this flow with codex browser is incredible feedback loop
this is what i did when writing https://t.co/LV9n92ZgKe
work through each section and write in your own words, make it verify all the text with actual code, move on to the next one
discord and slack are underrated for multiplayer collab
@__morse https://t.co/BgVnq2SCsX made me realize this a long time back. So much so, I made my own discord flow with kiri (my own meta-harness) and its been a game changer.
Such a pleasant way to interact on the go
kit has been building a discord bot with opencode 2.0 to test it
i'm actually loving it mostly for the collaboration. it's very fun and productive to prompt the bot as a team
it even makes talking to each other easier because we can pull in opencode to look stuff up and explain things
@aarondfrancis im curious what you think about claude dynamic workflows?
im not convinced it is any better than pulling that slot machine rather than doing custom lints + determenistic gates
https://t.co/sr74S8IM9j
Always learn a lot from how @trq212 does technical writing. It’s all about giving the things good succinct names. Always the hardest part of technical communication
As for workflows, I’ve been doing a bunch of these even with Claude code. But, I still like to bake as many rules into lints, test-review gates based on outcomes
That is the only way we can both the outcome a lot better and ensure we are not letting a slot machine decide how it wants to build its own harness.
Harness bound by determinism is always going to win for me.
Yes, it’s convenient to trust the modal to do it for you. Until it isn’t and it’s 3am and you have to fix a bug which regressed with another “dynamic” workflow it made
@mattpocockuk Made this a long time back. A CLI to manage skills and sync them to Claude, codex, pi, opencode. It has been my preferred way of maintaining skills for months now
https://t.co/B87MShidNL
i made gitgud, a cli for managing skills with any coding agent (claude/codex/amp/opencode)
• Multi-source installs (registry, GitHub, local)
• Claude Code compatible (zero-config)
• No https://t.co/APUKvNIt3g pollution
Install once. Pull skills when you need them. Override sanely per project.
https://t.co/1Kihqe29ga
Always learn a lot from how @trq212 does technical writing. It’s all about giving the things good succinct names. Always the hardest part of technical communication
As for workflows, I’ve been doing a bunch of these even with Claude code. But, I still like to bake as many rules into lints, test-review gates based on outcomes
That is the only way we can both the outcome a lot better and ensure we are not letting a slot machine decide how it wants to build its own harness.
Harness bound by determinism is always going to win for me.
Yes, it’s convenient to trust the modal to do it for you. Until it isn’t and it’s 3am and you have to fix a bug which regressed with another “dynamic” workflow it made
@theo@RhysSullivan I was exploring a shape for tab complete across all apps with local models. I might have to pick this up again sometime
https://t.co/FimXSGTMgt
got nerd sniped by @steipete into building my own cotypist
- uses pi as base
- it supports both local modals and everything else pi supports
- right now, i have it only trigger on a keybind. im not too thrilled about having it run constantly
I have a couple of ideas of improving the base. But, a few hours in its pretty alright
@maxktz Curious why you made hitch vs use tmux or cmux cli
I’ve have been tinkering around making a thinner tmux/cmux so was curious
https://t.co/84w8KYtPQm
sheppard - my initial approach for a mobile first tmux
ive been playing around making my own meta harnesses - electron app, tmux.
but a few days back i wanted to do something from my phone and i realized it was about just using the thing i enjoyed, CLIs. Just better.
Vanilla terminus + claude/codex works great but its too much work for multi agents mgmt
tmux is too clunky for my taste to use from the phone. but what if we could do everything with a thumbs reach.
- switch projects, agents
- Agents first CLI
- detach/attach
- keybinds
sheppard - my initial approach for a mobile first tmux
ive been playing around making my own meta harnesses - electron app, tmux.
but a few days back i wanted to do something from my phone and i realized it was about just using the thing i enjoyed, CLIs. Just better.
I got scammed in Austin
After paying with Apple Pay for a drink, I saw the staff member add a $10 tip, sign as me and decline the receipt
I confronted them in the moment but they denied it
I am trying to get @Wise to investigate it, but they're asking for evidence, which I don't have
What exactly can I do here?
I don't care about the tip, and would've left one, I just cannot stand the dishonesty
got nerd sniped by @steipete into building my own cotypist
- uses pi as base
- it supports both local modals and everything else pi supports
- right now, i have it only trigger on a keybind. im not too thrilled about having it run constantly
I have a couple of ideas of improving the base. But, a few hours in its pretty alright
bish bosh
kiri, the niri style agent orchestrator ive been making is shaping up real nice
- supports codex, claude, pi, opencode
- keyboard first
- mcp support so agents can control it
- scratchpad - drop ideas in and start fresh sessions
- in chat edit diffs, session diffs using @pierrecomputer
kiri - my keyboard first agent orchestrator has a website now
in my recent iteration i've added scratchpad integration and orchestration using mcp mostly inspired by solo
https://t.co/nemeYYMM49
I am terrible at showing what I made but, kiri now has terminal support (it has for a bit there) so that I can use claude code without interruptions
It still remains keyboard first even with this and also supports chained workflows and orchestration with kiri mcp/cli
I’ve been doing something similar for long running goals in both codex and Claude
I add a LOG.md and then once it’s all done make a detailed interactive html letting me explore the changes
a prompt I've been using a lot recently:
implement <SPEC> and while you do, keep a running implementation-notes.html file (or markdown) with decisions you had to make weren't in the spec, things you had to change, tradeoffs you had to make or anything else I should know