Chat Prompts

Chat Prompts

The Problem

AI confidently produces output — code, documentation, configs — but doesn’t automatically verify its own work. It writes a Helm chart and moves on. It documents a workflow and assumes it’s correct. Without explicit prompts to review, you get first-draft quality that might have bugs, inaccuracies, or missed edge cases.

The fix is simple: ask AI to review its own work before you accept it.

These are quick one-liners I use mid-conversation to catch issues before they become problems.


Verify What We Wrote

do some research online and think like a senior devops engineer,
does it make sense what we wrote? did we lie about anything?

When to use: After writing documentation, decision records, or technical explanations.

What you get: AI fact-checks claims, verifies technical accuracy, and flags anything that sounds wrong. It might find outdated information, incorrect assumptions, or statements that need nuance.


Self-Review Like a Senior Engineer

think like a senior [DOMAIN] engineer and review what we did

Replace [DOMAIN] with the relevant expertise:

DomainWhat it catches
DevOpsInfrastructure anti-patterns, CI/CD issues, deployment risks
Go developerCode style, error handling, idioms, missing tests
platform engineerArchitecture issues, security gaps, scalability concerns
UX engineerFlow problems, clarity issues, user experience friction

When to use: After completing a feature or significant piece of work. Don’t review after every small change — wait until you have something complete.

What you get: A second opinion that catches things you missed. Different domains surface different issues — a “senior Go developer” finds different problems than a “senior security engineer” reviewing the same code.


PR-Style Code Review

you're a senior Go developer, review the changes in our branch vs main
like you're reviewing a PR — check for bugs, style issues, missing tests

When to use: Before merging, or when you want a thorough review of accumulated changes.

What you get: Structured feedback like a real PR review — bugs, style issues, missing tests, potential improvements. AI compares your branch against main and reviews the diff, not just individual files.


The Pattern

These prompts share a common structure:

  1. Assign a role — “think like a senior X engineer”
  2. Define the task — “review what we did” / “fact-check this”
  3. Set expectations — “like you’re reviewing a PR”

The role matters. A generic “review this” gets generic feedback. A specific “senior platform engineer reviewing for production readiness” gets focused, relevant critique.


See also: Saved Prompts for longer, reusable prompts that persist across sessions.

Last updated on