Michael Ramos

Anthropic code review this, clanker review that ... why don't you shut up and review+annotate your own code.... (yes im a loser who still manually reviews code) Originally inspired by a bunch of feature requests and then seeing @dillon_mulroy tweet a similar cool ux. @plannotator for reviewing plans (primary focus) and code, fully oss. OpenCode, @badlogicgames 's pi.dev, and Claude Code and other clankers

@SIGKITTEN what is ideal ux?

whats a good tool to share codex and cc convos im sure someone had built something decent?

@Yuchenj_UW I think codex should. It’s important for provenance

I noticed something interesting: Claude Code auto-adds itself as a co-author on every git commit. Codex doesn’t. That’s why you see Claude everywhere on GitHub, but not Codex. I wonder why OpenAI is not doing that. Feels like an obvious branding strategy OpenAI is skipping.

@trq212 working with goog & gov agency pre/during covid era, we had to do budget/value accounting for any UX around transformer models & employee workflows (augmented retrieval). Industry wanted to deccouple application/platform/hardware. But AI bolted them all down together again.

an increasingly large part of the job of an engineer is deciding how much compute to spend on a problem

@vincent_koc Worth it

use git worktrees they said

@hichaelmart Tbf output averaged like 300lines of code. I would _think_ strength of types scale with the size of a codebase LLMs have to retain context of without hallucinating - and also steer the llm to always seek type information in active development.

Kinda fascinated how ppl keep trotting out "typed languages are better for codegen" even though we have evidence against it. It *feels* intuitively correct, right? They give better guardrails! But maybe types aren't all that amazing as guardrails for LLMs? dev.to/mame/which-pro…

@kinglycrow love this. I've been toying around with a similar skills-based approach. Often ask: what do I need to know about the changes, and what code does each specific thing touch? (like a QA list) x.com/plannotator/st…

lately i've been pretty interested in tools that help me digest and understand large amounts of changes from agents so i built a little thing 🧵

@dillon_mulroy @kinglycrow 🔥

@kinglycrow @backnotprop you might dig this

realized it might be better to show a video of the visualization, but i also added git diff signals last night!

Still a terminal user? Ever wanted to give detailed feedback on a long message? ... /plannotator-last Works for all agents. Thanks @_mjmeyer for the feature request!

Does anyone have a good agent skillfile that: - Spins up a screen recorder - Runs your app end to end - Edits and annotates the video to show all the core highlights of the feature working Plus if not limited to webapps/GUIs either

I've been using it to capture the plan and annotate it w @plannotator & automate feedback - Claude Desktop supports this now though. No need for external app. @amorriscode if the PermissionRequest(ExitPlanMode) could support approve with comments (additionalContext) that would be nice and you could enable here as well

@backnotprop @amorriscode @felixrieseberg what do you use hooks for? havent found good use cases for them

Claude desktop is one of the most important daily apps, as it combines Cowork (personal assistant ie Anthropic Openclaw) and Claude Desktop (coding). There some minor fixes that 10x's the UX for users, hopefully @felixrieseberg @amorriscode can push out: - notifications seem unreliable to me, especially in claude code desktop, when I'm on Cowork, i've noticed it not giving a notification when cc is done. - shortcuts to navigate between chats, not clicking to select chats - notifications for cowork/cc always bring me to top of convo and I need to manually scroll down each time - /plugins: cowork allows managing plugins on app, but claude code desktop doesn't - overall feature parity with claude code cli

@amorriscode @mattlam_ @felixrieseberg Ayy

@backnotprop @mattlam_ @felixrieseberg Claude Code on desktop supports hooks?

@amorriscode @mattlam_ @felixrieseberg hooks?

@mattlam_ @felixrieseberg What CLI features are you longing for?

For those adopting agentic first development - we need better review surfaces to maintain ownership of our code and what we ship. Any GUI harness has an opportunity to outclass the TUI, yet they are all very focused on chat organization and experience still. Some new harnesses are out of touch in terms of the design that's required for human in the loop: 1/2. Better Plan Review

@benhylak The app is also unusable bc the underlying runtime/cli has state (internal and cross session) leaking.

i'm sorry but codex cli is just unusable. i actually like the model.

@lydiahallie Dope would love for the session info and other context be made available to that context via Env var or similar

if your skill depends on dynamic content, you can embed !`command` in your SKILL.md to inject shell output directly into the prompt Claude Code runs it when the skill is invoked and swaps the placeholder inline, the model only sees the result!

I got $0 and a hackathon rejection after blessing Anthropic with this alpha. github.com/anthropics/cla…

@iamsahaj_xyz What falls short with permissions turned on? Gives you an opportunity to review every change?

I'd enjoy a semi-tab model generate spec / plan with an agent, then hop into the codebase with a great tab model to implement the spec WITH you

@_mjmeyer @plannotator @badlogicgames 0.14.0

@backnotprop @plannotator @badlogicgames epic! good to hear, can't wait 🙌

Underrated feature: annotate any file @plannotator demoed in @badlogicgames pi. works in any harness as slash command or shell (!plannotator annotate file.md)

I think there's a big Twitter dance going on all around this. And I would say your approach is different approach than human-on-top-of code telling the agent what to do (a new form of tab basically) which is what I hear Dax advocating for. QRDSPWIP seemingly centers on a very key artifact (ie plan) derived from context steering. You pitch the imaged process and tell me that it is different from spec-driven development. I tell you, tomato, tomato. The implied evolution of a plan from detailed requirements-vommiting versus the tactical document is a general misalignment of who's doing what and who's on first. I've never had this problem with planning (CC plan prompt has remained largely the same since introduction). I think we are 100% moving into the direction where a clear statement of intent (spec plan tomato tomato), combined with growing agent capability and tighter feedback loops, outperforms the traditional write-then-review cycle or full-ownership cycle.

Yea we still use Md docs for context engineering purposes eg implementing a big thing in small context windows but yeah capital S Specs require “declarative config” and we’re still in the Stone Age doing imperative things (but I do think there is a lot of leverage to be had if you set up the workflow well)

damn this is so good and encapsulates everything I've been seeing/saying in the last few months - a spec that is sufficiently detailed to generate code with a reliable degree of quality is roughly the same length and detail as the code itself - so don't review those things, just review the code at that point, if you care enough about that level of abstraction - unless you're vibing side projects or prototypes (yes, even zero-to-one software), you ABSOLUTELY SHOULD care about the code at that level of abstraction - you need to find SOME way to get more leverage over coding agents though, because just reading all that code is a pain, esp when a lot of it is slop - the default/dare-i-say-decel way is to go back to "i own the execution, and give little things to the agent, check it along the way" - the accel-but-safe-way is to find something - NOT A SPEC (the word "spec" is broken anyway) - NOT 3 INVOCATIONS OF AskUserQuestion - that lets you resteer the model *before* it slops out N-thousand LOC