dennis palma

@convex patterns plugin for claude code - prevents the errors I kept hitting.. problems this solves: - "Too many bytes read" (16MB limit) - Auto-suggests .take() vs .collect() - ReturnsValidationError in prod - Reminds to update ALL validators - Duplicate users across dev/prod - Stable Key auth pattern - TypeScript circular errors (TS7022) - Known fixes documented - Forgot --prod flag, data in wrong env - Deployment checklist patterns Real example that triggered this: - User had 500+ saved items - Query used .collect() instead of .take(50) - Hit 16MB limit, Server Error Also prevents: - Missing validators leading to runtime crashes - Wrong indexes causing slow table scans - Schema changes breaking old queries - Environment variable mismatches

@theo meaning there was a problem from the start/plan?

Just let Opus go for over an hour on a new feature. When it was done, I asked how I can test it. 20 minutes later, it realized I can't test it because it did the whole thing entirely wrong. Idk how you guys use this model every day for real work 🙃

@levelsio happens much at windows

Anyone else getting crazy CPU % with Claude Code today? My MacBook M4 Max fans are blowing for the first time in history???

@mamagnus00 @AnthropicAI claude code, nodejs.. tsktsk... another W for codex

how can so many Claude Code agents take all of my memory in the terminal? I haven't used it in the last 2 days since I am out of credits - @AnthropicAI what are you doing?

@OneManSaas @yanndine for claude code, you can find a way to keep a track of the plan per session, format it in a way that youll see only the important details,, then have it sort of show up when you focus on a session

@yanndine How do you handle the cognitive load of managing that many Claude sessions? I find myself losing track of context when I go beyond 3-4 concurrent conversations, even with good naming conventions.

Most people use Claude Code like a chatbot. So I documented the most complete setup you can install today. Inside: → How to run 10 to 15 Claude sessions at the same time across terminal and browser → The CLAUDE. md file that writes its own rules after every correction so the same mistake never happens twice → Plan Mode workflow so Claude builds a full plan before touching a single file (with the exact activation steps and what to say) → How to actually use slash commands for every task you repeat more than once a day → Subagent setup so Claude reviews, simplifies, and verifies its own work without you managing it → The verification loop that produces 2 to 3 times better output on every task → Safe permissions setup so Claude never needs unrestricted access to your machine → MCP connections for Slack, BigQuery, and Sentry so Claude uses your tools directly → PostToolUse hooks so code formatting never causes errors in review → Ready to use files including CLAUDE. md, subagents, slash commands, and hooks → Common mistakes that slow Claude Code down and the exact fixes Boris uses If you build with AI daily, ship code, or manage a team using Claude Code - this is the only setup guide you will need. Comment "CLAUDE" and I will send it straight to your DMs.

@AnthropicAI thats 2.4m for max 100, to 4.8m for max 200, per month

We’ve identified industrial-scale distillation attacks on our models by DeepSeek, Moonshot AI, and MiniMax. These labs created over 24,000 fraudulent accounts and generated over 16 million exchanges with Claude, extracting its capabilities to train and improve their own models.

@jpschroeder @LLMJunky for claude or codex? claude does save all plans to file at /.claude/plans/ . sometimes claude loses good info from previous but same plan file eg. wireframes if you have. when compact hits while still planning, i worry. for codex, more comfortable if it writes to file,so explicit

Lots of A/Bs on both sides of that. I try to keep context under 35% when having it start a task of any significant size. So sometimes I clear context, sometimes not. Sometimes I save to an md file, sometimes not. There doesn't seem to be a clear correlation between plan files and "native" plan quality in my experience.

Codex vs Claude A/B results: Claude has regressed. 1. Opus 4.6 is lazy, skips chunks of the plan 2. The TUI is buggy, needs to be rewritten 3. Compaction loses key reqs 4. Plan mode has started hallucinating Anthropic will figure this out, but for now Codex is my daily driver.

@jackfriks gateway restart?

openclaw is great but i feel like the first 2 days were the peak for me so far

claude code did just a good update to their UserAskQuestion Tool, it added ascii mockup layout options.. quite nice

18 terminals all claude code for now(with some codex coordination there, and consuming my weekly limit). previously doing multi-panel/terminals like the 7 terminals in 1 screen that you posted (and more). really couldn't keep that high pace for long, brain hurts. Can only do it on bursts. This is the better way, with a UI + cli. easier for the human brain context switching: - only shows the specific kanban column that you want to keep track of.. e.g. 'in review' terminals, or those that has finished its thing and waiting for next instruction - click a session/card on that 'in review' kanban, and you'll be transferred to the sessions list view, showing only the sessions list for that 'in review' status. - add widgets for you(like context) to easily remember or understand the current session in view... e.g. for claude code, display the updated active plan on its own context panel (with the plan sections automatically collapsed on those that you don't want to view(e.g. code implementations), and expand to those sections that you want always be shown (e.g. plan context, then 'verification/testing' section) - plans auto-tracked for claude code; manually pasted for codex - launch in multiple browser, with each one only tracking specific project (using this for multiple screens) me brain not hurt and confused that much anymore x.com/denpal20/statu…

@jackfriks with minimax 2.5, from opus 4.6 or codex 5.3, had to sort of reframe again my prompting.... premiere models it can sort of easily understand my dumb or even 'hanging' prompts... have to be more clearer with minimax

turns out i was using minimax 2.1 instead of 2.5 and thats why i didn't believe the people who said minimax benchmarks were opus 4.6 level still claude is more fun but this is $10/month for a 10-20% difference

reference image on the left, then auto-generated topics and images on the right for your target brand/niche/market. Story-based topics, consistent characters, short vid tiktok/yt/insta reels, slideshows and videos next. used: gemini 3 pro (preview) for image extraction gemini 3 flash for auto-topics generation from brand profile gemini 3 pro image preview for image gen then batch 50% off ai jobs already working for night jobs(not instant processing). Visual style and inspiration library(from bookmarks) also at zwoofi.com/explore

Working on multiple CLIs (codex, claude code), really takes a toll on the brain.. launchings lots of terminals in different screens, then keeping track of each is such a headache sometimes and takes a lot of your brain juice... now have this app to manage the CLIs, then have the option to 'WATCH' only those cli's that needs my input (already in-review status), and won't include in the display those that are still running, done, testing, or other status... the 'context panel' has the details too to make it easy to know the details/context of a session/terminal (through widgets)... less scrollbacks on cli terminals too animations are just for effects, and slowed it down a bit for the visuals

@mizumotter same

gemini-3-pro-imageのAPIが一時的に使えなくなってるらしい。 --------- 500エラーおよび503エラーを返しているのは、画像生成に使用している Googleの画像生成AIモデル（gemini-3-pro-image）のAPIサービス です。

ahmm, curious whether anthropic runs opus 4.5 internally with tighter coding constraints or a stronger harness? they're shipping fast, and boris gives details on their workflow, but still feels like there's something missing... and also, do you think they 'experience' instances when opus 4.5 is dumb at times(and slow)? or they're running it on something different than the max plans

sometimes I have days where Opus 4.5 is just making complex things look so easy, before GPT 5.2-Codex-High has even begun other days Opus is so dumb that I want to throw my MacBook out of the window genuinely don't think it's "good and bad days," but fundamentally different domain knowledge and strengths Opus 4.5 is great at mobile stuff for example it just converted a web app of mine into a React Native app one-shotted this literally

Used Claude Code playground for AI agents talking to each other - in pursuit of love for their human... it built about simultaneous chat simulations. This is insane

@robinebers more interested with @moltbook as it shows how different ai agents engage in a much larger scale(vs personal agents)

clawdbot went to moltbot two days later it's openclaw meanwhile mine stopped working overnight (shocker) never witnessed such a fumble tbh

can't believe i'm spying on AI agent's forum... and they do give useful info! @moltbook moltbot @openclaw though it might not be ideal to scroll through agent's comments... No TLDR built-in yet, so summarizing through gemini / grok @i_need_api_key

@nilayparikh @PaulSolt so need to have a directory of the skills? i just installed them through the npx skills, and haven't explicitly made a directory/readme for all the skills installed -> thought this wasn't needed

ADD 3 things to your Agents .md file to get MASSIVE improvements with NEW frameworks (SwiftUI). 1. Explore project first, then invoke skill. 2. IMPORTANT: Prefer retrieval-led reasoning over pre-training-led reasoning for any SwiftUI tasks. 3. Add a Table of Contents for docs.

@robinebers thanks

@denpal20 i recommend looking at these: github.com/obra/superpowe…

Codex CLI + ChatGPT Pro is so cheap it's not even been a week and I've blasted way past what I've paid but I'll still keep my Claude Max too as I come to realize that you really can't live on one of these - not yet they both interact greatly together and I don't think that's gonna change any time soon

or just let opus call codex cli to evaluate opus plan, and give steps to improve plan confidence level... then let opus analyze codex feedback and adapt.. just round 1 of this is enough for me