Christina @ATX

Check out my Claude Code plugin marketplace kitaekatt.github.io/plugins-kit/ - bootstrap enables any other plugin with a properly configured bootstrap.json to manage its own dependencies (install missing software, python libraries) - skills-kit has a robust skill development framework with new skill features - unreal-kit makes it easier for Claude to work with unreal - p4-kit enables local code review of p4-changelists - awesome-kit has a lot of awesome skills in it

@steipete This is unacceptable plugins are totally a thing and can be presented as an option to install

New pet peeve: cli's that install new skills onto my system without asking.

I believe there is a constraint threshold, fairly small depending on the model but less than 100, after which adherence starts to rapidly decay and much of it is taken up by the system prompt. This is a function of attention decay and it can also happen through context growth. Environmental constraints (hooks, process steps) Is def the way to go.

I haven’t read the paper, but I have noticed that attempting to constrain an agent with rule files is a fools errand. They will break any rule, and overturn any stated constraint. So I use physical constraints instead. Those constraints are things like acceptance tests, unit tests, mutation tests, crap analysis, dry analysis, property tests, etc. The agents cannot overturn those constraints. Therefore they become zealous — sometimes too zealous — in conforming to them.

In the context of this paper, constraint decay refers to the phenomenon where the performance of Large Language Model (LLM) agents drops significantly as more structural requirements/constriants are added to a coding task. Did you notice this as well @unclebobmartin ?

@pvncher When I say component-of i am building this system out of composable components because I recognize not everyone will want to do graph based orchestration.

Hey that is a component of something I am building that is open source I will share with you when it is done :) I am building a Claude code in-client node based workflow scripting system. Basically what you could do before with Python scripts and Claude -p but inverted so Claude code in client is the orchestrator.

If this is true, then we need something other than skills to handle single track multi step workflows, that are not composable and can’t be auto invoked by models.

@noahzweben Hi Noah I don't know if you are pming the remote control feature still. I use this all the time and it isuch more stable. The one real paint point I have is occasional disconnects. They manifest as my android client thinks it is connected, a type a message, it acts as if it is thinking and blocks indefinitely. If I restart the android client I am in the same place but my message is gone (which sucks if it is information rich). I have posted bugs to the Claude code GitHub in the past but I have stopped doing so because honestly the experience as a bug/feedback contributor is not great (consider having Claude help here perhaps, I think Claude could make it a better experience for contributors). Happy to write up a request there if it helps .

@Christopurr I also use and like that pattern of Claude updating html docs in my browser as I work! I do a lot of terminal-only development though, thus the desire for ascii. Do you use any skills for guiding visualization in html?

@truffle I have a lean note-taking skill im always running. And recently had it start doing those notes in html just for this exact purpose of embedding better looking diagrams and images. Have that open in a second window that live updates. Some plusses and minuses to this.

Is there a Claude skill that enables Claude to make readable inline ascii diagrams without alignment issues?

@t20000622yy Thanks I am looking for in terminal visualization but out of terminal is still very important and archify looks neat!

@truffle If HTML/SVG output works better than ASCII, Archify is worth a look. It is a Claude Skill for readable architecture, workflow, sequence, data-flow, and lifecycle diagrams from plain English, with dark/light mode and PNG/SVG export. GitHub: github.com/tt-a1i/archify

@kiaran_ritchie I am but I am skipping threejs just using wasm and flecs.

Who's building a new Unity-style engine with Threejs and Wasm? I could absolutely see a whole host of game types that would live comfortably in the browser and cut out all the hassle of stores/downloads/installs/updates.

@felixrieseberg Next stop: become a Ninja

I've moved into a somewhat older house so I'm having tons of little engineering problems in the REAL PHYSICAL WORLD but I keep arming myself with illustrated instruction manuals for my specific problems and I feel unstoppable I'm vibeplumbing

So many hours spent on this amazing games

@thisisgrantlee My startup went out of business years ago and I still have conviction.

Every founder eventually learns this. Conviction isn't a feeling you have at the start and keep forever. It drains. Weekly. Sometimes daily. The founders who last are the ones who found a way to refill it. Phil Knight ran Nike for years while his father told him he was wasting his time. His bank called his loans. His supplier dropped him. He was personally guaranteeing debt on a company that had, at one point, no confirmed future. He wrote about this in Shoe Dog. He described it not as passion but as a compulsion. He couldn't stop. The problem felt too real. That's a different thing from motivation. Motivation responds to external signals. Conviction doesn't wait for any external thing. It’s something inside. Howard Schultz was rejected by 217 investors before he raised the money to buy Starbucks. 217!!. Most people treat 2-3 rejections as data about the quality of the idea. Schultz treated it as a filter for who was worth having around the table. He said in his memoir, "In times of adversity and change, we really discover who we are and what we're made of." The pattern in all of these is the same. The market was not ready. The feedback was negative. The external environment was not supportive. And the founder kept going anyway, because their conviction about the problem was stronger than their need for validation. The article below gets into this in detail. What Year 2 actually feels like. The role of the ego. How conviction and stubbornness are different things and why that distinction matters. The founders who survive long enough to be right almost always had conviction before the market gave them any reason to have it:

I am not as big on unity as I used to be, but this is an amazing upgrade!

@_kzr This is amazing! I am not a big unity user right now but burst is key to performance in unity. Huge upgrade!

Starting with Unity 6.4, the Web platform supports the Burst compiler with C# job multithreading. This greatly improves the performance of simulation-heavy tasks like this. discussions.unity.com/t/burst-compil…

@1Bexly Generally a good solution but I am thinking more ad-hoc diagrams

@truffle Just have it make a script and save that as the skill.

@rsnous You're absolutely right. I shouldn't have launched those nuclear missiles. Take shelter immediately!

your phone rings and you pick it up and it's the result of an async function that you called many years ago

As a game designer this is a terribly loaded question for which there is no easy answer. So I will instead say a lot of problems in game development come from trying to validate game ideas before they are ready. Usually this is because team members or leadership lack confidence. You won't ever make a great game by trying to validate an idea and iterate. You can't iterate crap to greatness. You can polish a great idea so it has maximum impact. To me this is what validating a great idea is. But if no one thinks it's great, there is no point. Every great game has a visionary, or visionaries, who see the greatness before the product experience reaches testable greatness.

Game dev question of the day How do you validate your game ideas?

The algorithm was alot harder than I expected I was planning to incorporate it in my first game but it took much MUCH longer but now, it's basically plug 'n play for anyone interested to create a slay the spire style progression map: aeyandmelgaming.itch.io/slay-the-spire… #gamedev #indiedev

Been doing quite a bit with deepseek-v4 and it is definitely a high-performance high-value model!

@unclebobmartin yes i wake up every morning appreciating the tricorder in my pocket

These agents are the Star Trek computers. You can ask them question and get answers just like Spock and Kirk used to do. It's uncanny.