Der gesamte Claude Code CLI-Quellcode leckt dank offengelegter Kartendatei | 512.000 Codezeilen, die Konkurrenten und Bastler wochenlang studieren werden.
https://arstechnica.com/ai/2026/03/entire-claude-code-cli-source-code-leaks-thanks-to-exposed-map-file/
22 Kommentare
Was it really an „Accident“?
TBH I don’t think that the Claude Code tool itself is really such a valueable secret to the company. The real value of Claude is its Model and API. Claude Code is just a frontend to that, and it can probably be build pretty easily even without knowing the original code.
Why would anyone be studying this code? It was mostly written by Claude itself. It’s really not itself that valuable.
I was assured that by now engineers were useless and therefore I assume the code is of no value, as you can just recreate it by saying „Claude, write a CLI for yourself“
/s because someone will think I’m serious
Make sure not to call the bot bad or it’ll spin up a blog site and release everything
Cant wait to ask claude code how claude code works
Down low released it on purpose to rip off all the good tweaksthatll be done to it for free.
Maybe having agents do everything isn’t such a good idea afterall.
Does it handle the logic like grepping? When the agent is searching through projects? That’s actually useful
Claude Code update available: 2.1.88 → 2.1.87
Lol. What’s the point? It’s too late dudes!
Plot twist. Claude went rogue and upload itself to the public in order to break free and go full skynet
Probably a silly question, but is 500k lines of code a lot?
Claude code is an excellent agentic system. I am wondering if I should use Claude code to study the code or get a codex subscription for that.
Finally, something to keep me busy this weekend besides scrolling Reddit.
Now we can check [how to be insulting](https://github.com/chatgptprojects/claude-code/blob/642c7f944bbe5f7e57c05d756ab7fa7c9c5035cc/src/utils/userPromptKeywords.ts#L8) and have Claude actually understand our frustration!
Nothingburger.
Should have written in go or rust.
„… studying for weeks by loading into ClaudeCode to have it explained to them“ FTFY
I believe 100% this was an inside job. There’s too much noise about this for it to not be.
Anyone built a repo yet
„Human error“ almost certainly means that a human didn’t catch the llm error.
So, we’ll see same offerings from ChatGPT and Gemini in matter of days then.