Agents Configuration

Prompts & Instructions

Prompts and instructions define how an agent thinks and behaves.

In Sprig, instructions are not casual chat prompts — they are behavioral contracts that shape reliability, tone, and outcomes.

Well-written instructions make the difference between:

A helpful, predictable agent
An inconsistent, hard-to-trust one

Instructions vs Prompts

Sprig separates instructions from runtime prompts.

Type	Purpose
Instructions	Define permanent agent behavior
Prompts	Provide situational input at runtime

This separation keeps agents stable while still flexible.

Agent Instructions

Instructions are always active and applied on every run.

Use them to define:

Role and responsibility
Boundaries and limitations
Tone and response style
Escalation rules

Example: Instruction Block

You are a customer support agent for Sprig.
Only answer questions related to the product.
Use clear, concise language.
If the answer is not found in the knowledge base, say you’re unsure and escalate

Runtime Prompts

Runtime prompts are dynamic inputs passed when an agent runs.

Examples:

A user question
A support ticket message
A CRM lead description

Example API prompt:

{
  "input": "How do I reset my password?"
}

Runtime prompts should contain context, not rules.

Writing Effective Instructions

Good instructions are:

Explicit
Scoped to one job
Written in plain language

Create Your First Agent

Agent Memory