△ Warning · claude-mem
The 808 GB disaster
Old claude-mem versions had a memory bloat bug — mine grew to over 800 GB before I noticed. If you installed it months ago, fix it today.
npx claude-mem@latest install Frisco · Plano · Claude Meetup · May 15, 2026
Sponsored byTonight · Frisco · Plano
Three talks. One night. Tap a row to expand.
How Cowork changes the loop between you and the model — and why Markdown beats HTML as the contract between humans and LLMs for almost every workflow you'd reach for a frontend to build.
How a working web-dev agency wires Claude into the daily build — prompts that earn their keep, the guardrails that keep client work clean, and the practices that survive contact with a real Friday deadline.
Everything on this page, live. A course, a hotel pipeline, a portfolio, a Chrome extension, a daily AI digest, a toolkit, three resources, and one experiment. Stack chips, proof of revenue, and the parts that broke first.
The course · ccforsf.com
Claude Code for Salesforce Admins. A practical course showing admins how to build Flows, Apex, and validation rules from plain English. Pitched as the $97 alternative to a $125/user/month Agentforce license.
The whole product runs on a five-tool chain. Built with Claude Code inside Cursor IDE, pushed to GitHub, deployed on Vercel. Course content lives on Systeme.io. Payments flow through ThriveCart. No CMS. No agency.
Live at
ccforsf.com
The client work · hotel AI
A hotel operator hired me to lead AI research and find cost savings across their operations. They have a long list of problems worth solving — this one was first up.
Every night, 20+ audit PDFs hit the inbox from the property management system. Someone used to read all of them by hand. Now they get ingested automatically, parsed by Claude, and rolled into a single daily dashboard.
Revenue, rooms sold, occupancy, ADR, RevPAR — the metrics hoteliers actually care about, surfaced before the GM's first coffee. Built on n8n with AgentMail handling intake and Claude doing the parsing.
The portfolio · amit.so
amit.so — the personal site, built end to end with Claude Code + Cursor IDE. Hero, experience, skills, 8 Salesforce certifications, career timeline, side projects, contact. Everything someone hiring a strategic GTM consultant needs to see.
One detail worth flagging: the animated hero header came from motionsites.ai — a great example of borrowing a polished component instead of building from zero. Claude handled the rest.
Live at
amit.soThe Chrome extension · BetterSkillsMD
BetterSkillsMD fixes the part of vibe-coding that breaks: describing UIs to your LLM. You lose hex codes, spacing, type. You iterate five times. You burn tokens.
Click the extension. Pick a capture mode — viewport, full page, drag a region, or click an element. GPT-4o vision reads the screenshot. A streaming design.md writes itself in the side panel. Copy. Paste into Claude. Ship.
5 free captures, then $10 once. Bring your own OpenAI key — screenshots never touch my servers.
Live at
betterskillsmd.comThe daily loop · Claude Cowork
Two scheduled Cowork sessions, both fire before I'm out of bed.
The Daily AI Feed. Cowork monitors my favorite X accounts — Claude, OpenAI, Gemini, Matthew Berman, Peter Steinberger, Tom Dörr, klöss, Alex Finn — pulls the day's signal, and lands a digest in my inbox at 5:42 am. I read it with coffee. No more doomscrolling.
The ccforsf social engine. Cowork drafts three posts a day promoting ccforsf.com — different angles, different platforms. I review, schedule them through my scheduling tool, and they ship. Distribution on autopilot.
In progress · tactical wins
Live experiments — small enough to swap out, useful enough to keep using. These three earn their keep every week.
Take these home · resources for the room
These are the ones I actively use and recommend without caveats. Bookmark them tonight.
Persistent memory · claude-mem
claude-mem gives Claude Code session-spanning memory. It's been one of the most useful upgrades to my setup — and also the one that nearly ate my disk. Here's what happened, and how I run it now.
△ Warning · claude-mem
Old claude-mem versions had a memory bloat bug — mine grew to over 800 GB before I noticed. If you installed it months ago, fix it today.
npx claude-mem@latest install My fix for the bloat
Even with the fix in place, I didn't want to trust the cleanup. So I built a daily automation: it reads the claude-mem JSON, converts every entry into an Obsidian note, then deletes the original JSON from the claude-mem folder.
My memory survives in a format I actually own. My disk stays alive. Win-win.
Skills · CLAUDE.md packages
Each one is a small CLAUDE.md package — drops into a project, reshapes Claude's defaults for a specific job, then gets out of the way. Stealable from anywhere they help.
amit-voice Writes in your voice (no em dashes, no lecturing) for emails, posts, scripts.
youtube-video-package Transcript → titles, description, thumbnail prompts, tags.
youtube-video-prep Pre-record talking points and explainer artifacts from a source or transcript.
linkedin-video-post Transcript → viral LinkedIn post + first comment + thumbnail prompt.
ccforsf-lesson-package Lesson transcript → ThriveCart-ready title, description, and metadata.
interview-research Calendar event → company + interviewer research brief.
job-search Tailors your resume to a JD without inventing experience.
salesforce-mcp-setup Walks through connecting Claude Code to a Salesforce org via MCP.
The experiment · future self prompting
A Berkeley psychologist published a protocol called Future Self Prompting. The premise is simple. You write a letter to yourself from the version 10 years ahead — the one who already has what you're chasing today. Then you let that version do the talking.
The research underneath it is real. UCLA's Hal Hershfield spent 15 years proving one thing: the more vividly you connect to your future self, the better every decision you make today gets. Money. Health. Relationships. Career.
Six prompts. Run them in order. The first one took me 20 minutes to sit with. The second one is brutal — it shows you the version of yourself you'll become if nothing changes. The sixth one ended up taped to my monitor.
The six prompts · run in order · click to expand & copy
You are me, 10 years from now. You already have everything I'm working toward right now. You're writing me a letter from that future. Tell me 3 things I'm currently doing that you're grateful I didn't quit. Tell me 3 things I'm doing right now that you wish I'd stopped sooner. Be specific. Be honest. Don't be polite.
Same future self. Tell me about the version of me you almost became. The one who didn't make the changes. What does his life look like at 45? What did he settle for? What did he tell himself to make the settling feel okay?
Future self, walk me through one ordinary Tuesday in your life. Wake up to bedtime. Don't skip the boring parts. I want to feel what it's like to be you.
Future self, I'm currently stuck on [specific decision]. You already know how this turned out. Tell me what I chose, why I chose it, and what I'd be losing right now if I'd picked the other path.
Future self, what does the version of me right now believe about himself that's no longer true for you? What story did I have to drop to become you?
Future self, write me one sentence I should tape to my monitor for the next 12 months. One sentence. Make it the one I'll most need to hear on the days I want to quit.