anonfunction

There's this: https://platform.kimi.ai/docs/guide/kimi-k3-quickstart

This is something I've been working on exposing to AI labs through my startup LatentEvals[1], and found similar results in other industries from lending to insurance claims.

Happy to share some sample reports if anyone is interested!

1. https://www.latentevals.com/

There are multiple "sandbox backends"[1] available besides Vercel currently. I'm sure more will follow, and it's open-source so you could always implement your own.

1. https://eve.dev/docs/sandbox#backends

One of the preeminent Zig software projects, Tigerbeetle[1], has a great website[2] and markdown file[3] for what they call TigerStyle.

It's honestly a great read as a human but also I packaged it as a skill and saw coding agent improvement, especially around memory usage.

1. https://tigerbeetle.com/

2. https://tigerstyle.dev/

3. https://github.com/tigerbeetle/tigerbeetle/blob/main/docs/TI...

Love these types of projects, really useful and helpful. Sharing with my daughter!

I'm the creator of hty, as part of my work for building custom eval suites at LatentEvals.com I found that AI agents struggled with interactive CLI / TUIs and would resort to setting flags that may be outdated or not exist at all.

I needed a way for AI agents to use all the software that humans do, things like htop, k9s, create-next-app, and even nethack.

How it works: hty wraps any interactive cli in a persistent PTY session. Your agent "sees" the rendered terminal the way you do and types the way you would. I've had a lot of success with agents being able to get through interactive npx setups, play terminal games, and create logical commits from a bunch of work in an long claude code session by using git add -p for interactively patching the git stage.

It's available to use as a simple skill so it works for almost all agents, but you can also use it as the human via the CLI. It's written in zig and built on top of libghostty.

I've been doing bias and misaligned behavior research, creating custom private eval suites to test and compare models. Claude Opus 4.7 is heavily biased and presents clear regulatory and reputational risk.

It seems the initial product footprint tries to sidestep this problem by not giving the agents control on who to lend to or which applications to approve. Even so I think it's quite an optimistic read on their end. Happy to share reports to anyone who's interested (montana@latentevals.com), especially if you work at a frontier model lab and are interested in plugging my evals into your RL systems!

Author here, I built this because my agents could not use TUI or interactive CLI software. Think puppeteer for TUIs.

If you've ever asked Claude Code, codex, opencode, etc... to make logical commits from a bunch of changes you might have seen them struggle with moving files around, deleting files when a simple `git add -p` would have let them work through the changes interactivly. ht (the binary name) allows them to spawn their own terminal and drive it similar to puppeteer, sending keys, viewing a snapshot of the terminal and making a decision.

Some other use cases are also solved from CI/CD to making nice demos (it can export images or the format used by asciicast for making videos of the session.

It's powered by libghostty so it has support for all the stuff a modern terminal can do (full color, mouse clicks, etc...) and I've had a lot of fun letting claude run wild and even play games like nethack.

Happy to answer questions, and of course let me know if you run into any issues!

Fourth thing in a row they've announced that I wanted to try and couldn't.

Previous comment with the prior 3: https://news.ycombinator.com/item?id=47794419

Start designing at claude.ai/design.

That link is redirecting me to https://claude.ai/404, anyone else?

Thanks, I was already on the latest claude code, I just restarted it and now it's showing 4.7 and xhigh.

xhigh was mentioned in the release post, it's the new default and between high and max.

That sets it to opus 4:

/model claude-opus-4.7 ⎿ Model 'claude-opus-4.7' not found

/model claude-opus-4-7 ⎿ Set model to Opus 4

/model ⎿ Set model to Opus 4.6 (1M context) (default)

Seems they jumped the gun releasing this without a claude code update?

     /model claude-opus-4.7
      ⎿  Model 'claude-opus-4.7' not found

It seems they nerf it, then release a new version with previous power. So they can do this forever without actually making another step function model release.

Same, if we're punished for being on the highest tier... what is anthropic even doing.

     /model claude-opus-4.7
      ⎿  Model 'claude-opus-4.7' not found

Related features that were announced I have yet to be able to use:

    $ claude --enable-auto-mode 
    auto mode is unavailable for your plan

    $ claude
    /memory 
    Auto-dream: on · /dream to run
    Unknown skill: dream

A little off topic, but did Anthropic distill from an older OpenAI model? All the sudden over the last few days I'm getting a ton of em dashes in claude code responses!

It gets really fun when you prompt them to update their BOT.md, also they get access to previous run results so they continually "learn" from mistakes or investigate changes.

This is very cool, I just wonder where they get the crossing data!

Thanks for sharing, I think it's kind of a different type of system although a lot of similarities for sure.

I made botctl explicitly to not have to talk to my agents, I wanted them running autonomously based on their BOT.md file's prompt / goals / skills.

I'll take a counterpoint, coding has historically not been cheap. Software engineers have been one of the highest paid professions for a long time. Personally while working a full time job, raising a family and trying to have some semblance of a social life my open source contributions fell off a cliff until recently with the popularization of coding agents. I've created more projects and software in the last 12 months than the past 10 years combined. Not to say a lot of it wasn't total slop or provided little utility, but it's been a fun and exciting time.

Another interesting point is that until recently most average people thought "code" was out of reach or they didn't have time / energy to learn it. My mother made a webapp with the help of claude code the other day to generate books, which she thought of and completed in the course of 3 days all the while learning about terminals, localhost, ports, APIs and more.

I think there's also something about being able to make it exactly how you want. For example I didn't like openclaw, I felt it was extremely over-engineered and I honestly couldn't even get it running on my first attempt. So I made botctl to be a generalized version where it doesn't rely on complex setup and have every bell and whistle, you just install it, create a directory with a BOT.md file and off it goes.

Here's two popular open-source projects that have combined 200k+ stars and 1,300+ contributors, with slightly different goals:

- https://github.com/anomalyco/opencodeagents

- https://github.com/OpenHands/OpenHands

Thanks for the feedback, I'll do that! I think different screens and system settings affect this a lot, on all the devices I've looked at the site on I felt it was quite comfortable to read, not to discount your experience, I'll keep that in mind for the future as well.

So I'm the creator of botctl which I created to satisfy my personal need for running many agents on a cron style loop. The whole premise is you don't need to be there to chat with the agent for it to do it's job.

I'm definitely not A/B testing anything, it just cycles through a couple titles I thought conveyed the project well. The sub-title directly under it explains it pretty clearly, would you agree with that?

Manage persistent AI bots with a terminal dashboard, web UI, and declarative configuration.

There's also an interactive emulation of the TUI directly next to both titles in the hero.

Take your upvote, love the old school hn spirit.

Reminds me of when I was speaking at a conference back in the mid 2010's with a presentation titled "Join the dark side of APIs" or something akin to that, where I showed many of the newly popularized "single page apps (SPAs)" relied on undocumented and public APIs, often without any authorization. Immediately followed by a talk by a business type guy on API copyright or something or other.

That’s actually a great find, this page[1] from them I found seems to have more recent data than I’ve otherwise been able to find. I’ll reach out to them to see if they can provide some data to power my site as I don’t like the current setup of old data and prediction market odds. Thanks!

1. https://insights.windward.ai/

I mean you obviously could, the url is a little harder to remember and it doesn't have crossing data. This was just a small fun project I did, so you're free to do whatever you like. The reason I thought of using polymarket data is I didn't have live ship tracking data which is what I originally intended to use.

Thank you for this comment, I just fixed it[1], don't know why claude code decided to hide it, I actually should have known this requirement and checked!

1. https://github.com/montanaflynn/ishormuzopenyet/commit/70a8c...

Ahh, thanks I'll remedy that now, wasn't intentional I'll blame Claude.