I have finally realized I have two styles of ICM that fall outside of the five flavors of ICM in the classroom. There have been times that I have an ICM that is a factory of results that I use multiple times. Then there are times I use an ICM for a new skill/research/brainstorming. I get VERY frustrated when Claude moves ahead on a brainstorming ICM without me. So I have written specific instructions that I put into brainstorming ICM's to be interactive with me at each step, and that I will ask many questions to determine the direction. Two of my factory ICM's - Creating brand one pagers, and putting sample data into a salesforce sandbox. One of my Brainstorming ICM's - Building out a sample dashboard for a mentoring program in real estate. Steal this if it helps anyone.

Do we still need dashboards, menus, buttons and complex applications? Or will the AI itself become the main interface, with a simple visual layer only for things like monitoring, approvals and seeing what’s happening? I’m curious how you see this evolving over the next 5 years. Will we still use software, or will we mostly talk to software? @David McDermott @Ari Evergreen @Jake Van Clief @David Vogel @Matthew Creamer @Bas Rosario

Just a simple poll interested in what people think and to spark some discussions....

Hello everyone, I just recently came across Jake's videos and I learned a lot from them. I just installed this or the ICM skill and let it rework my whole Claude code folder. After seeing how effective it is and making it more my style, i came up with two questions: 1. My first question is: does it matter how big the main folder gets? So can I have one main folder with every app, personal workflow, or project in there? With the whole ICM process? Or is it better to split that up and make seperate folders for different uses cases? 2. My second question: how can I use ICM with Obsidian efficiently? ICM already creates a bit of a memory/context for every part of the folder, but how can I make Obsidian a strong extra pillar of memory in there. So i can have a shared vault with my business partner and our projects are in line with the company's state. Please help me out with this. I'm open to learning and thank you for reading.

spire-pipeline is a plugin for Claude Code / Cowork that takes a project from a vague idea to a finished, verified build - across as many sessions as it takes, without losing the plot between them. Four skills: DESIGN-SWARM ▸ give it a hard problem and instead of one answer it generates several competing approaches, each deliberately biased a different way, attacks them all at a critique gate, and hands you the survivors to pick from. ends at a design you signed off on. never builds. RUN-LIST ▸ turns that design into one planning doc: the build split into ordered sessions, what depends on what, a paste-ready prompt per step, live status. the document IS the project state — any future session reads it cold and knows exactly where things stand. CONDUCTOR ▸ executes it like a foreman who never touches the tools. collects every decision up front in one sitting, then runs for hours without pinging you, farms each step to worker agents, and refuses to tick anything done unless a mechanical probe proves it happened. a worker saying "finished!" counts for nothing; a file existing with the right contents counts. SETUP (optional) ▸ first-run onboarding - detects what it can about your machine, asks two questions, writes a minimal config. you never need it; zero-config runs fine on safe defaults. Works on anything with 3+ steps - code, docs, content, systems. This is the actual engine I run my own work through, published as-is: the house-specific parts swapped for a config layer you fill in yourself (or don't — the defaults are safe, and every skill prints whether it's running on your config or on defaults). CREDIT ▸ two of the mechanisms are borrowed and credited, both MIT, both from @Ari Evergreen - https://www.aris-space.com/ the brainstorming-swarm that design-swarm is adapted from, and the context-pack contract the conductor uses to brief its workers. I built the pipeline around them - the planning, the multi-session state, and the probe-verified "done" are mine.