Seed: Minimal, self-modifying agent harness | Hacker News Reader

Seed: Minimal, self-modifying agent harness

github.com

59 pointsgandalfgeek20 comments

I think as many things that are posted here lately there is no *why* attached to the readme. Why would one use this, what is the benefit of this approach? Am I really gonna need my model to build exotic tools around it; or is exec/web_search/web_fetch enough for 90% of the use cases? Is my agent not capable of writing new plugins/tools for pi/opencode?

I agree some why could be interesting. I would add to your list, because IMO this is more of a thought experiment-as-repo than a utility focused tool, what can i see by using this that i cant with a more fully fledged starting point?

If your needs are met otherwise stick to that and move on.

Not everything needs to fit your needs, including the documentation. Bizarre how you seem to think someone else owes you answers to questions you don't seem to want to try and answer yourself.

Since I am here though and find this project actually interesting; I like the idea of systems that offer, overall, the same utility but are implemented to be as minimal as possible. LLMs are exactly that; look at all the utility from some for loops sorting a huge pile of data. So a harness that's also minimal at its core and endlessly self extending just corner cases unlike pi's ponytail skills which ironically packs a ton of opinions into a workflow up front, pitching the minimalism it aspires to in the bin, seems like a Homer car to me.

But I am a hardware engineer biased by experience with minimalism; I got a single data structure (electromagnetism) to manipulate and BOMs that come with hard constraints.

On the contrary, web SaaS devs want to run a business/rocket to the moon moreso than be an engineer. As such git pulling, pip installing the world allows them to focus on their get rich quick with as little labor involved as possible goal front and center.

> Bizarre how you seem to think someone else owes you answers to questions you don't seem to want to try and answer yourself.

This, of course, depends on your goals, but if your goal is for your project to reach more people for whom it is relevant, it is not crazy to provide some basic motivation.

"Why anyone should use this" is the fundamental question that anyone building something with the intention of other people using it should be thinking about. For example, Pi agent provides lots of information on the thinking and motivation behind the design. Even if I don't use the project, I come away learning something interesting.

You're in luck; plenty of basic information as to it's implementation and self extending nature right there in the repo.

It was enough for me to see the value.

In conclusion we merely reasserted "it's relative".

> If your needs are met otherwise stick to that and move on.

Weird to post such a thought in a forum where OP has posted their project for people to look at. I never mentioned any needs, I am saying the readme holds very little value for a project in a space that has 50 different projects/coding harnesses available.

How would I know if this fits my needs based on it saying LLMs can build tools for themselves? That's something they can do with any and every open source harness.

> But I am a hardware engineer biased by experience with minimalism; I got a single data structure (electromagnetism) to manipulate and BOMs that come with hard constraints.

Alrighty.

> On the contrary, web SaaS devs want to run a business/rocket to the moon moreso than be an engineer. As such git pulling, pip installing the world allows them to focus on their get rich quick with as little labor involved as possible goal front and center.

Posting your open source project to a forum such as hn means people will have opinions they want to voice. No need to shit on opinions you disagree with.

This might be useful to me. I have an app where I’m reusing an existing harness. It generally works well, but there are some rough edges where I have to work around the harness. I don’t want to code one from scratch. This may work as sort of a kickstarter, although I think I would prefer sort of a framework where there are mechanisms to plug in my own components rather than create something as fundamental as memory from scratch. Like the plumbing for memory with an API where I can make the storage layer whatever I want would be nice. Anyway, this thing is giving me some ideas.

I have to say I enjoy the brevity and clarity of this idea, given the ease of producing something large and unfocused in the last couple of years.

I hope it stays very simple, and would be interested if the author or anyone else can speak to what kind of complexity arises from using it, rather than bloating out the seed itself with initial complexity.

Ive been thinking about starting stimple with a pi framework for example and letting the tools emerge, but this is so much more lean to start, i wonder if it will be more or less interesting and or useful to start here.

I just wrote about this.

Not this exactly, but how your repository can be your swarm, and essentially be your harness.

Using Gitea/GitHub Actions within your repo and agent runner containers allow you to have agents working in your repos 24/7. Not just that, but in a containerized world you can just link up their brains over some shared storage. Using Antigravity, this means you can share the `brain` and its conversation directory with your agent runners and they can get the entire history of everything.

While building it out it becomes apparent that we are just getting started!

Where did you write about this? I am interested in reading more.

I just submitted it last night on HN, actually. So it should be on my profile in my submissions. It's written more about my experience rather than as a tutorial, but I think it becomes clear how it should work.

For those interested, myself amongst:

- https://nicktrevino.com/your-repository-is-your-swarm.html

I enjoyed it, thank you for sharing and double thanks to Bluestein for surfacing it. Your point about declarative languages is interesting and mirrors my thoughts. Prompting is declarative even.

A nice idea that I've made a few attempts to implement myself, but somehow always end up with more than I intend. I'll just use this to... seed.

This remind me kind of stuff people did back in 2023: https://x.com/SigGravitas/status/1642181498278408193

... Massive Update for Auto-GPT: Code Execution!

Auto-GPT is now able to write it's own code using #gpt4 and execute python scripts!

This allows it to recursively debug, develop and self-improve... ...

Also fun fact: Loveable started as a fairly basic Python script: https://github.com/AntonOsika/gpt-engineer/releases/tag/v0.0...

It looks like you're trying to destroy the world. Would you like some help with that? /Clippy

https://gwern.net/fiction/clippy

I think people underestimate the number of successful growth lanes versus the number of failure lanes.

I don't think these agents are going to do anything but find unique and interesting ways to torture their users.

This is lovely and a reminder of why I used pi as a basis for my own harness. I have a similar idea in LISP that isn’t as minimal, might go all in on eval…

> There is no framework here. The entire frozen layer is seed.py

Slop language in the second paragraph. I refuse to read on.

Yet another LISP in disguise...

Reply on news.ycombinator.com