• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Claude Code's Source Code Leaks Via npm Source Maps

erek

erek

Fully [H]
2FA
Joined
Dec 19, 2005
Messages
17,421
“Architecture Highlights
The Tool System (~40 tools): Claude Code uses a plugin-like tool architecture. Each capability (file read, bash execution, web fetch, LSP integration) is a discrete, permission-gated tool. The base tool definition alone is 29,000 lines of TypeScript.
The Query Engine (46K lines): This is the brain of the operation. It handles all LLM API calls, streaming, caching, and orchestration. It's by far the largest single module in the codebase.
Multi-Agent Orchestration: Claude Code can spawn sub-agents (they call them "swarms") to handle complex, parallelizable tasks. Each agent runs in its own context with specific tool permissions.
IDE Bridge System: A bidirectional communication layer connects IDE extensions (VS Code, JetBrains) to the CLI via JWT-authenticated channels. This is how the "Claude in your editor" experience works.
Persistent Memory System: A file-based memory directory where Claude stores context about you, your project, and your preferences across sessions.

Key Technical Decisions Worth Noting
Bun over Node: They chose Bun as the JavaScript runtime, leveraging its dead code elimination for feature flags and its faster startup times.
React for CLI: Using Ink (React for terminals) is bold. It means their terminal UI is component-based with state management, just like a web app.
Zod v4 for validation: Schema validation is everywhere. Every tool input, every API response, every config file.
~50 slash commands: From /commit to /review-pr to memory management -- there's a command system as rich as any IDE.
Lazy-loaded modules: Heavy dependencies like OpenTelemetry and gRPC are lazy-loaded to keep startup fast.“

Source: https://developers.slashdot.org/sto...e-codes-source-code-leaks-via-npm-source-maps
 
Ya, but Claude has mostly figured out coding, so developers are not needed... and yet they have 50+ roles open on their website and some 5K+ issues in their git. And with Anthropic pushing token's costs and more limits on people who bought unlimited Pro/Max accounts, and then shutting down access to those plans to stop them being used with OpenClaw or other 3rd party interfaces, because they want you to ONLY use their interfaces, or pay more for their API plans...


Claude-Screenshot_20260319_151626.png
 
MrGuvernment said:
Ya, but Claude has mostly figured out coding, so developers are not needed...
Click to expand...
that seem such a strange way to interprety what he said, he is just saying he does not write single line of code anymore (I barely do either), does not mean a non developer could control its coding agents like he do. Is prediction is in ~2 year, engineer that are not coding expert would be able to control coding agents, but not yet, at least in the 2 minutes I did listen of that interview when he talk about that.

When you read old books of the 70s, they talk as if programming will not be needed really, because compiler will be so good you will be able to just right COBOL and SQL code instead, people writting those ended up being called programmer.Would program stay as complex as they were in the 70s, non decade long coders would have been able to write them in Basic is probably not far from the truth, will see the same things, for a lot today programs Ai agents in the hands of a competent small teams, they will be able to do them, but software will explode in complexity again will be my guess.
 
Last edited:
Eliminate coders so only AI's really code............so who owns the AI's who will code for you? Right. And that's a 2 dimensional not 3D Chess question, I'm sure there are far deeper implications......
 
I discovered recently that saying you are "not writing a single line of code" is just a semantic trick, yes technically you are not "writing a single line", the AI spits out pages of code and the engineers have to go through all of it and fix it. Not sitting back in their chairs while things get done. This is arguably a worse way of working but if you dare say that no one will give you a billion dollars.
 
myzz said:
I discovered recently that saying you are "not writing a single line of code" is just a semantic trick, yes technically you are not "writing a single line", the AI spits out pages of code and the engineers have to go through all of it and fix it. Not sitting back in their chairs while things get done. This is arguably a worse way of working but if you dare say that no one will give you a billion dollars.
Click to expand...

As a complete Powershell doofus, I have been using one of the AI systems to write Powershell code for my personal use. Things like ID all flagged posts in an Outlook PST file, list all installed programs and create a table that includes all winget calls, rename folders, copy highlighted text in a PDF, etc., etc. It usually takes me 3-4-5 iterations to get the description right and then to debug the code. It seems the AI I use hasn't had enough PowerShell training. (I'm not naming this AI because I don't want to embarrass myself.)

Now I know that I'm no developer and my applications run maybe 500-1500 lines of code, some with a simple GUI. That's nothing for a real commercial system with input validations, a proper GUI, security checks, etc,. etc that are absolutely necessary in a commerical product. Me, I wouldn't even know how to describe those to the AI.
 
I like Claude, but the increased restrictions on a paid subscription is pretty stupid. However, Claude recently built a full on app for me from scratch for macOS and that's pretty amazing. I tried doing that with ChatGPT and Gemini and all they could make me was Apple Scripts that couldn't run.
 
Shoganai said:
I like Claude, but the increased restrictions on a paid subscription is pretty stupid. However, Claude recently built a full on app for me from scratch for macOS and that's pretty amazing. I tried doing that with ChatGPT and Gemini and all they could make me was Apple Scripts that couldn't run.
Click to expand...
Just out of curiosity, have you tried Grok?
 
philb2 said:
As a complete Powershell doofus, I have been using one of the AI systems to write Powershell code for my personal use. Things like ID all flagged posts in an Outlook PST file, list all installed programs and create a table that includes all winget calls, rename folders, copy highlighted text in a PDF, etc., etc. It usually takes me 3-4-5 iterations to get the description right and then to debug the code. It seems the AI I use hasn't had enough PowerShell training. (I'm not naming this AI because I don't want to embarrass myself.)

Now I know that I'm no developer and my applications run maybe 500-1500 lines of code, some with a simple GUI. That's nothing for a real commercial system with input validations, a proper GUI, security checks, etc,. etc that are absolutely necessary in a commerical product. Me, I wouldn't even know how to describe those to the AI.
Click to expand...
Yes I think this is the accurate description of the workflow. There have been several studies where it takes developers using AI X% longer to do tasks but they actually think they are working X% faster. Roughly around 20ish percent for both IIRC.
 
myzz said:
I discovered recently that saying you are "not writing a single line of code" is just a semantic trick, yes technically you are not "writing a single line", the AI spits out pages of code and the engineers have to go through all of it and fix it. Not sitting back in their chairs while things get done. This is arguably a worse way of working but if you dare say that no one will give you a billion dollars.
Click to expand...

And in 10-15 years when all those senior developers retire there won't be anyone who had experience writing code to properly review and fix what AI wrote. Yes, the AI will improve over time but someone needs to program it to improve. Someone needs to tweak it when it starts doing weird stuff. If the world goes the way the AI tech bros insist it will go then we'll quickly end up with a bunch of industry with little to no talent left when the AI doesn't do the job correctly. That said, I don't think it will go that way and their hiring practices indicate they're still hiring people while claiming they don't need as many. They're just bragging to get people to spend more money on AI, not because they honestly think it'll wipe out all developers or other jobs.
 
sleepeeg3 said:
FTFY. It wil do it far less than 10-15 years.


Claude > Grok > Copilot/ChatGPT. Have not tried Gemini for code, but it's probably one of the better ones, based on its comprehensive answers to other questions.
Click to expand...
Personal experience: Gemini isn’t great. Used it to merge two non-version controlled branches of source code and spent more time tracking down errors and fixing bugs than if I did the merge manually line by line with a diff tool.
 
Shoganai said:
I like Claude, but the increased restrictions on a paid subscription is pretty stupid. However, Claude recently built a full on app for me from scratch for macOS and that's pretty amazing. I tried doing that with ChatGPT and Gemini and all they could make me was Apple Scripts that couldn't run.
Click to expand...
at one point, usage should match cost and an API per tokens model is quite straight forward, pay for what you use, nothing more, nothing less, is rarely stupid and something people tend to love (people do complain about subscription everything all the time), how high volumen their token usage is for tasks is more the issue.

sleepeeg3 said:
Claude > Grok > Copilot/ChatGPT. Have not tried Gemini for code, but it's probably one of the better ones, based on it's comprehensive answers to other questions.
Click to expand...
GPT codex 5.3 (and now 5.4) was quite the jump in coding agents ability, I doubt Grok get that close to it.
 
bigdogchris said:
When AI starts coding itself, that's when the problems start :)
Click to expand...
I would say when they start combining AI with quantum computers ... which will probably be around 2030 ... a lot of weird things are going to start happening.
 
sc5mu93 said:
Personal experience: Gemini isn’t great. Used it to merge two non-version controlled branches of source code and spent more time tracking down errors and fixing bugs than if I did the merge manually line by line with a diff tool.
Click to expand...
This is the dirty secret, like those studies I mentioned. Extra sinister because it makes people feel like they're going faster.
 
Vermillion said:
https://techtrenches.dev/p/the-snake-that-ate-itself-what-claude

Great read about how bad the Claude code actually is. This is pure AI slop. To fix slop they generate more slop.

They knew there was a bug that was burning through 250,000 API calls a day. 3 lines of code fixed it but they left it there and moved on.
Click to expand...
It's important to point out that Claude Code is a separate product from the Claud chat app we use. This issue has nothing to do with the latter.
 
Shoganai said:
I like Claude, but the increased restrictions on a paid subscription is pretty stupid. However, Claude recently built a full on app for me from scratch for macOS and that's pretty amazing. I tried doing that with ChatGPT and Gemini and all they could make me was Apple Scripts that couldn't run.
Click to expand...
Our team has made/making 6 different products by strictly using Claude since October. 3 are fully released, 1 is in the finial stage, 2 are in beta stage.

It has a much faster turn around time, more stable and less time consuming than using the hired people that we have used in the past.

Is it perfect? No, sometimes it hallucinate giving you bs results/responses that you have to call it out on and deal with it going full r-tard mode, overall not bad. Ive used it to make scripts that I could of done by hand, but would of took an hour or two (or more if I made a typo!), in 3 minutes. It's not bad if you know how to use it.
 
myzz said:
This is the dirty secret, like those studies I mentioned. Extra sinister because it makes people feel like they're going faster.
Click to expand...
I programmed for a few months,(long story). As a %100 noob (only learning by a programming by a book) My first assignment I wrote from scratch, like every single line was mine. They complained about the time it took. I then figured it out the trick was finding programs that did about what I wanted and modified them to do exactly like I wanted. It was way faster, BUT if AI gets better, why not let AI find and modify the program and let a programmer or even an analyst validate the code? Obviously this would only be good for entry level programming. My manager's manager had software that could build simple code that he would then read and update as needed and this was 10 years ago. I also was witness to software that wrote sql queries based on plain text requests (the sample was how many abc did I buy in 2015) . None of the software indicated AI but I would bet money it is now called AI.
 
Nothing wrong with using finished code to build something new, that's what libraries and packages and open-source is. We would get nowhere without this, everything is built on top of other things: HTTP, TCP/IP, Unix, etc. The trick is up until now those were made by humans who care.
 
sc5mu93 said:
Personal experience: Gemini isn’t great. Used it to merge two non-version controlled branches of source code and spent more time tracking down errors and fixing bugs than if I did the merge manually line by line with a diff tool.
Click to expand...
Yeah I had a year of free gemini from the deal for university students and I recently tried Claude, man Claude is just so much better. On the other hand though, I immediately run out of tokens on Claude within like an hour. I rarely run out on Gemini.
 
You must log in or register to reply here.
Back
Top