Civil engineer turned self-taught builder. I direct AI agents to write code while I do architecture, QA, and decisions. Currently building MSCodeBase Intelligence, an MCP server for codebase-aware AI
Since the source is available, I did the thing I usually do with tools I like: checked out 1.8.0 and wrote failing tests before touching anything. Six edge cases came out, all reproducible — path containment through links, a lighter write path than the read path (.git/.env), matching the disk copy while the editor had unsaved changes, undo overwriting later edits, CRLF, and a swallowed ======= inside a SEARCH.
Fixed all six red-first: 14 commits, tests 16 → 35, re-ran the series from a clean ab89034 checkout. Opened issue #3 with the summary and PR #4 with the fix. Happy to share the exact reproduction privately.
One product thought, not a bug: the gap is that "Apply Diffs" asks for trust without showing what it will touch. A small "3 modified, 1 created, 1 deleted" list before applying would close most of it.
No rush on the PR — it's there when you have time.
Thanks for taking the time for all of this 😭😭😭😭😭 This is literally the first PR i have ever got! I gave you my first gem too!
Let me review the PR as close as possible. Also the product thought will also be implemented! Next 2 days are holidays! I will start work right when college ends today (I am typing this from the college bus lol).
Thank you so much again!!! Can I put a link to your profile in the README?
bro this is sooo goood, turning free chatbots into agent is a greatt idea nd undo redo is a nice touchhh 🔥 starred it alreadyy keep buildingg 🚀 hehehe
also those recordings made by ur own giffypy tool is next level(peak marketing too lol)
JavaScript enthusiast with a growing passion for Python. I love exploring the world of programming, experimenting with code, and discovering new ways to turn ideas into reality.
JavaScript enthusiast with a growing passion for Python. I love exploring the world of programming, experimenting with code, and discovering new ways to turn ideas into reality.
JavaScript enthusiast with a growing passion for Python. I love exploring the world of programming, experimenting with code, and discovering new ways to turn ideas into reality.
AI testing storyteller. Writing the 36 Stratagems series — ancient Chinese military tactics meet modern AI system meltdowns. 16yr QA → building AI test framework.
Clever workaround for teams without a paid agent API — the clipboard bridge keeps the loop local and cheap. The main friction points are exactly what you'd expect from a manual handoff: context bloat from over-broad globs, and making sure pasted output lands back in the right file without clobbering unsaved edits. A glob allowlist plus a short manifest of which files were included makes the round-trip auditable; that's the same boundary discipline we keep in desktop extension messaging, where a handoff channel has to state what data it carries and who writes it.
Yes! This is the way in which i use AI for coding currently. Thanks for the comment 😊 The context includes the relative path to each file before the file contents as a heading. Is that enough? We can see how many files were included in the context at the bottom status text (that's what i use to ensure the globs matched correctly). But it doesn't list all files so its not fully reliable.
This is a smart way to keep using free chatbots without losing the editor workflow. I liked the all-or-nothing diff step most, because a half applied change is usually harder to fix than a failed one. The session undo for the last ten diffs also feels like the right safety net for experimenting with a model that can be confident and wrong at the same time. Nice work shipping a practical bridge instead of waiting for another paid promo.
AI testing storyteller. Writing the 36 Stratagems series — ancient Chinese military tactics meet modern AI system meltdowns. 16yr QA → building AI test framework.
AI testing storyteller. Writing the 36 Stratagems series — ancient Chinese military tactics meet modern AI system meltdowns. 16yr QA → building AI test framework.
JavaScript enthusiast with a growing passion for Python. I love exploring the world of programming, experimenting with code, and discovering new ways to turn ideas into reality.
JavaScript enthusiast with a growing passion for Python. I love exploring the world of programming, experimenting with code, and discovering new ways to turn ideas into reality.
Since the source is available, I did the thing I usually do with tools I like: checked out 1.8.0 and wrote failing tests before touching anything. Six edge cases came out, all reproducible — path containment through links, a lighter write path than the read path (
.git/.env), matching the disk copy while the editor had unsaved changes, undo overwriting later edits, CRLF, and a swallowed=======inside a SEARCH.Fixed all six red-first: 14 commits, tests 16 → 35, re-ran the series from a clean
ab89034checkout. Opened issue #3 with the summary and PR #4 with the fix. Happy to share the exact reproduction privately.One product thought, not a bug: the gap is that "Apply Diffs" asks for trust without showing what it will touch. A small "3 modified, 1 created, 1 deleted" list before applying would close most of it.
No rush on the PR — it's there when you have time.
Thanks for taking the time for all of this 😭😭😭😭😭 This is literally the first PR i have ever got! I gave you my first gem too!
Let me review the PR as close as possible. Also the product thought will also be implemented! Next 2 days are holidays! I will start work right when college ends today (I am typing this from the college bus lol).
Thank you so much again!!! Can I put a link to your profile in the README?
bro these guys dont comment or speak with me in my articles just they are going duo like they are just iam a invisible
Which guys? 🤔
@akhourianmolkumar and @effessdev
bro what?
Yeah we are the ones who comment on your posts lol.
Anyway i am going to sleep. Good night! 🌃
Yeah you didn't even debate the new features in the articles its technically jour job
bro this is sooo goood, turning free chatbots into agent is a greatt idea nd undo redo is a nice touchhh 🔥 starred it alreadyy keep buildingg 🚀 hehehe
also those recordings made by ur own giffypy tool is next level(peak marketing too lol)
hehe I learned marketing while being CTO @ your company! 🔥🔥🔥
bro this is appreciation or taunt lol I'm confused.
APPRECIATION no doubt!
hehehe then thank you lol, bro today is the best day for me in 2026.....
Yeah! Now you have 2 laptops (@koda2026 doesn't even have one lol poor Harun 😢 but I hope he'll get one soon!)
Bro first time I give you.....
Wow!!! 🤯🤯🤯 Thank you so much 🥲🥲🥲🥲🥲 How do I rate the quality of a post like that? Does DEV.to have such features?
Because bro now I am a trusted member!
Overcame discouragement from doubting classmates
A 10 years old developer who has become a DEV.to Trusted Member! 😱
Bruh I have to become a trusted member to rate the quality of a post?
This DEV.to is very time consuming... :(
@xulingfeng guide him😀
😂What is trusted member?
Clever workaround for teams without a paid agent API — the clipboard bridge keeps the loop local and cheap. The main friction points are exactly what you'd expect from a manual handoff: context bloat from over-broad globs, and making sure pasted output lands back in the right file without clobbering unsaved edits. A glob allowlist plus a short manifest of which files were included makes the round-trip auditable; that's the same boundary discipline we keep in desktop extension messaging, where a handoff channel has to state what data it carries and who writes it.
Yes! This is the way in which i use AI for coding currently. Thanks for the comment 😊 The context includes the relative path to each file before the file contents as a heading. Is that enough? We can see how many files were included in the context at the bottom status text (that's what i use to ensure the globs matched correctly). But it doesn't list all files so its not fully reliable.
This is a smart way to keep using free chatbots without losing the editor workflow. I liked the all-or-nothing diff step most, because a half applied change is usually harder to fix than a failed one. The session undo for the last ten diffs also feels like the right safety net for experimenting with a model that can be confident and wrong at the same time. Nice work shipping a practical bridge instead of waiting for another paid promo.
Thank you so much! 😊
And WorkBuddy also has free large models available to use. I think I was the one who recommended Qoder too, hahaha 😂
No no no no @unitbuilds recommended Qoder to me first. He was literally the first person to comment on my post.
I can't visit WorkBuddy:
Haha, I’m the one who recommended it to him.🤣
Ooookayyyy 🤯 Completely outsmarted me lol
Okay bro, I gave it a star! 😀
Thanks Star! 😊😊😊😊 As I promised I am sooo sooo grateful!!! 🥳
Bro😃
Hi Star!
You started THREE threads. Where am I supposed to talk now?