AI UX PlaygroundNewsletterJoin 2K+ AI designers and PMs on Substack. New teardowns, patterns, and prompts as they drop.

AI Workflows

Audit and test my Claude Code setup

Have Claude Code compare your skills, commands, and rules to a best-practice repo, then test every suggested change with a before/after prompt loop.

Prompt by

Brian Lovin@brian_lovin

Product Designer, Notion

Post

Prompt

Read the README on this repository: https://github.com/ChrisWiles/claude-code-showcase 

Explore the code base to learn about best practices and patterns for using Claude Code effectively. Take what you learn and bring it back into the context of our codebase. Look for opportunities to improve our own Claude Code setup (in the ./claude directory). Read our skills, commands, subagents, and rules files to get the full picture before making any suggestions.

When you're ready, create a list of your top recommendations and explain why you think they would be meaningful improvements to our project setup.

For each suggestion, create a testing framework so that you can measure how different prompts would behave or change the outcome given the new rule/setting/skill/etc.

The testing framework will let you run a sample prompt before making changes, take notes about the response/tool use/quality in your own response, and document everything in a BEFORE file. 

Then apply the suggested change to the rule/skill/command/setting file and re-run the exact same prompt. Observe the outcome and and save your notes in an AFTER file. 

Next, analyze the two documents to figure out if the changes to the Claude Code settings had a meaningful impact. If you aren't sure, run this loop with one more test prompt case (come up with the best possible prompt you can think of that would adequately test the changed setting).

Write a final before/after report that we can review together. When we find high impact changes, we'll implement the suggested change and commit. Don't stop until you've run the testing + evaluation loop on all of the recommendations from your first-phase exploration.

How to use

  1. 01

    Open Claude Code in a repo that already has a Claude setup (skills, commands, subagents, or rules).

  2. 02

    Paste the prompt. The agent reads the public claude-code-showcase repo, then your own setup.

  3. 03

    Let it run the before/after loop for each recommendation. Expect a long session.

  4. 04

    Review the final before/after report together and commit only the high-impact changes.

Tips

  • The prompt says "./claude directory". Claude Code keeps its config in `.claude/`, so point the agent there if it gets confused.
  • Work on a branch. The loop edits your rule, skill, and settings files while it tests them.