Ben Freed

hi frens! I published my Claude Code workflow system as a package called Kata! github.com/codevibesmatte… This system integrates fully with CC's native task system, no beads (though its an amazing project) or other external task managers needed. Also, it relies heavily on hooks and has a pretty ingenious stop hook mechanism that is fully flexible and changes based on what mode you're in. The number and type of modes and individual mode instructions are fully customizable and the package has examples and a good set of starter modes. I'll post more when I get a chance but the main reason i built this is because a lot of systems felt too heavy handed or black-boxy for my liking. Anyway check it out and let me know what you think!

@dillon_mulroy no, not those artifacts

🤨

@threepointone Always.

is there interest in a 4k+ word deep dive in building reliable agent loops (on cloudflare and elsewhere) writing down what I've done for building agents resilient to catastrophic failures on clients/servers/inference (with zero user code) and I need to get it out of my brain

Who's gonna be the first big platform built for this feature???

@dhh is there a way to upgrade omarchy without breaking my hy3 setup??

@sinasanm A banger post for the history books

"i never wanted a company i just wanted a home"

@kunchenguid Yeah and I've seen it using memory to track task progress...

pro tip: disable Claude Code "auto memory" to improve its quality you can do so by typing "/memory", select auto memory and turn it off. this can also be done with setting env var CLAUDE_CODE_DISABLE_AUTO_MEMORY=1 why? - i've seen so many times that stale memory caused the agent to make bad decisions. the information it puts into those memory files are inherently stale in a constantly-evolving project, and the agent would often trust the memory instead of looking up latest information - it stores memory into a claude-only location that other agents don't share. this is anthropic deliberately trying to create vendor lock-in. as a user, it's much better to disable "auto memory" and make claude store memory into standard locations such as AGENTS.md or AGENTS.local.md that is agent-agnostic and portable

@sinasanm @Cloudflare Pushing the frontier with a single convo

encouraged them to list on @Cloudflare workers AI and AI gateway to go viral on X lol

just met someone at a wework in tokyo who works at @cohere a Canadian open weight foundation model company. surprisingly low profile but their latest models seem super legit. apparently their top models can be hosted on just 2 H100s !

@verycracked North Jersey

Alright nyc folks, let’s say hypothetically I move to nyc, should I be looking at Queens or Brooklyn? We’re a family of three and a large dog.

@ipla03 Certainly you can get a first pass?

@ivebenfreed Hard to automate since this process heavily includes my own judgement

Now that codex writes a lot of the code i ship and i find myself reading through a lot of code that i have not written myself i think that code quality is more important than ever I often find myself asking codex for changes to improve the organization and splitting of the code it produced so it's easier for me to read through it and make sure that it does what i want

@jonathan_wilke Because that's the fastest way to be close to the file system. When you put it in a browser you have to solve a lot more problems.

I don't get the hype around CLI coding tools like Claude Code. Human-computer interaction evolved past the terminal 30 years ago for a reason. UIs won. Why are we regressing?

@mattpocockuk This is the unlock

Playing around with the idea of decision maps 1. Figure out the 'frontier' of decisions with a short grilling session 2. Fan out to multiple grilling/prototyping/research sessions, uncovering the fog of war as you go Here I go, grilling on three aspects of a huge build:

@sinasanm Exactly

To make sure I understand, you mean in this context, let's say if a user has 5 sandboxes and there's a headless claudem command in each running some instance of claude code and the agents want to log in to some tool within that sandbox? If that's how you mean, then yes. It's possible with what I'm building. I want to make sure that these sandboxes are very easily configurable, so that no matter how many you launch, they can have the right tools, right skills, right access, and etc

been working on an open-source self-hosted sandbox sdk github.com/sinameraji/sbx purpose: run as many parallel agents as possible on a hardware (mac mini, linux server etc.) and give each agent a dedicated docker container + harness and have full visibility over agent cost and logs. work in progress (pre MVP) but building in public. uses: - @opencode as harness (makes it easy to use whatever else tho) - @OpenRouter - @mastra

Right I'm thinking about the logins for the agents themselves like Claude code I set up multiple home directories that I route the launch commands through that lets me use multiple accounts without having to fiddle around with logging in and logging out could be useful concept in this

@ivebenfreed there's no auth, u just run it locally on whatever device

@robinebers @udiWertheimer Do this all the time

@udiWertheimer recently had that conversation in some X post and was told I have skill issue for telling people not to use the 1M models not my job to educate them but holy shit

never do this kids

Bro... Claude code team is red teaming in the system message and we don't even know???

@zebassembly @dillon_mulroy Stealing this

what about a knowledgebase platform built using Durable Objects+ CRDTs for realtime edits, artifacts to make it git clonable, MDX + RSC so you can build custom components into the wiki platform itself, and then Dynamic Workers to isolate the custom components could be agentic af

when the system prompt leaks thru

@fitchmultz #clankermaintenance

@codevibesmatter Bro is feeling bogged down by the number of tasks I'm throwing at it. Good life advice for a human though.

GPT-5.5 is tired today. "Reflecting on rest needs I'm thinking I really need to take some time to rest, especially around my focus and drift. It’s easy to get caught up in tasks and forget that downtime is essential for rejuvenation. Balancing my work and relaxation can help me be more effective in everything I do. I wonder how others manage their rest and stay productive. It's definitely something I want to improve."