so @LVNilesh sir told me to try breaking into his network as a challenge
found exposed backend internals, user systems, config, and way more than I expected
and apparently the backend was happy to let me speak for it
Reached 100 stars on Codra.
Thank you to everyone who starred it, opened issues, or gave it a try. Means a lot for a side project.
Onward. https://t.co/7jPoxlZ2HD
I'll be posting 1 free tool daily on Codra. These utilities help developers ship better code faster.
Day 1: An instant .gitignore generator for any stack. Just type the environment and copy the config.
This open-source project makes AI code review free by letting you run your own reviewer on GitHub π₯
Most AI code review tools charge you for every repository, PR, or review.
This takes a different approach.
Connect your GitHub repo, and it automatically reviews pull requests, understands the repository context, and posts findings directly inside the PR.
You can run the whole thing yourself and bring your own model.
No separate code review SaaS. No per-review bill.
What you get:
> Automatic reviews on new and updated PRs
> Inline comments + review summaries
> Open-source model support
> Per-repository review rules
> Custom skipped paths and triggers
> Model chains and fallbacks
> Self-hosted infrastructure
> Review history and job monitoring
Itβs still in beta, but the idea is simple:
Your code. Your models. Your code reviewer.
ππ‘ππ§ πππ«π―π’ππ π πππ₯π₯π¬ πππ«π―π’ππ π, ππ§π πππ«π―π’ππ π π‘ππ§π π¬...
Service A exhausts its thread pool and crashes too.
This is a cascading failure.
Bitcoin gets called digital gold, internet money, a scam, a revolution.
My new post is none of that. It is just how the machine works: hashes, keys, UTXOs, proof of work, built up one layer at a time.
With demos you can run yourself.
https://t.co/8SHcCUgFF5
Codra's evidence grounding discards every hallucinated finding: of ~14,800 findings generated across the sweep.
Check it out: https://t.co/7jPoxlZ2HD
PS: These numbers are from the v0.9.5 work, not yet released.
PPS: Special thanks to @Gautam_A_k for helping me with this!
Across 787 runs, Codra posted 0.003 false positives per review, 99.97% precision, matching Opus.
At its default settings, Codra running a flash-lite model posted the same ten true positives per review that Claude Opus did, at ~1,700 tokens and 24 seconds per defect found.
This is the one π₯
Codra = open-source, self-hosted AI PR reviews with inline comments, model chains/fallbacks, per-repo rules, and a real dashboard.
Star + try it:π https://t.co/7jPoxlZ2HD
I built Codra, an open-source self-hosted AI code review engine for GitHub pull requests.
Built on Cloudflare Workers, posts inline comments, tracks reviews, and supports per-repo settings, model fallbacks, and queues.
For teams that want AI review without giving up control.
@abhxy03 As @sunnykgupta mentioned, do take help from AI to get the work done!
Iβd go as far as saying let AI do the heavy lifting, but with one base rule: You must understand every line it generates.
If you're blindly copying code you can't explain, pause and learn the basics first.
Recreated the iconic Vercel Ship 3D card for my portfolio!β‘οΈ
Always loved the interactivity of the original, so I had to build my own version. The floaty physics are so satisfying to play with.
Check it out:π https://t.co/X4YFPXexeI
I thought my code was perfect, until a user tried it.
The UI was flickering. Every keystroke triggered a request: Type "H" β Request sent. Type "He" β Request sent. Type "Hel" β Request sent.
Fixed it with a 300ms debounce. Now the app waits for a pause before fetching data.