How would you separate instructions, context, and user data in a prompt?

Builds a readable request with stable instructions and clearly delimited, provenance-aware untrusted inputs.

1Say itAnswer out loud first

2CompareCompare with Core answer

3RefineRefine with Stronger and Advanced

TL;DR

I keep stable task instructions separate from the current user request and from supporting data. Each context item gets a clear label, source, and boundary, and untrusted text is presented as content to analyze rather than instructions to follow. I pass only the data the task needs instead of mixing configuration, history, and arbitrary documents into one string.

Full answer below

Practice focus

Whether the candidate can construct context without confusing data with control or hiding important boundaries inside a monolithic prompt.

Answer first

Answer path

Core answer first, then checkpoints toward a stronger answer.

Start
Core answer
Start here. This is the direct spoken answer to practice first.
Overview
Clear structure improves maintainability and model behavior, but the application must still enforce the trust boundary the structure describes.
I keep stable task instructions separate from the current user request and from supporting data. Each context item gets a clear label, source, and boundary, and untrusted text is presented as content to analyze rather than instructions to follow. I pass only the data the task needs instead of mixing configuration, history, and arbitrary documents into one string.
Checkpoint
Stronger answer
Optional upgrade with examples, trade-offs, and sharper phrasing.
A practical request might contain a stable classifier instruction, a small category definition, then a labeled ticket body and relevant account facts. I would escape or serialize variable fields consistently and state what to do when the inputs conflict or lack evidence. This makes failures easier to inspect and prevents accidental prompt construction bugs such as one field swallowing the next section.

Review signals

Use these to catch weak phrasing, shallow reasoning, and risky claims.

Common mistakesCandidateCommon mistakesMistakes that weaken the answer or signal shallow understanding.Open details

Details

Concatenating raw inputs with no labels or escaping.
Sending all available history because it fits in the context window.
Presenting retrieved documents as application instructions.
Logging complete prompts containing sensitive data by default.

How would you separate instructions, context, and user data in a prompt?

Practice focus

Answer path

Overview

Review signals

Follow-up prompts

Continue practice

Next interviewer turn

Tech stacks

Skill areas

Human editorial direction

Was this drill useful?

Found this answer useful?