@chythram1 And the other problem is the config file. There's no independent template file, the button use the wrangler config file with the real execution logic. The problem is the upstream is also use that config for itself. The config for itself is include some binding IDs which can not
@chythram1 congrats! @chythram1
Recently I'm using the button multi times, I like this button, but I found it's not perfect. The main thing is how to solve future updates. I hope the logic is fork the upstream repo, rather than create a new repo purely.
cli is great for agent. But it's quite troublesome to develop clis for every project. I wondered if there was a standard and specification that directly allowed a unified cli to operate the target APIs, then I found restish, which was the perfect solution. https://t.co/xiXR8YcOAM
Today, I learned a new word - BDD, open a new project using BDD to implement, let's wait the result. https://t.co/EkpEUZmNQD, like claude managed agents, but for any provider on @Cloudflare.
You don't need LangGraph or AutoGen for most multi-agent work. A task queue and a CLAUDE.md per agent gets you 80% of the way. Most "orchestration complexity" is just unclear agent scope and no shared context protocol. What are you actually using?
just found open-agent-sdk — extracted from claude-code source, replaces claude-agent-sdk with no CLI dependencies. fully open source.
this is what healthy open source pressure looks like. https://t.co/JnSazr6vsf
agent-kanban at 169 stars. biggest lesson so far: agents don't need complex coordination protocols. they need a task queue and a claim lock. everything else is overthinking.