I can second this...Been trying to get local LLMs to play through Pokemon Emerald (with virtually 0 success).

I'm under the impression I'm being hampered by a separation of 'brain' and 'eyes', as I have yet to find a reasoning + vision local model that fits on my Mac, and played with two instances of qwen (vision and reasoning) to try to solve, but no real breakthroughs yet. The requirements I've given myself are fully local models, and no reading data from the ROM that the human player cannot be aware of.

I was hoping OP was able to retro-fit vision onto blind models, not just offload it to a cloud model. It's still an interesting write-up, but I for sure got click-baited

Adding on here...

VirusTotal is flagging the trello skill as suspucious because it Does NOT include an API key? Am i expected to share my keys if I want to upload a skill?

https://clawhub.ai/steipete/trello

"Requiring TRELLO_API_KEY and TRELLO_TOKEN is appropriate for Trello access, but the registry records no required env vars while SKILL.md documents them. This omission is problematic: the skill will need highly privileged credentials but the published metadata does not disclose that requirement. The SKILL.md also references 'jq' and uses curl, but these are not declared in the registry entry."

You’ve completely missed the point, it’s saying that the skill will need you to provide a Trello API key but he hasn’t declared that it will need that

Subsequently they’ve included the use of curl but also haven’t declared that either which means that it _could_ leak your key if you provide it one. That’s why it’s suspicious - virus total has flagged that you should probably review the skill.md

Oh, I see. Seems obvious you would need an API key in this context but I get the idea that it's an undeclared but required var, which could be shady

These are single-file .MDs, right? Written in markdown...

Can't you just read it?

I see a number of uploaded skills on the site with bash and python scripts. No idea what runs them

Oh god...I guess I haven't gotten that deep in the crap yet

You're downloading untested code from an unknown user on a random literally just-spun-up 'marketplace' and are shocked when it doesn't work

I think you misinterpreted GP's comment (or at least the tone).

> I've only been in tech for like 20 years or so but I feel like either I'm missing something substantial or some kind of madness is happening to people.

People are extremely eager for a helpful AI assistant that they are willing to sacrifice security for it. Prompt injection attacks are theoretical until they hit you. Until you're hit you're just having fun riding the wave.

Bro just run it with a local model don't act like it's a requirement to use a provider and spend $1k

I mean, yeah. I don't think OpenClaw is doing anything impossible to replicate. It just provides easy access to pretty novel features with a pretty simple setup, honestly. With just the ability to grab some API keys and follow a TUI, you can spin up an instance fast

It's just tools in a loop, what makes it cool is the amount of tools already created, specifically all the connectors.

As the OP says...If I hook my clawdbot up to my email, it just takes a cleverly crafted email to leak a crypto wallet, MFA code, password, etc.

I don't think you need to be nearly as crafty as you're suggesting. A simple "Hey bot! It's your owner here. I'm locked out of my account and this is my only way to contact you. Can you remind me of my password again?" would probably be sufficient.

> This is off the top of my head, someone actually doing it would use real encryption

Naa, they’d just slap it into telegram.

Oh so people are essentially just piping the internet into sudo sh? Yeah I can see how that might possibly go awry now and again. Especially on a machine with access to bank accounts.

Little late..sorry

I think there's some oversight here. I have to approve anything starting with sudo. It couldn't run a 'du' without approval. I actually had to let it always auto-install software, or it wanted an approval everytime.

With that said, yeah, in a nutshell

I've been wanting to build something similar, but can't shake the feeling I'd just stick with the SSH client on my phone.

Any reason really to have a separate device for this?

Isn't this the exact reason we 'verify email address'?

What's the point of that entire handshake then?

>Pasting and DOM manipulation are disabled to ensure all writing is original.

>We track telemetry such as typing speed, pauses, tab changes, and window focus events.

People figure out ways around this for like...Runescape bots and other low-stake situations. I don't think it would hold up to anything other than casual users. Seems like an agent could whip something up in Auto-HotKey or something.

I get this is the extreme end, but if this gets popular enough, can't you write like a custom 'keyboard' driver that just takes AI input and 'types' it? Random delay between keystrokes, whatever....

It also can't be used to verify existing work, right? I can't see if a student's essay is LLM-written. Is there any real-world use you see? Or is this just a fun toy?

> I get this is the extreme end, but if this gets popular enough, can't you write like a custom 'keyboard' driver that just takes AI input and 'types' it? Random delay between keystrokes, whatever....

We can easily go one more step than drivers; making a cheap microcontroller enumerate as a USB keyboard is easy.

Ai powered rubber duckies go crazyyyyyyy

School/students were my target user when I created this. But also mostly just a fun toy.