The decision

What can an approved assistant help draft without receiving restricted operational detail?

This workbench creates the instruction set, not the finished security document. It separates the work objective from the site context so a practitioner can decide what information is approved before using an external assistant.

Input

Work product, audience, sensitivity, format, approved context, and constraints.

Output

A structured prompt with explicit evidence, uncertainty, privacy, and review requirements.

Owner

The practitioner who approves the context and reviews the resulting work product.

Before you begin

Prepare four pieces of governance evidence.

  • Approved service: Name the assistant your organization allows for this use.
  • Data boundary: Confirm what is public, approved internal, restricted, or prohibited.
  • Source set: Identify the policy, record, or reference the output must use.
  • Named reviewer: Assign the person responsible for factual, policy, and operational review.

Working tool

Build the governed prompt

Restricted context is withheld from the generated prompt by design.

Select a work product and generate a reusable prompt.

Interpret the output

A strong prompt has eight inspectable parts.

Role and objective

Defines the job without granting decision authority.

Audience and format

Sets length, structure, and level of detail.

Data-handling rule

States what information may and may not be used.

Evidence rule

Requires sources for consequential claims.

Unknowns

Prevents missing facts from becoming invented facts.

Required output

Names the sections the reviewer expects.

Stop conditions

Withholds restricted detail and autonomous decisions.

Human approval

Names the accountable final reviewer.

Method and limits

This tool governs a prompt, not a model.

The generator is deterministic and browser-local. It does not contact an AI service, verify a model, inspect organizational policy, or determine whether a particular dataset is lawful or appropriate to use.

Review the generated prompt against your AI policy, records rules, privacy obligations, collective-bargaining requirements, and incident procedures before use.

Read the complete methodology