Editor Workflow

Editor Workflow#

Overview#

EditorWorkflow is an automated workflow that uses an LLM as a supervisor to provide feedback and editing capabilities, reducing the need for human intervention. The EditorWorkflow implements an automated supervision system where a separate LLM (editor) acts as a supervisor to review, provide feedback, and edit the outputs of the main LLM worker. This creates a hierarchical AI system with automated quality control.

Its key features include:

  • Automated Supervision:

    • LLM Supervisor: Uses a separate LLM model as an editor/supervisor

    • Automatic Feedback: Provides automated feedback without human intervention

    • Quality Control: Ensures output quality through automated review

    • Iterative Improvement: Allows the worker LLM to improve based on supervisor feedback

  • Editor Capabilities:

    • Reviews worker LLM outputs for accuracy and completeness

    • Provides constructive feedback for improvement

    • Can edit and modify worker outputs when necessary

    • Supports “LGTM” (Looks Good To Me) approval for satisfactory outputs

Node Flow#

The dataflow among nodes is as follows:

InitPromptNode → LLMNode ↔ EditorNode ↔ TerminalNode

  1. InitPromptNode: Sets up the initial prompt and context for the workflow

  2. LLMNode: Generates responses and commands (the worker LLM)

  3. EditorNode: Reviews and provides feedback on the worker’s output (the supervisor LLM)

  4. TerminalNode: Executes commands and returns results

Use Cases#

  • Automated Development: Fully automated AI agent development with minimal human oversight

  • Quality Assurance: Automated quality control for AI-generated content

  • Scalable Operations: Large-scale operations where human supervision is not feasible

  • Consistent Standards: Maintaining consistent output quality across multiple executions

Configuration#

The EditorWorkflow requires:

  • Worker LLM: The main LLM model for task execution

  • Editor LLM: The supervisor LLM model for review and feedback

  • Supervisor Prompts: Specialized prompts for the editor’s supervision role

Implementation Details#

The EditorNode inherits from UserActionNode but replaces human interaction with LLM supervision:

  • Uses SUPERVISOR_PROMPT and SUPERVISOR_SYSTEM_PROMPT for editor guidance

  • Implements automated feedback generation

  • Supports iterative improvement cycles

  • Maintains the same workflow structure as AutopilotWorkflow but with automated supervision

Contents