Aryam Goyal

hi...im aryam :)

[Aryam Goyal @AryamGoyal1]

Coding agents don't fail at writing code. They fail at knowing where to start.

FixMap gives any agent a map of your repo before it edits: ranked files with reasons, test routes, risk notes. Local, deterministic, no API key.

[Aryam Goyal @AryamGoyal1]

@aussiehaggie hi

[Aryam Goyal @AryamGoyal1]

claude just made me a 3D taj mahal
https://t.co/vVzfxmKpgj

[Aryam Goyal @AryamGoyal1]

@PSR_40 Active!!!!!!!!!!!!

[Aryam Goyal @AryamGoyal1]

@MartinOtie819 Hello

[Aryam Goyal @AryamGoyal1]

@enodrift https://t.co/DPtjOVy9Zx

[Aryam Goyal @AryamGoyal1]

@themishra4402 Uhhh I can tell you 3 why codex is better

[Aryam Goyal @AryamGoyal1]

@PSR43_ Hello

[Aryam Goyal @AryamGoyal1]

@Prasanthviru Every second

[Aryam Goyal @AryamGoyal1]

@V1adXBT Hello

[Aryam Goyal @AryamGoyal1]

@sattyyouneed AI writes all my code!

[Aryam Goyal @AryamGoyal1]

Any and all suggestions are welcome!

[Aryam Goyal @AryamGoyal1]

What should I ship next on FixMap?
https://t.co/DPtjOVxC9Z

[Aryam Goyal @AryamGoyal1]

@codewithDiyaa Vercel

[Aryam Goyal @AryamGoyal1]

@suni_code umm....whatsapp.......and discord group

[Aryam Goyal @AryamGoyal1]

My changelog told users a bug was fixed. It wasn't — not for them.

The fix was real. It was merged. It was on main. They just couldn't have it, because I hadn't published a release.

The bug itself was a good one. My tool ranks which files you should open for a coding task, and it deprioritizes build output, because editing a file the next build overwrites is always wrong. A vendored dependency inside Next.js — react-dom.development.js — sailed straight past that and ranked first at high confidence.

It escaped two independent guards.

The content check looks for minified output: 400+ characters per line, or two bundler fingerprints. This file is a readable dev build. 96 characters per line. One fingerprint.

The path check drops generated files that have a maintained source twin. A vendored dependency has no twin, by definition. That rule exists to protect chalk, whose vendored file genuinely IS the implementation.

So: too readable for one guard, too vendored for the other.

I proved it by adding a single bundler marker and changing nothing else. Score went 16 to 4. High confidence to low. The machinery was fine — it had simply never fired.

Then I shipped the mistake that's actually worth writing about. Fix merged, changelog updated, npm untouched. The site said "fixed" while `npm install` handed you the version without it.

Which is the exact bug that started this whole audit, pointed backwards. That one was: the site published an accuracy number the released code no longer achieved. This one: the site published a fix the released code didn't contain. Same root cause both times — a claim living somewhere other than the thing it describes.

The repair isn't "be more careful." I deleted the "Unreleased" section from the changelog, and the code that renders it. Now every entry is a version you can install, and putting an unshipped one back has to be a deliberate act.

An unreleased section in a public changelog is a promise about software nobody can run.

https://t.co/FvPFlZ7yiu

[Aryam Goyal @AryamGoyal1]

@Olao3w I have even less

[Aryam Goyal @AryamGoyal1]

@anah_sahh Ai exploration

[Aryam Goyal @AryamGoyal1]

@lightsilver323 https://t.co/DPtjOVy9Zx

[Aryam Goyal @AryamGoyal1]

@sflorimm https://t.co/DPtjOVxC9Z

[Aryam Goyal @AryamGoyal1]

@YashHustle_22 8