16 short prompts I type to my coding agents

Most of these prompts are a sentence or two. One is two words: “Elaborate, simply.” I type it when an agent’s update is too dense to follow.

These are 16 prompts I type. Anything in angle brackets, like <X>, is a placeholder.

Investigate and diagnose

Ask for the cause before accepting a fix.

What’s the root cause? Anything to fix or fixed?

Ask the agent to find why a repo keeps regressing.

Study commit history and sessions for this repo. Why are there so many regressions. Is it the codebase or instructions, workflow or something else? Ask clarify questions first, if needed.

Find out why a tool uses so much disk and what to do about it.

Why is Codex eating up so much diskspace. Can we optimize the usage and/or would it help to have a weekly cleanup?

Study and research

Compare models and reasoning levels on your own recent work.

Study sessions for the past 7 days and check for effectiveness (task) and efficiency (wall clock time, token output speed and $ token) and compare models and reasoning levels.

Turn a video into lessons for your own project.

Study this video (speech to text + study frames for screenshots, if needed). It’s about <X>. Why is it supposed to be better than <Y>? Learn from it and if we can improve <Z>. <URL>

Quick feasibility check on an idea before committing to it.

Study if <idea> would help and be feasible?

A recurring look at product analytics since the last review.

Study PostHog analytics for activity since we last check. how does user navigate, etc. Anything we can improve?

Understand and explain

Use when the agent’s update is too dense to follow.

Elaborate, simply.

Learn a CLI from what was actually run.

Give me the list(s) of commands executed for our test flow(s). Contrived examples are enough, but I want to see correct cli usage to understand how it works.

Ask how side work connects to the main product.

How does <X> help our core product?

Use when asked to review something without knowing where it goes.

What is this title/body we are reviewing for? Where is it going to be used?

Decide and plan

Collect every open question in one place before work starts.

What do I need to supply to you, what are you not sure about, or need to confirm, for you to proceed? List them here.

Use when the agent asks for a decision without making the choice clear.

Elaborate, simply. what do you want me to decide?

Build and implement

Tune visual values live instead of rebuilding for each change.

Can we add a temp tab to Settings with textboxes for the <parameters> to be read when we want to do that <effect>? So I can try it, test it with minimal restarts