Prompting Guides

The system prompt is the single biggest lever you have over how a Cloudax Connect agent behaves on a call or in a chat. These guides distil OpenAI's official prompting advice for the model families that power your agents, with ready-to-paste snippets you can drop straight into an agent's system prompt.

Pick a guide

Which guide should I read?

Both guides share a common philosophy: be explicit, avoid contradictions, and iterate against real examples, but they target different model generations:

• GPT-4.1 is a fast, non-reasoning model that follows instructions extremely literally. Reach for its guide when you want low-latency agents and are happy to spell out every step.
• GPT-5 is a reasoning model with a tunable reasoning_effort. Its guide focuses on calibrating how much the model thinks and explores before it acts.

Universal principles

No advice is one-size-fits-all; prompting is an empirical discipline. Still, a handful of principles hold across every model we support:

• Be specific and literal. Modern models follow instructions closely. If behaviour is off, a single clear sentence stating the desired behaviour is usually enough to correct it.
• Remove contradictions. Conflicting or vague instructions force the model to waste effort reconciling them. Review long, multi-author prompts for rules that fight each other.
• Give examples. Demonstrate the exact behaviour you want, and make sure any behaviour shown in an example is also stated in your rules.
• Iterate with evals. Build a small set of representative test cases and re-run them after every prompt change so you can see whether an edit actually helped.

A reliable starting structure

Both guides recommend the same skeleton as a starting point for a system prompt. Add or remove sections to suit your use case:

# Role and Objective

# Instructions

## Sub-categories for more detailed instructions

# Reasoning Steps

# Output Format

# Examples
## Example 1

# Context

# Final instructions and prompt to think step by step