@harpercarrollai: Look no further for your first Claude Code demo!! Demos were WIDELY requested (& those requests were even followed up on — I see you & I got you 🤗!). You can build this /skill in just a few minutes and have a working daily brief skill that gets you prepared - with backgrounds + customized talking points - for all your meetings, every day, that improves over time. Would you like more demos like this? I would be happy to go into building an app end-to-end, setting up users & the backend (i.e., where your app’s data lives) and deploying it to the web… I also think it would be fun to fine-tune an open source model to talk like you, with the help of Claude Code to gather & clean the dataset, choose the model & hyperparameters, and train & deploy! (& the finished model would be completely yours afterwards!) ⚠️ Security considerations before connecting your email Connecting email to any AI agent carries risk. Malicious actors can embed hidden instructions in emails that attempt to hijack Claude's behavior (prompt injection). A few mitigations: 1. Give Claude a dedicated email address. Forward only relevant emails to it and invite it to meetings you want analyzed. This limits what Claude can access if something goes wrong. 2. Run in sandbox mode with read-only permissions. When invoking this skill, tell Claude Code: "Run this in /sandbox with read-only email access." 3. Audit this skill before running it. Ask Claude to review the skill file itself for injection vulnerabilities: "Review /daily-brief for prompt injection risks before executing." Today we used the Claude Code app, but I’m particularly partial to using Claude Code in the terminal. It has many benefits & I’d like to teach you how to do it. I promise it’s MUCH less scary than it looks… with AI today, the terminal is not much different than a chat interface. Let me know your questions & comments below! @Claude #ClaudePartner

Harper Carroll AI

Open In TikTok:

Region: US

Friday 03 April 2026 14:39:32 GMT