Hey HN — I built this after getting fed up with AI-generated PRs slipping
through code review unnoticed. TODOs everywhere, placeholder variables,
empty except blocks — the usual slop.
roast-my-code scans your repo with static analysis rules specifically tuned
for AI-generated code patterns, then calls an LLM (Groq free tier by default,
so $0 to try) to generate a brutal, specific roast referencing your actual
file names and issues.
Stack: Python + Typer + Rich + Jinja2. The HTML report exports a shareable
shields.io badge with your score.
Try it: pip install roast-my-code
Would love to hear what patterns you'd add — especially if you've spotted
AI slop in the wild that my analyzer doesn't catch yet.
Haha — it's surprisingly therapeutic to get roasted by your own tool.
I ran it on the repo itself and it called out my own placeholder names
in the test fixtures. The fallback roast lines weren't safe either.
Let me know what score you get if you try it! The worst I've seen so far
was a 12/100 on a legacy codebase with 200+ TODOs.
Hey HN — I built this after getting fed up with AI-generated PRs slipping through code review unnoticed. TODOs everywhere, placeholder variables, empty except blocks — the usual slop.
roast-my-code scans your repo with static analysis rules specifically tuned for AI-generated code patterns, then calls an LLM (Groq free tier by default, so $0 to try) to generate a brutal, specific roast referencing your actual file names and issues.
Stack: Python + Typer + Rich + Jinja2. The HTML report exports a shareable shields.io badge with your score.
Try it: pip install roast-my-code
Would love to hear what patterns you'd add — especially if you've spotted AI slop in the wild that my analyzer doesn't catch yet.
— I see what you did there.
I'm terrible about placeholder variables and functions. This thing might rip me to shreds.
Haha — it's surprisingly therapeutic to get roasted by your own tool. I ran it on the repo itself and it called out my own placeholder names in the test fixtures. The fallback roast lines weren't safe either.
Let me know what score you get if you try it! The worst I've seen so far was a 12/100 on a legacy codebase with 200+ TODOs.