromSchema) that already handle edge cases. The addition of childWorkflowId is a straightforward schema extension. All changes are defensive with proper null checks and early returns.
• No files require special attention

Important Files Changed

Sequence Diagram

sequenceDiagram
    participant Executor as Workflow Executor
    participant BlockData as collectBlockData()
    participant GetSchema as getBlockSchema()
    participant Evaluator as getEvaluatorMetricsSchema()
    participant ResponseFmt as getResponseFormatSchema()
    participant Resolver as Variable Resolver

    Executor->>BlockData: Collect block data and schemas
    
    loop For each block in workflow
        BlockData->>GetSchema: Get output schema for block
        
        alt Block has inputFormat outputs
            GetSchema-->>BlockData: Return merged inputFormat schema
        else Block is evaluator
            GetSchema->>Evaluator: Check evaluator metrics
            Evaluator-->>GetSchema: Return metrics schema (even if not executed)
            GetSchema-->>BlockData: Return evaluator schema
        else Block has responseFormat
            GetSchema->>ResponseFmt: Parse responseFormat param
            ResponseFmt-->>GetSchema: Return parsed schema (even if not executed)
            GetSchema-->>BlockData: Return response format schema
        else Block is trigger or has outputs
            GetSchema-->>BlockData: Return block outputs
        end
    end
    
    BlockData-->>Executor: blockOutputSchemas for all blocks
    
    Note over Resolver: Variable references can now be validated<br/>even for non-executed blocks in deactivated branches