How to build a Control Checker and Writer Agent

Workflow Overview

1. User Input: Control Description

• Node: Control Input (in-0)

• Role: This is where the user provides a control description or an existing control statement they want to check or improve.

• How it works: The user types or pastes their control statement into this input node.

2. Knowledge Base Search: Control Standards

• Node: Google Drive Knowledge Base (knowledgebase-0)

• Role: This node searches a curated set of documents in Google Drive that contain your organization’s control standards, templates, and best practices for writing controls.

• How it works:

  • The input from the user (from in-0) is used as a query to search the knowledge base.

  • The node retrieves the most relevant sections, templates, and examples from your control standards documentation.

3. LLM Analysis & Drafting

• Node: OpenAI (llm-0)

• Role: This is the “brain” of the workflow. It uses a large language model to:

  • Draft a new control description that meets your standards, or

  • Review and suggest improvements to an existing control statement.

• How it works:

  • The LLM receives two main inputs:

    • The user’s control description (in-0)

    • The relevant control standards content from the knowledge base (knowledgebase-0)

  • The LLM is guided by a system prompt to focus on the “Who, What, Where, When, Why, and How” of control design, and to compare the user’s input to the gold standard.

  • The LLM then generates a response that either drafts a new control or provides feedback and suggestions for improvement.

4. Output

• Node: Output (out-0)

• Role: Presents the final drafted or reviewed control statement to the user.

• How it works:

  • The output from the LLM is displayed here for the user to review, copy, or use as needed.

How the Nodes Connect

• User Input (in-0) → feeds into both the Knowledge Base and the LLM.

• Knowledge Base (knowledgebase-0) → provides context and standards to the LLM.

• LLM (llm-0) → processes both the user input and the standards, then sends its output to the Output node.

• Output (out-0) → displays the LLM’s drafted or reviewed control statement.

What This Flow Achieves

• Automates the process of drafting or reviewing internal controls.

• Ensures that all controls are aligned with your organization’s gold standards and best practices.

• Provides actionable, specific, and auditable control statements, reducing ambiguity and improving audit readiness.