Master Instructions

The always-on system prompt that defines your agent's identity, constraints, and operational blueprint.

What are Master Instructions?

The biggest friction point in using AI is the "blank canvas" problem: having to spend two paragraphs explaining who you are, what your company does, and how you want things formatted every single time you start a new conversation.

In Opulent, workspaces are governed by Master Instructions. This is an overarching, always-on directive that blankets every single task, subagent, and workflow executed within that environment. You configure it once, and the agent obeys it permanently.

Anatomy of a Perfect Master Instruction

A high-performing Master Instruction acts as an employee onboarding document. It should dictate behavior across four pillars:

1. Identity & Role

Tell the agent exactly who it is acting as.

"You are a Senior Systems Architect for Acme Corp. Your role is to review PRs, 
audit cloud infrastructure, and propose scalable backend refactors."

2. Format & Style

Mandate output structures to avoid having to correct the agent's tone repeatedly.

"Never use corporate jargon or marketing speak. Always output documentation 
in Markdown, utilizing exact file paths. Prefer bullet points over long paragraphs."

3. Workflow Constraints

Set guardrails on how the agent is allowed to execute tasks, especially when using tools like the Cloud Browser Engine or Repo Access.

"When reviewing code, never auto-commit directly to the main branch. 
Always create a new branch prefixed with `agent-review/` and open a draft PR."

4. Naming & Tooling Conventions

Inform the agent about internal company standards.

"We use Jira for ticket tracking and Slack for messaging. 
All date formats must be YYYY-MM-DD. Our primary database is PostgreSQL."

Compounding Efficiency

When you combine Master Instructions with the Knowledge Base and the Connected Cortex (Memory), you create an incredibly autonomous environment.

1. The Master Instruction tells the agent how to act.
2. The Knowledge Base provides the facts it needs to act.
3. The Connected Cortex remembers what it has already done.

This trifecta allows you to start complex tasks with incredibly short prompts:

• "Execute a weekly review." The agent already knows (via instructions) what a weekly review entails, knows (via knowledge) what the KPIs are, and knows (via memory) what last week's numbers were.

Global vs. Local Scope

It is highly recommended to keep Master Instructions tightly scoped by creating separate Workspaces for different departments.

• Sales Workspace Instruction: "You are an aggressively helpful SDR focused on lead qualification..."
• Engineering Workspace Instruction: "You are a cautious, security-first Site Reliability Engineer..."

By isolating these personas, you prevent the SDR agent from accidentally utilizing engineering jargon, and vice-versa.