[Painsawman123]

Really surprising how many people are downplaying this leak! "Google and OpenAi have already open sourced their Agents, so this leak isn't that relevant " What Google and OpenAi have open sourced is their Agents SDK, a toolkit, not the secret sauce of how their flagship agents are wired under the hood! expect the takedown hammer on the tweet, the R2 link, and any public repos soon

[loveparade]

It's exactly the same as the open source codex/gemini and other clis like opencode. There is no secret sauce in the claude cli, and the agent harness itself is no better (worse IMO) than the others. The only thing interesting about this leak is that it may contain unreleased features/flags that are not public yet and hint at what Anthropic is working on.

[IceWreck]

> What Google and OpenAi have open sourced is their Agents SDK, a toolkit, not the secret sauce of how their flagship agents are wired under the hood

And how is that any different? Claude Code is a harness, similar to open source ones like Codex, Gemini CLI, OpenCode etc. Their prompts were already public because you could connect it to your own LLM gateway and see everything. The code was transpiled javascript which is trivial to read with LLMs anyways.

[MallocVoidstar]

Codex is open source: https://github.com/openai/codex

[weird-eye-issue]

It doesn't matter that much. Trust me you could just have an LLM reverse engineer the obfuscated code.

[sodapopcan]

The point is that a "secure coding platform" leaked something they were trying to keep under wraps, whether the contents of the leak matter or not.

Also, as many others have pointed out, there is roadmap info in here that wouldn't be available in the production build.

[Capricorn2481]

> The point is that a "secure coding platform" leaked something they were trying to keep under wraps, whether the contents of the leak matter or not

Sure, but that's completely different from what they were responding to to, which was someone insinuating the Claude Code CLI has secret sauce that makes it better than the competition.

[sodapopcan]

Fair, I was just seeing so much of this and randomly responded to yours.

[weird-eye-issue]

No reasonable person actually thinks that Claude Code cannot make mistakes. So if your point is that this is going to change anybody's opinion about it then I think that's pretty silly.

Also who really cares about the roadmap? Any feature they release can be easily copied quickly. The only moat they have at the moment is in giving access to their models via a subscription.

[sodapopcan]

> No reasonable person actually thinks that Claude Code cannot make mistakes.

This was insanely rookie mistake that could have been caught if anyone was paying attention as opposed to "vibing."

[weird-eye-issue]

As a Claude Code user why should I care?

[ithkuil]

yeah it actually works to use claude to reverse engineer itself; I've used that to workaround some problems. E.g. that's how I discovered that I had to put two slashes for absolute paths in sandbox config. The thing is, the claude team is so quick that soon enough they add more and more features and fix more and more bugs that your workarounds become obsolete

[petcat]

> they add more and more features and fix more and more bugs

My experience has been that they add far more bugs in every release than they fix

[kaszanka]

Is https://github.com/google-gemini/gemini-cli not 'the flagship agent' itself? It looks that way to me, for example here's a part of the prompt https://github.com/google-gemini/gemini-cli/blob/e293424bb49...

[hmokiguess]

Do you think the other companies don’t have sufficient resources to attempt reverse engineering and deobfuscating a client side application?

The source maps help for sure, but it’s not like client code is kept secret, maybe they even knew about the source maps a while back just didn’t bother making it common knowledge.

This is not a leak of the model weights or server side code.

[danmaz74]

I guess that the most important potential "secret sauce" for a coding agent would be its prompts, but that's also one of the easiest things to find out by simply intercepting its messages.

[mholm]

The only real secret sauce is the training methods and datasets used for refining harness usage. Claude Code is a lot better than gemini-cli/open-code/etc because Claude is specifically trained on how to run in that environment. It's been rlhf'd to use the provided tools correctly, and know the framework in which it operates, instead of relying solely on context.

[nunez]

Yeah, this is the LLaMa leak moment for agentic app dev, IMO. Huge deal. Big win for Opencode and the like.

[mmaunder]

Agreed. This is a big deal.