cat-whisperer

I think the blog is slightly incomplete, but I would like you to fill in the gaps. What do you think?

Do you think software is just conservation of search?

your website uses the exact tech as the openAI website that they had for Astra. Interesting

We might be using decisions models (Jev) in there somewhere ;D

I literally wanted to by Iphone 13 mini because of it's size, but the camera sucks so I had to reconsider.

I agree that there's poor taste :)

It's interesting that they haven't published any benchmarks yet.

Why doesn't apple have a setting for that, or am I doing something wrong?

I think everyone should take some time out of their lives to learn parser combinators; they give you a good grasp of how languages work.

After building a few parsers with `nom`, my appreciation for English grew, and I took an interest in learning all the nuances of the language I speak.

I really like the concept of dependent types, all type systems that I use rust or haskell have a very limited set of dependent types, but I do like the things that it let's you do

you get me!

that’s fair; I was suggesting something even more drastic than this https://github.com/uji-labs/ito/issues/4

maybe it’s a bad idea :X

it does, but I am a fan of the composability that ratatui has, and I wanted to copy it over…

I trust you would make that happen! And I would be right there to use it in my next project

feel free to raise issues if anything feels off. The plan is to make it the rice harness :)

*oh! I meant the right harness for you

the discord jam was awesome! kudos to these guys!!

I doubt that. I don't think AI are trained to find elegant solutions, they are trained to find solutions. if you look at the things that OpenAI published, so seem to be micro improvements riddled with chaos. They are better solution, but they aren't drastic.

I have noticed this with autoresearch too, the solutions and improvments that you get are incremental and not groundbreaking. Humans, have a way find elegant solutions and also seem to solve that problem. Then again, beauty is in the eye of the beholder, and I might be wrong.

+1 to that. I didn't even feel like doing anything I was just there flabbergasted by the background

I agree that the next session wants conclusions, not the whole transcript. What I've seen working with people on this is that a summary only has what the summarizer prompt asked for, and what they need from an old session changes depending on what they're doing when they go back to it. With the full session still around, they can just ask it again.

Porting is also cheap, since it's just a format conversion. Getting a new session back up to speed from notes usually isn't.

Concurrent sessions is a separate problem, and I agree that a message is the right shape there. Where portability helps is that different harnesses are good at different things, so you can start a task in one and move it to another that's better at the next part, without starting over.

Already merge #53, thanks for the assist.

I'll add support for this. On another note, if you know what the issue is, our core is open and I would love for you to contribute: https://github.com/skillsynchq/txcript

it's interesting to know that archived sessions are stored different as opposed to the codex cli and desktop app session.

keeping tool_results help the agent figure out what is being referred too. But, as you said we also faced the problem of context overhead from reading markdown that's why we made continue into a cli command instead.

the mind meld thing sounds interesting, it sounds like git merge but for the transcripts

Yes the name is a giveaway, that's how we started out. We wanted to make it easy for non-technical people to collaborate with technical ones. git pull does bring in some friction here. Skillsync can be used as a shared skills library out of the box. Once someone updates a skill, it hot-reloads and make the latest version available to everyone instantly

Curious about your /resume skill. You said merge, so are you pulling a colleague's session into one you already have open? We only do the other direction today, where you reopen their session and continue from it. I'd love to know which one your team reaches for more.

we do support antigravity!

we publish a npm package of our core library too: https://github.com/skillsynchq/txcript

you can use that internally for this

isn't that what they did to opencode? when they were using the subs by reverse engineering their APIs

That's a clever idea, and a different direction from where we are right now. Today it only goes one way, from chat into your coding agent.

The case I hit all the time is that an idea strikes on my commute, so I pull out my phone and start brainstorming in the Claude app. By the time I'm home I've done the research and chased down the tangents in that conversation. I open my laptop, use Skillsync to continue it in Claude Code, and start on the implementation.

We do that, and it's really the point of the portability work. Once every harness's sessions live in one schema, you can manage and search all of them in one place.

The markdown journal is a fair way to do it, and a popular one. It has a couple of limits though.

A journal comes out of a prompt, so it records one way of looking at the session. Whatever that prompt didn't ask about is gone, and you usually find out which detail you needed much later. But the raw session still has it.

Being able to unify the schema lets you do things notes can't. You can continue a teammate's session on your own machine, or pick up a claude.ai conversation in Claude Code or Codex. The new agent gets the transcript: all the decisions in the trajectory, not a summary of it.

In some cases, harnesses try to obfuscate reasoning, but almost everything goes through.