@nexxeln @AdamAugustinsky How is it the first time I’m hearing about this? Do you guys have some resource explaining it? Sounds very interesting
English
Antoine
19 posts
Antoine
@antoinecojp
Vibe coding in Japan, shipped nothing to this day
Tokyo-to, Japan Katılım Ekim 2023
25 Takip Edilen1 Takipçiler
turns out codebase architecture was always more important than AGENTS.md
nexxel@nexxeln
i’m starting to think agent-friendly codebases are more about constraints migrating parts of opencode to effect has made agent-written code noticeably less cursed good architecture boxes agents into writing specific, constrained code fewer ways to go wrong
English
@oljimenez @mattpocockuk @hhg2288 Problem is that some skills should be opt-in by agent automatically. But yeah a lot of skills could just be commands. It’s just overhead to think about which should be which.
English
@mattpocockuk @hhg2288 But that won't make skills like a little useless?
Why not use commands (because they aren't opt in by the agent, only humans can call them) + rules for forcing the agent to follow the codebase patterns.
I don't understand why I should let my agent opt in a skill.
English
quick question @mattpocockuk
what's the difference between having a skill like /grill-me and include it in your AGENTS.md file?
English
@MichaelArnaldi @EffectTS_ Yeah that makes a lot of sense, especially for effect. Tho maybe we don’t want to end up cloning all libraries that we use and sometimes skills can be useful for different kind of libraries, maybe CLIs or for best practices for tools.
English
@antoinecojp @EffectTS_ It’s because skills are genuinely useless, LLMs are good at consuming code so you should give them code:
Michael Arnaldi@MichaelArnaldi
English
People keep saying @EffectTS_ is so good for coding with LLMs, but I don’t see them shipping any official skills. The docs are not in the node modules either. Should we point to the llms.txt in our AGENTS md and hope for the best?
English
@kitlangton @mattpocockuk Nice! Definitely worth exploring, maybe that usecase could also be solved in a different way, something to think about.
x.com/antoinecojp/st…
Antoine@antoinecojp
@ericbdev @mattpocockuk True, that’s what I usually do but it’s not super convenient. I wish there was some kind of native handoff feature where you can send back the summary to the original thread. Or maybe native sub-threads that would be used precisely for that.
English
@antoinecojp @mattpocockuk I’ll explore adding this feature to opencode 🤔
English
A flow I just tried and LOVED:
1. /grill-with-docs, talking about a new bit of UI
2. Asks me a question I can't answer unless I prototype
3. /prototype
4. Iterate on the prototype, burning tokens freely until we get a good spot
5. /rewind to the question, and select 'summarize' (Claude Code feature), saying 'summarize what we learned from prototyping'
6. Continue the grilling session, retaining the prototype
Smoooooooth
English
@ericbdev @mattpocockuk True, that’s what I usually do but it’s not super convenient. I wish there was some kind of native handoff feature where you can send back the summary to the original thread. Or maybe native sub-threads that would be used precisely for that.
English
@antoinecojp @mattpocockuk The flow seems similar to forking a session, then either summarizing or compacting that fork and re-ingesting back into the 'parent' session.
English
@antoinecojp what type of naming convention do you use for your design system?
English
@shadcn It’s just different. We match the design tokens used by our design team. It’s not structured the same way. For example we have interactive tokens, typography ones etc.
Maybe a mapping config on shadcn side could map each token and use the right classes when importing components.
English
@shadcn Honestly not sure if possible to solve but I often want to use shadcn components with my own company design system. The problem is that they use different naming for tokens. Having a mix of both shadcn tokens and personal design system is a bit dirty and tricky to deal with.
English
English
@refactorordie opencode v2 will contait `import from "@opencode-ai/core/effect"
English
Someone port this to Effect, please! I'd love to build with this
fks@FredKSchott
Introducing Flue — The First Agent Harness Framework Flue is a TypeScript framework for building the next generation of agents, designed around a built-in agent harness. Flue is like Claude Code, but 100% headless and programmable. There's no baked in assumption like requiring a human operator to function. No TUI. No GUI. Just TypeScript. But using Flue feels like using Claude Code. The agents you build act autonomously to solve problems and complete tasks. They require very little code to run. Most of the "logic" lives in Markdown: skills and context and AGENTS.md. Flue is like Astro or Next.js for agents (not surprising, given my background 🙃). It's not another AI SDK. It's a proper runtime-agnostic framework. Write once, build, and deploy your agents anywhere (Node.js, Cloudflare, GitHub Actions, GitLab CI/CD, etc). We originally built Flue to power AI workflows inside of the Astro GitHub repo. But then @_bgiori got his hands on it, and we realized that every agent needs a framework like Flue, not just us. Check it out! It's early, but I'm curious to hear what people think. Are agents ready for their library -> framework moment?
English
@UtkarshUsername @jullerino Thanks, was confused because I was used to CMD + « . » on opencode desktop. Shift tab seems like an accessibility thing in a desktop app.
English
@iamsahaj_xyz I don’t want to pay for two different subscriptions. And as you said v0 is not a replacement for claude/cursor.
English
Ok, sorry. ClawdBot is a war machine. Just merge my first pull request from Telegram. Feels like magic.
Sorry @steipete you was 100% right
English
i come back to sonnet for most task and i must say that it's do pretty good job for 95% of the task, like opus
can't wait for sonnet 5 💃
English
@mattpocockuk That makes sense, you give it one more way to debug its code and even check the UI looks correct.
English
Playwright MCP + Ralph is incredible
Article coming soon
English
Tried Ralph for the first time today with @opencode and following @mattpocockuk method. Iterating on it I could get some very good results.
But feeling its something still early that deserves a clean framework and "quick start" way of doing it.
English