📝 SUMMARY A technically rich call spanning terminal tooling, agentic architecture, and practical Claude workflows. Patrick showcased his Claude Code plugin ecosystem — including a status bar plugin and CMUX integration — and shared a detailed Cowork folder structure for building persistent Claude memory across sessions. The group converged on a key insight for production agentic systems: standard SDLC principles apply directly, with one critical addition — designing explicitly for LLM non-determinism through idempotency and retry-safe patterns. Paul demonstrated Stagehand for authenticated browser automation, Morgan shared progress on a cemetery data visualization app, and the group explored tools from Pencil.dev for AI-assisted UI design to 1CLI for agent API key management. The emerging theme: Claude Code workflows are maturing rapidly from individual productivity tools into multi-machine, multi-agent operational infrastructure. 💡 KEY INSIGHTS Cowork as a PM layer, Claude Code as a coding layer Patrick separates conceptual and planning work (Cowork) from implementation (Claude Code). Cowork builds persistent context over time, reducing re-explanation overhead across sessions. The IDE is now just a markdown viewer Patrick no longer uses VS Code or Cursor as a primary workspace — only as a file viewer for markdown and generated output. All active work happens inside CMUX. Stay below 60% context window usage Experienced users deliberately keep context below 60% even when 1 million tokens are available. Output quality degrades in the middle of very long contexts — more tokens does not always mean better results. Back up your Cowork sessions Cowork session data is stored locally and can be corrupted by power outages mid-session. Patrick is building a backup system. Anyone using Cowork seriously should be aware of this risk. Plugin marketplace pattern for Claude Code A marketplace can be added with a single command and individual plugins installed from it — enabling composable, shareable Claude Code extensions across the community.

Impromptu Zoom Meeting - March 17 VIEW RECORDING - 91 mins (No highlights) Meeting Purpose A roundtable on recent AI development projects and tools. Key Takeaways - New Agent Dev Environment: Patrick demoed a new setup using CMUX (a modern tmux alternative) as a lightweight terminal host for Claude Code, enabled by two new plugins: CC Status Line and CMUX Integration. - Complex Data Challenges: Morgan's "Heritage Plot" project highlights the difficulty of migrating data from legacy systems (e.g., AccessDB) and managing complex permissions (e.g., GRAMMA requests) for public vs. private records. - Production-Ready Agents: The group agreed that agentic solutions require standard SDLC practices, with the key challenge being LLM non-determinism, which can be mitigated by treating agents as non-deterministic "users" within the system. - "Claudebernetties" Concept: Patrick proposed a "Kubernetes for Claude Code" to manage and replicate skills, prompts, and configurations across multiple projects and machines, solving a growing friction point. Topics Agentic Development Environments - Claude Code Plugins (Patrick): CC Status Line: Displays Claude Code version, model, token usage, and Git info in the terminal status bar. CMUX Integration: Wraps the CMUX CLI to make Claude Code aware of its terminal environment, enabling features like live progress bars and browser splits. Plugin Marketplace: A custom marketplace for publishing and installing plugins. - CMUX Terminal Environment: A modern tmux alternative built on Ghosty, chosen as a lightweight host for Claude Code over heavier IDEs like Warp. Key Features: Multiple, customizable panes and windows. Browser splits for agent interaction and testing. CLI-driven, making it fully scriptable for agents. - Cowork Agent: Patrick uses Cowork as a project manager, finding it superior to OpenClaw for security and context management. Workflow: Use Cowork for high-level planning → use Claude Code for detailed development. Challenge: Cowork's local session files are prone to corruption; Patrick is building a backup system.

Hey everyone, Quick preview for Tuesday. Alongside whatever discussions and rabbit holes we get into on the call, I've got a short show & tell segment I'm looking forward to. Lately I've been using Claude's Cowork mode as my dev project manager, it tracks my projects, remembers context between sessions, knows my stack, my terms, my priorities. It's basically become the place where I think, plan, and build. In fact, the post you're reading right now was drafted there. One of the things that came out of that workflow is a Claude Code plugin ecosystem I've been putting together. That's what I'll be walking through: CC-StatusLine: a plugin that gives you a live dashboard in your Claude Code session. Context window usage with a color-coded progress bar, session cost, git info, model details, all at a glance. cmux-plugin: hooks Claude Code into the cmux terminal. Auto-named workspaces, completion notifications, live progress bars, browser splits. Knows when to stay silent if you're not running cmux. patchoutech-plugins, the branded marketplace that ties it all together. One command to add, one command to install. I'll show how the registry works and how you can spin up your own. If you've been curious about how Claude Code plugins work, hooks, skills, commands, the whole structure or you just want to see what it looks like when you let an AI help you manage the whole dev cycle from planning to shipping, this should be a fun one. See you Tuesday!

Here is a new tool from the team behind LightRAG and Nanobot lets you auto-generate a CLI for any open source software. Point it at a codebase, and it builds a fully tested command line interface that Claude Code can use natively. https://www.youtube.com/watch?v=Uzd2ckXnsg0

Last week, Patrick's automated recap pipeline ran live for the first time — and it basically wrote this post. If that doesn't make you want to see what this community is building, I don't know what will. 📞 HOW THE CALLS WORK The calls can run 2+ hours. We want to make sure we're respecting everyone's time. Especially those of you who actually show up. Here's the structure: 👉 Reply to this post with your questions before the call 👉 If you submit a question and you're on the call, you go first 👉 We work through questions in the order they came in 👉 Then we open it up for everyone else If you can't make the call but want your question answered, drop it in the comments. We'll get to it. But priority goes to people who are there. The goal is simple: if you're taking the time to show up, you shouldn't have to wait behind questions from people who aren't even on the call. We've got some good threads to pick back up this week. Morgan committed to having a demo of his Heritage Plot cemetery platform ready — and after last week's UX conversation, that should be a great one to dig into together. Ty is mid-rollout on his Contributor Model across two businesses and we'd love to hear how staff are responding. And the community agreed to give Claude Code Review a week to breathe before passing judgment — that verdict is due. 🔗 ZOOM LINK (save this) https://us06web.zoom.us/j/81995207847?pwd=Xe6u6LmIQOmCP5VTnOwWYjDBfZNKGB.1 📅 WHEN Tuesday March 17th at 6PM ET Looking forward to seeing you on the call!