TechnicalCollectible prompt
Cleanup plan for a messy AI-built codebase
Act as a senior engineer brought in to tidy up a {{tech_stack}} app that was built quickly with AI tools. The area that's hardest to change right now: {{page_or_feature}}. Read the whole project before suggesting anything, and don't change code until I approve the plan.
Health check first:
1. Very large files and components doing too many jobs
2. Duplicated components, helpers and styles that should be one
3. Different ways of doing the same thing, such as data fetching, forms or error handling
4. Dead files, unused packages and leftover experiments
5. Missing types, hard-coded values, and settings that belong in environment variables
6. Anything fragile that has no test
Then the plan:
- Phases ordered by risk and payoff, each small enough for one AI prompt or one commit, with the app working after every phase
- For each phase: files involved, what changes, and how I can check nothing broke
- A handful of smoke tests to add first, locking in the key flows before anything moves
Finally, write a short project rules file for my AI tools: folder structure, naming, which components and helpers to reuse, how to fetch data and handle errors, and things never to do, such as adding a new library for something we already have.
Tell me which issues are cosmetic and can safely wait.
Technicalnvoka.com/library/nvoka-cleanup-plan-for-a-messy-ai-built-codebaseScan to open
Technical
Cleanup plan for a messy AI-built codebase
Get a phased plan to untangle an app that grew fast with AI, plus rules that keep future edits tidy.
Curated by Nvoka
Claude CodeCursorCodexWindsurf
Make it yours
Fill in the blanks and change any word. Only your copy changes, never the card.
Fill in the blanks
0 of 2 filledYour prompt
Act as a senior engineer brought in to tidy up a {{tech_stack}} app that was built quickly with AI tools. The area that's hardest to change right now: {{page_or_feature}}. Read the whole project before suggesting anything, and don't change code until I approve the plan.
Health check first:
1. Very large files and components doing too many jobs
2. Duplicated components, helpers and styles that should be one
3. Different ways of doing the same thing, such as data fetching, forms or error handling
4. Dead files, unused packages and leftover experiments
5. Missing types, hard-coded values, and settings that belong in environment variables
6. Anything fragile that has no test
Then the plan:
- Phases ordered by risk and payoff, each small enough for one AI prompt or one commit, with the app working after every phase
- For each phase: files involved, what changes, and how I can check nothing broke
- A handful of smoke tests to add first, locking in the key flows before anything moves
Finally, write a short project rules file for my AI tools: folder structure, naming, which components and helpers to reuse, how to fetch data and handle errors, and things never to do, such as adding a new library for something we already have.
Tell me which issues are cosmetic and can safely wait.