Hacker Newsnew | past | comments | ask | show | jobs | submit | usrbinbash's commentslogin

And in a skill, I can store the secret in the skill itself, or a secure storage the skill accesses, and the agent never gets to see the secret.

Sure, if I want my agents to use naked curl on the CLI, they need to know secrets. But that's not how I build my tools.


what stops the agent from echoing the secure storage?

what i see is that you give it a pass manager, it thinks, "oh, this doesn't work. let me read the password" and of course it sends it off to openai.


> what stops the agent from echoing the secure storage?

The fact that it doesn't see it and cannot access it.

Here is how this works, highly simplified:

    def tool_for_privileged_stuff(context:comesfromagent):
        creds = _access_secret_storage(framework.config.storagelocation)
        response = do_privileged_stuff(context.whatagentneeds, creds)
        return response # the agent will get this, which is a string
This, in a much more complex form, runs in my framework. The agent gets told that this tool exists. It gets told that it can do privileged work for it. It gets told how `context` needs to be shaped. (when I say "it gets told", I mean the tool describes itself to the agent, I don't have to write this manually ofc.)

The agent never accesses the secrets storage. The tool does. The tool then uses the secret to do whataever privileged work needs doing. The secret never leaves the tool, and is never communicated back to the agent. The agent also doesn't need, or indeed can give the tool a secret to use.

And the "privileged work" the tool CAN invoke, does not include talking to the secrets storage on behalf of the agent.

All the info, and indeed the ability to talk to the secrets storage, belongs to the framework the tool runs in. The agent cannot access it.


OpenAI is not the worst it could or would send it to.

> The core philosophy of MCP is simple: it’s an API abstraction. The LLM doesn’t need to understand the how; it just needs to know the what.

Wrong. It needs to "understand" both these things. The only difference is where and how the strings explaining them are generated.


That's an important point.

Whether it's tools, MCP or skills: they are fundamentally all just prompts. Even if the LLM is trained to recognize those and produce the right shape of tokens that validate most of the time.

But I wouldn't use the word "understand" here, because that builds the wrong intuition. I think a more useful term would be "get guided by" or "get nudged by". Even "recognize" is slightly misleading, because it implies too much.


> I don’t understand how people can remember all these custom scripting languages.

We can't.

Why do you think the `man` command exists?


Same here, so I wrap the post into an agent skill. Hope I can use them next time. https://github.com/yujiachen-y/codebase-recon-skill

As the old saying goes: "This too will pass."

Consumer hardware will always be a market worth serving for companies who don't see their stock price as their product.

If the existing companies are unwilling to make a sale, I am sure new players will arise picking up their slack.

https://www.youtube.com/watch?v=SrX0jPAdSxU


> iran's dickhead move...

Remind me again, which country started this whole mess?

> what choice do the gulf nations, or even all the asian+european (strait users) nations have?

They can go "yeah, you know, the US has been less than reliable as an ally recently, what with absurd tariffs, saber rattling around greenland, belitteling NATO, etc., and they seem unwilling to change, so we're just gonna pay the piper, and get oil, and make arrangements with the Chinese (aka. the worlds most powerful industry), and if they US doesn't like it, that sounds like a them-problem..."

What's very likely not gonna happen, is other countries fighting the US's war for them. NATO already told trump no, other countries won't give different answers.

And anyone who wants to actually invade Iran...well, let's put it this way: Iran is 3-4 times the size of Afghanistan, with even more difficult terrain, and has a standing army of 600,000 men, with over 300,000 in reserve. They have an air force, are proficient in the manufacture of drones, have a working intelligence network. And they've had 4 decades to dig into defensive positions.

In short, it's not gonna happen.


Don't think there is much of a point replying to this person seriously as he is obviously a troll. You can take half a minute to check his profile


People having worldviews you disagree with does not make them "trolls".


> which country started this whole mess?

what has already started, is already started -- I agree on Trump being dick, but does that make iran's "making new enemies" a wise move?

> NATO already told trump no, other countries won't give different answers.

of course it said no BEFORE IRAN started the $2M toll (and other countries don't like trump due to tariff-for-everyone)

if the current iran regime was strategically wise, iran should have fired everything it got to Israel, and make the missile interception rate down to 40%. That would have actually showed it's power.

now, with even UAE's missile interception rate of 96%, iran actually showed its missiles are nuisances, not some existential threat.

600,000 men and 300,000 in reserve -- well that would have mattered a lot in medieval wars... "they have an airforce" -- well do they actually have planes? "have a working intelligence network" -- hmm...

no you're way way way over-estimating iran

the only strategic move for iran was selecting one specific target (israel) and focusing all its might, not becoming a rambo


Their win condition isn't destroying Israel, its outlasting the American will for the war until a leadership change happens. They aren't the attackers in this war. They need to just defend until America and Israel give up because it is too costly at home.


> its outlasting the American will for the war until a leadership change happens

well even in the best-case scenario (trump impeachment), I highly doubt any democrat president can actually stop at status quo -- rather, the next president has exactly zero choice but to wipe out iran MORE than trump (and call trump a weakling)

just leave Iran be and get out? well he/she could, GIVEN that Iran didn't show its potential to be bully on the gulf states and didn't even think about that $2M toll...

now? well even if a pirate has a sad back story, doesn't mean the navy can leave them be.

by missiling everyone nearby, iran just became too dangerous to nearby neighbors...

by even talking about $2M toll, iran just became too financially dangerous even to strait users... I mean, even if it's "just $2M", what will stop iran from asking $5M, $10M, or even $100M ?


> iran's "making new enemies"

Those countries were already enemies of Iran by virtue of housing US bases, military installations, etc.


> what has already started, is already started -- I agree on Trump being dick, but does that make iran's "making new enemies" a wise move?

There is no downside on making the Gulf states enemies. Quite to the contrary: they might lobby the USA to end this madness. It's a serious damage to the importance of the USA in the region if it can't or doesn't want to open the strait again, either by force or by making a deal.


Absolutely true, but there is a silver lining:

When people rewriting open source libs with a bot then come crying to maintainers that their rewrites have bugs, and they would like for someone to fix said bugs for free, there is absolutely no one who will feel obligated to help them out.


> Because it takes a massive amount of developer work

You know what else takes "a massive amount of developer work"?

"any LLM-generated code must be reviewed by a good programmer"

And this is the crux of the matter with using LLMs to generate code for everything but really simple greenfield projects: They don't really speed things up, because everything they produce HAS TO be verified by someone, and that someone HAS TO have the necessary skill to write such code themselves.

LLMs save time on the typing part of programming. Incidentially that part is the least time consuming.


The submitter is supposed to be the good programmer; if not, then maintainers may or may not review it themselves depending on the importance of the feature.

And yes of course they need to be able to write the code themselves, but that's the easy part: any good developer could write a full production OS by themselves given access to documentation and literature and an enormous amount of time. The problem is the time.


> The submitter is supposed to be the good programmer;

And how will that be assured? Everyone can open a PR or submit a bug.

> The problem is the time.

But not the time spent TYPING.

The problem is the time spent THINKING. And that's a task that LLMs, which are nothing other than statistical models trying to guess the next token, really aren't good at.


Well, assuming you care about verification, of course. If it's got that green checkmark emoji, it ships!


> Not sure how they can expect to make a viable full OS without massive use of LLMs, so this makes no sense.

Every single production OS, including the one you use right now, was made before LLMs even existed.

> What makes sense if that of course any LLM-generated code must be reviewed by a good programmer

The time of good programmers, especially ones working for free in their spare time on OSS projects, is a limited resource.

The ability to generate slop using LLMs, is effectively unlimited.

This discrepancy can only be resolved in one way: https://itsfoss.com/news/curl-ai-slop/


There are only 4 successful general purpose production OSes (GNU/Linux, Android/Linux, Windows, OS X/iOS) and only one of those made by the open source community (GNU/Linux).

And a new OS needs to be significantly better than those to overcome the switching costs.


> There are only 4 successful general purpose production OSes

Feel like you are using a very narrow definition of "success" here. Is BSD not successful? It is deployed on 10s of millions of routers/firewalls/etc in addition to being the ancestor of both modern MacOS and PlaystationOS...


What about IBM i and z/OS, and Stratus VOS, and Burroughs MCP, and Tandem GUARDIAN, and VxWorks and OS-9 and… These all not only still exist but run huge transaction volume (for the mainframe and minicomputer systems) and run a huge amount of embedded systems (for the embedded OSes).


> And a new OS needs to be significantly better than those to overcome the switching costs.

Who cares if nobody switches to it as their daily driver? The goal you proposed was "viable", not "widely used". The former is perfectly possible without LLMs (as history has proved), and the latter is unrelated to how you choose to make the OS.


None of this counters the argument I made above :-)


Just because they have been made before LLMs doesn't mean it can be done again, since there was just one success (GNU/Linux) and that success makes it much harder for new OSes since they need to better then it


> Just because they have been made before LLMs doesn't mean it can be done again

Erm...no? That's exactly what that means.

Earth-Ovens haven't been in widespread use for hundreds of years. People can still use them to bake bread however: https://www.youtube.com/watch?v=WAJqGVxuJPo


Well, by this logic there have been 0 successful OSes made with LLMs so far...


The sad part is, how infinitely more functional these simple, static HTML documents are, compared to much of the shit that floods the "modern" web.

Ofc these pages cannot replace SPAs. That's not the point. The point is: Much of the web isn't SPAs. And much of what is SPAs shouldn't be SPAs. Much of the web is displaying static, or semi-static information. Hell, much of the web is still text.

But somehow, the world accepted that displaying 4KB of text somehow has to require transmitting 32MiB of data, much of it arbitrary code that has no earthly business eating my CPU cycles, as the new normal. Somehow everyone accepts that text-only informational pages need to abuse the scroll-event, or display giant hero-banners. Somehow, having a chatbot-popup on a restaurants menu-page is a must (because ofc I wanna talk to some fuckin LLM wrapper about the fries they sell!!!), but a goddamn page denoting the places address and telephone number is nowhere to be found.

https://idlewords.com/talks/website_obesity.htm

This talk was given over a decade ago, and its takeaways are as relevant today as thy were back then, and in fact maybe even more so.


> Somehow everyone accepts

Everyone did accept that because when you needed information from a page that pulls that shit, you don't have a choice, and when you did have a choice, all the others did it too.

Nowadays people just ask ChatGPT for the information they need so they don't have to visit those awful sites anymore.


Some of the stuff we have been adding since then is GOOD though.

Some examples:

We now have to accommodate all types of user agents, and we do that very well.

We now have complex navigation menus that cannot be accessible without JavaScript, and we do that very well.

Our image elements can now have lots of attributes that add a bit of weight but improve the experience a lot.

Etc.

Also, things are improving/self-correcting. I saw a listing the other day for senior dev with really good knowledge of the vanilla stuff. The company wants to cut down on the use of their FE framework of choice.

I cannot remember seeing listings like that in 2020 or 2021.

PS.

I did not mean this reply as a counterpoint.

What I meant to say is, even if we leave aside the SPAs that should not be SPAs, we see the problem in simple document pages too. We have been adding lots of stuff there too. Some is good but some is bad.


> We now have to accommodate all types of user agents, and we do that very well.

Simple websites don't even care about the UA.

> We now have complex navigation menus that cannot be accessible without JavaScript, and we do that very well.

Is there an actual menu which is more than a tree? Because a dir element that gets rendered by the UA into native menu controls would be just so much better.


Websites do care about the UA. They don’t care, at least most don’t care, about the User-Agent string. That is different.

About an element that gets rendered into native menu controls, I am not sure. I haven’t been following closely for the last two or three years. But that seems like a good candidate for a native element. 9 out 10 websites need it.


> That's great, but it's always just one agency, or one very local bit of government.

Transitioning every system wholesale at once, is not gonna happen.

I rather have our governents and agencies do it step by step than not at all.


It won’t but it creates a sense of urgency.


Not exactly the best conditions for making good and measured choices, I'd prefer if we didn't add more urgency than what most of us (Europeans) feel already. Everyone already have it on their mind when making purchasing decisions now, no need to also make those people do rash decisions.


The reason Europeans feel the urgency is because of rigid minders and failure to act at least 10-15 years ago. So now it’s ok to bite the bullet a bit. It’s a lesson for the next time.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: