AI Codebase Auditor

Built your game or app with ChatGPT, Claude, Copilot or Cursor? This finds the mess they leave behind.

AI coding assistants are fast, but they leave things behind: half-applied edits that break a whole script, API keys pasted into code, errors swallowed by empty catch {} blocks, "TODO: implement" stubs that silently do nothing, the same function written three times under different names, game_old.js copies everywhere, and sometimes packages that don't even exist.

Drag your project folder onto the auditor. A few seconds later you get a report with a health score from 0 to 100, every problem with its file, line and code, and a plain-English explanation of why it matters.

What it finds

• Broken JavaScript: unbalanced brackets and unterminated strings from half-applied edits, including scripts inside .html files
• Hardcoded secrets: API keys, tokens and passwords in your code, which every player can read in a web game
• Made-up packages: Python imports that don't exist on PyPI, names an AI invented that attackers can register
• Missing files: images, sounds, scripts and stylesheets your code points to that aren't there
• Swallowed errors: empty catch blocks, bare except:, errors ignored with no message (harmless best-effort cases are left alone)
• Unfinished code: empty stubs, "TODO: implement" placeholders, leftover debugger statements
• Duplicates and dead code: functions rewritten under new names, functions nothing calls, leftover _old / - Copy files
• Dependency mess: imports missing from requirements.txt, listed packages nothing uses, several libraries doing the same job

Plus the Handoff Pack

Switching AI assistants, starting a fresh chat, or coming back after a week? Drag your project onto Make Handoff.bat and give the next AI one short HANDOFF.md instead of letting it re-read (and guess at) every file. It lists the key files, your project's rules, build copies not to edit and old leftovers not to copy from, how to run it, what changed since the last handoff, open TODOs, known problems, and a file map. It's kept short enough for small local models too.

Fix it with your AI

Run it with --ai for a compact text version with the code around each problem. Paste it into your assistant and ask it to fix them, most important first.

Try it on the demo

The download includes Space Miner, a tiny project with typical AI mess planted in it: a broken script, leaked keys, a made-up package, swallowed errors, stubs, leftover copies and missing files. Audit it and see what the report looks like.

Works with

• HTML / CSS / JavaScript: web games and apps, plain scripts or modules, scripts inside .html files
• Python projects
• Build copies (web/, www/, Android assets) are recognised and reported once, not as duplicates. Third-party libraries and minified files are skipped.

Requirements

• Windows, macOS or Linux
• Python 3.10 or newer (free from python.org). Nothing else to install.
• Runs on your own computer. Your code never leaves it; the only internet use is looking up Python package names on PyPI, and --offline turns that off.

What it doesn't do

It points at problems; it doesn't change your code. It reads JavaScript without running it, so code built from strings at runtime can't be followed. Well-kept projects score 90+ (the open-source projects it was tested on scored 87 to 100), so a low score really means there's cleanup to do.

License

Use it on all your own computers, for any projects you work on, including commercial ones. Please don't share or resell it. Studios: one copy per person who uses it.

Questions, or a false alarm on your project? Leave a comment.