Get live statistics and analysis of dennis palma's profile on X / Twitter

Why pay $200/mo for SaaS bills when you can run lots of apps on a <$10, $20 Hetzner server? n8n,lovable,supabase,etc and here's the kicker: without Claude Code: spend days burning brain cells, setting up servers, juggling configs With Claude Code (+ o3): gave it SSH access and prompts — it set up Hetzner, Supabase, n8n, Coolify, Cloudflare Tunnels, etc… everything!(or not, since main nameserver/s actual edit) checklist for launching multiple web apps 🚀 ✅ Hetzner (cheap, fast, but with KYC) + extra volumes. budget with scaling options ✅ n8n? go self-hosted ✅ supabase? also go self-hosted ✅ supabase resource-heavy? schema-type approach -> 1 web app = 1 schema... note: split once app validated/grows ✅ vercel / git actions / auto-deploy? Coolify self-hosted ✅ Cloudflare tunnels - no public IP exposure ✅ buckets / s3? - hetzner local dev: ✅ Docker Desktop • Next.js/React • shadcn/ui • GitHub ✅ lovable, v0, etc? claude code with context md framework, mcp ref

Having multiple test web apps with Supabase — each with its own project — was straining my Hetzner server. Why? Every Supabase project spins up a full Postgres instance, APIs, and background services. So I switched to a schema-based approach — each app now lives in its own private space inside one Supabase database. ✅ Clean separation ✅ Faster & lighter ✅ Easier to manage Thanks to Claude Code for setting everything up. Now I can ship ideas way faster 🚀

What if your PRD, prompt, or any info, was debated by AI? or not prd, but any prompt/info? just built some interface for easier viewing - multi-LLM prompting to refine product ideas. Letting Claude vs ChatGPT battle it out to improve: ✅ Prompts ✅Product specs ✅other metrics Can LLMs do it better, "together"? 👀

doing multi-modal LLM - this is why i always come back to o3. Claude code even with ultrathink, conceeded... I did prompt it to be sort of devil's advocate, be critical etc...

just a reminder, it's been 69 days since Nano banana was released. What have you built? 🙄

Problem: You lose context or get confused when running multiple claude code CLIs (3, 4, 5, x) on the same project folder (or different worktrees / projects). Additional cognitive load and scrolling on to find out which of the cli panes is doing what. I'm not using tmux, other ide's, but just powershell since i'm on windows (so using claud code cli on windows native) Solution: Add 'manual' sort of identifier in claude code identifier for easy display, through windows $env. Before, my statusline already shows the current project folder, so can already easily identify what project (or worktree) a current cli is running on. I also tried hooking a current 'implementation plan + task' to reflect in statusline, but still, have to think through on what it displays, and also it sometimes just doesn't work (hooks from the internal tool calls of claude code e.g. exitplanmode, entering of the plan mode / ToDos etc). So easiest way that works for me, which I can easily read, is add that sort of identifier, e.g. !sl CheckingAPICalls, which will add that to statusline. Together with the project foldr + other info, then can easily tell what that cli instance does. cd C:\apps\MyProject cc # Launch Claude Code with PANE_NAME=CC-20251106-143022 Inside Claude Code: !sl FixingAuthBug # Set manual identifier Custom statusline then displays session context: \apps\MyProject | [M] FixingAuthBug | Sonnet 4.5 | Plan >> Active task Shows: Project folder - Know which worktree/project you're in Manual label - Your custom identifier from !sl command AI model - Current model (optional - requires ccusage) Active work - Auto-extracted from TodoWrite/Task tools

Why pay $200/mo for SaaS bills when you can run lots of apps on a <$10, $20 Hetzner server? n8n,lovable,supabase,etc and here's the kicker: without Claude Code: spend days burning brain cells, setting up servers, juggling configs With Claude Code (+ o3): gave it SSH access and prompts — it set up Hetzner, Supabase, n8n, Coolify, Cloudflare Tunnels, etc… everything!(or not, since main nameserver/s actual edit) checklist for launching multiple web apps 🚀 ✅ Hetzner (cheap, fast, but with KYC) + extra volumes. budget with scaling options ✅ n8n? go self-hosted ✅ supabase? also go self-hosted ✅ supabase resource-heavy? schema-type approach -> 1 web app = 1 schema... note: split once app validated/grows ✅ vercel / git actions / auto-deploy? Coolify self-hosted ✅ Cloudflare tunnels - no public IP exposure ✅ buckets / s3? - hetzner local dev: ✅ Docker Desktop • Next.js/React • shadcn/ui • GitHub ✅ lovable, v0, etc? claude code with context md framework, mcp ref

when doing front-end with claude code, just give it your drawing ... excalidraw is enough

Problem: You lose context or get confused when running multiple claude code CLIs (3, 4, 5, x) on the same project folder (or different worktrees / projects). Additional cognitive load and scrolling on to find out which of the cli panes is doing what. I'm not using tmux, other ide's, but just powershell since i'm on windows (so using claud code cli on windows native) Solution: Add 'manual' sort of identifier in claude code identifier for easy display, through windows $env. Before, my statusline already shows the current project folder, so can already easily identify what project (or worktree) a current cli is running on. I also tried hooking a current 'implementation plan + task' to reflect in statusline, but still, have to think through on what it displays, and also it sometimes just doesn't work (hooks from the internal tool calls of claude code e.g. exitplanmode, entering of the plan mode / ToDos etc). So easiest way that works for me, which I can easily read, is add that sort of identifier, e.g. !sl CheckingAPICalls, which will add that to statusline. Together with the project foldr + other info, then can easily tell what that cli instance does. cd C:\apps\MyProject cc # Launch Claude Code with PANE_NAME=CC-20251106-143022 Inside Claude Code: !sl FixingAuthBug # Set manual identifier Custom statusline then displays session context: \apps\MyProject | [M] FixingAuthBug | Sonnet 4.5 | Plan >> Active task Shows: Project folder - Know which worktree/project you're in Manual label - Your custom identifier from !sl command AI model - Current model (optional - requires ccusage) Active work - Auto-extracted from TodoWrite/Task tools

Me=> Claude you're too slow launch 6 parallel agents please Claude Code: 💪😠

your future future self already forgot your 100s of ideas. start shipping. and keep away from alcohol

What if your PRD, prompt, or any info, was debated by AI? or not prd, but any prompt/info? just built some interface for easier viewing - multi-LLM prompting to refine product ideas. Letting Claude vs ChatGPT battle it out to improve: ✅ Prompts ✅Product specs ✅other metrics Can LLMs do it better, "together"? 👀

Having multiple test web apps with Supabase — each with its own project — was straining my Hetzner server. Why? Every Supabase project spins up a full Postgres instance, APIs, and background services. So I switched to a schema-based approach — each app now lives in its own private space inside one Supabase database. ✅ Clean separation ✅ Faster & lighter ✅ Easier to manage Thanks to Claude Code for setting everything up. Now I can ship ideas way faster 🚀

Is MCP really the best way to keep Claude Code up-to-date with APIs, docs, and info? I’ve tried scraping docs and feeding them in, but it gets bulky fast — context bloat, stale info, selective loading issues. Goal: Claude should always know the latest tools/stacks it’s building with — without drowning in clutter. o4-mini suggests RAG + auto-scraping + tool-calling… but honestly, I’m overwhelmed.

just a reminder, it's been 69 days since Nano banana was released. What have you built? 🙄

Been using worktrees with ai coding(claude code, codex) - launching different features, etc. Now when database is convex, I couldn't find an easy way to sort of *share* same dev db to the different worktrees. been using separate dev database for worktrees, but really the simple setup of having the same db across different worktrees also works and is easy, for me(depending on the tasks). Problem is, each npx convex dev results in a worktree's schema change to overwrite other worktrees schema definitions. Then thought, why not use claude code skills as well to define our workflow when creating new worktrees? and even merging etc So, just did a setup with Git Worktrees + Isolated Convex Deployments: - Separate git worktree (isolated working directory) - New Convex project with independent dev database - Data snapshot copy from main dev - Auto-assigned port (4031, 4032, etc.) - Configured environment variables Port Management Main repo for dev uses port 4030. Worktrees get sequential ports starting at 4031. When you delete a worktree, its port becomes available for reuse. Tracking happens in .worktree-ports.json. Merge Workflow The merge process excludes package.json port changes while keeping code changes and dependency updates. This prevents committing the wrong port (4031+) to main/staging (which uses 4030). Steps include: - Pre-merge testing (TypeScript, lint, Convex deployment) - Smart git staging that excludes port changes - Schema migration verification - Data migration prompts - Works with any target branch Why We Use Skills loads automatically when working in worktrees. include troubleshooting commands, the workflow on what we want, etc. When to Use Worktrees Use for: - Experimental/new features (easy to discard) - Parallel feature work (no database conflicts) Don't use for: - Quick bug fixes (setup overhead) - Frontend-only changes (no database interaction) So with a simple prompt, 'create 3 new worktrees pls for {feature1}, {f2}, {f3} pls', claude code loads the skills, and does its tasks like creating the local folders/worktrees, assigning ports, creating the isolated dev databases, migrating/copying the data, and others, based on our workflow. Works also with merging etc (auto-discards the temporary worktree ports assignments, discards the specific worktrees, convex dev db etc).