After my API keys and tokens were exposed repeatedly in chats with Claude, I built skills to enter secrets without showing them, plus guardrails to stop exposure. I packaged those as Secret Guard, a Hermes Agent plugin. It is for anyone who uses an AI agent with real credentials and wants to keep those keys away from the model. 1. Enter the key through a hidden dialog instead of the chat. The key is stored in the macOS keychain. 2. Tool calls that would read a key back are blocked before they run. 3. Any exact key value that shows up in tool output or model replies is scrubbed. Result: API keys stay out of everything the AI model sees. https://www.loom.com/share/c36f600ce8364a488ab3aed89e8ec157

I use Claude as an SEO helper for my WP site. It reads my search performance data and suggests improvements, so I don't have to dig through the numbers myself. This is useful for anyone who runs a website and wants to improve it using their own search results. 1. Connect your WP site to Search Console. 2. Let Claude read the Search Console results. 3. Ask Claude to improve the content. 4. Let Claude make the improvements. 5. Republish the updated content on your site. The result is improved content for your WP site, based on what your search results actually show.

I have around 150-200 online courses spread across platforms, and I don't enjoy learning from video. I wanted to collect the useful content in its original wording so I could read it at my own pace. While designing a local app with ChatGPT for this, we also developed a way of working together, and I turned that into a reusable skill called Appontwerp (Dutch for "app design"). 1. Start with a real project. I brought an existing Excel inventory and files from a previous working setup. These gave us concrete material to inspect and reuse. 2. Make corrections specific. "Don't rush" became: understand the goal and scale, inspect what already exists, ask about unclear decisions, offer your own recommendation, then make a plan together. 3. Separate decisions from experiments. We tracked what was decided, what still needed my input, and what a technical test had to establish. 4. Capture the method while it's visible. I asked ChatGPT to extract our preferences from the conversation and make a skill for local app design. In my setup, it packaged this as a private plugin. 5. Keep reusable guidance separate from project details. Our light-blue background and purple buttons didn't become mandatory for future projects. My existing personal-context plugin remains responsible for tone. The prompt I used: "Review how we've worked together in this conversation. Extract my explicit corrections and confirmed preferences into a reusable skill for local app design. Cover understanding the goal, reusing existing work, clarifying meaningful decisions, distinguishing decisions from test questions, agreeing a plan, and executing it step by step. Preserve my existing tone/context guidance. Separate general working preferences from project-specific choices. Don't claim anything was implemented or tested unless it was." The result so far is an early local inventory app and a private skill describing how I want us to design the next one. The course-harvesting automation is still unfinished, and the new skill hasn't been tested across future projects yet.

I built TEINT (teint.online), a free foundation shade finder for Filipino shoppers. It matches your skin tone from one selfie against 32 brands, 16 Filipino and 16 international, and runs right on your phone. I used Claude as my coding partner for the color science and face detection code. 1. Take or upload one selfie. 2. Face detection runs on your phone, so the photo is never uploaded. 3. The app corrects for the lighting, whether it is warm bulbs, daylight, or a dim room, before reading skin color. 4. It samples the skin tone in CIELAB color space. 5. It ranks shades from all 32 brands using CIEDE2000, a formula that measures how different two colors look to the human eye. 6. You get your closest matches in seconds. I also used Claude to generate shade guide pages for every brand so people can find the tool on Google. I wrote it in TypeScript and tested matches under different lighting before launch. The result is a short list of close foundation shades from a single photo, and the photo itself is never uploaded.

Frustrated with juggling multiple subscriptions and disconnected tools, I started building my own 'Everything app' a couple of months ago. I call it my portal. I built it for my own business. MCP helped me connect services, but it still wasn't clicking for me, so I built one integrated system instead. 1. Set up the core tools: CRM, task manager, time tracking, second brain, and bookmarks management. 2. Add money handling: invoicing with Stripe integration, payment reconciliation, recurring payments, expense tracking, and taxes. 3. Add sales and admin tools: sales and outreach, proposals and contracts, calendar, newsletter management, and website and app uptime tracking. 4. Add clients and collaborate with them inside the portal, and add my human team members. 5. Connect all of my LLMs and AI harnesses into one work ecosystem. 6. Operate the portal headless from Claude, and use the graphical interface when I want to look at records myself. The result is that I cancelled four or five paid subscriptions, and every data point now connects, so my data works for me. My Everything app includes the following: - CRM - Task manager - Time tracking - Client collaboration - Second brain - Bookmarks management - Virtual office visualisation (for lolz) - Invoicing with Stripe integration - Payment reconciliation - Recurring payments - Expense Tracking - Sales and outreach - Proposals and contracts - Calendar - Taxes - Collaboration - Newsletter management - Website uptime tracking - App uptime tracking - AI harness collaboration Inside this Everything app, which I like to call my portal, I can add clients and collaborate with them. I can add human team members, and I can also connect all of my LLMs and AI harnesses in one big work ecosystem. It's headless and allows me to communicate with and operate it directly from Claude, but also gives me a graphical user interface for when I just need to look at records and see something for myself. For many, this might be overkill. For me, it's allowed me to cancel four or five paid subscriptions and make my data work for me by having it all in one integrated system where every data point connects.

I built a handwriting app for Mac for Tamil, my language, using Claude Pro. I had already picked up a few ways of using AI at work, and that helped me give Claude clear instructions for this project. 1. I gave Claude Pro step-by-step instructions describing what I needed. 2. I had Claude set up the input and output folders. 3. I had Claude set up Git, both locally and online. 4. Claude created an installable Mac OS app for me. Result: a working, installable Mac OS app for handwriting Tamil, built from my step-by-step instructions to Claude. Github link for anyone who wants to try it out: https://github.com/kazhugu/Tamil_Handwriting_Beta

I have logged every minute of my life since December 2017, about 79,000 entries so far. I run four companies, from delivery logistics to knowledge-base software. This year I moved those spreadsheets into Well Lived, an app I built and have in beta at welllived.io, and put Claude Code on top of it. This is for anyone who wants a daily and weekly loop that adjusts how they spend their time. 1. Log as I go. Whenever I switch activities, I type or say what I did, with a note about how it went or what I learned. Well Lived turns that into entries with times, project, and how drained or energized I felt, from -10 to +10. 2. Each morning, before planning anything, Claude reads my inboxes, summarizes each email, and numbers it with a recommendation: draft a reply, make a task, or archive. 3. The plan is saved to a file, and the script that changes my inbox can only run what is in that file. Every change asks for my Mac password, so I have to be at my desk, but it is more secure. 4. Every morning, Claude reads yesterday's log, last night's notes, my calendars, and today's ClickUp tasks, then puts the most important task in my first block. 5. Each night I rate the day against my values. On Sunday, Claude compares where my hours went with what I said I wanted and makes a small change. This summer the log showed music had dropped to about 12 minutes a week. I knew I wanted more time for it, but seeing that number every Sunday is what finally got me to change it. The last three weeks averaged 7.7 hours of music. What I'd do differently: save every recommendation to a file from day one. Early on, Claude acted from its memory of a long conversation, and that memory drifted. If you're starting out, build the habit slowly. Logging is helpful, but the regular reviews are what make the data impactful in your life.

I run a solo handyman and property maintenance business, and this is the AI back office I built to run it without a full office team. I wear a Fieldy recorder during customer meetings, property inspections, and subcontractor talks. ChatGPT then combines those recordings with my Gmail, Google Calendar, and QuickBooks Online data into one daily operations briefing. 1. Record customer meetings, property inspections, and subcontractor discussions with the Fieldy recorder. It transcribes job details, measurements, promises, pricing, and follow-up commitments. 2. ChatGPT connects to my business and personal Gmail, Google Calendar, QuickBooks Online, Google Drive, and available Fieldy recordings, so information that used to be scattered across apps is in one place. 3. Every morning at 6 AM, a ChatGPT automation reviews the previous day's conversations and current records to identify outstanding jobs, estimates, approvals, appointments, payments, and customer commitments. 4. ChatGPT keeps tracking each unresolved item until it is confirmed completed, cancelled, or superseded, so verbal promises and unfinished maintenance requests do not disappear. 5. It compares customer emails, recorded conversations, and calendar appointments to find missing bookings, scheduling conflicts, and approaching deadlines. 6. It labels low-value promotional emails, marks them read, and archives them, while protecting customer inquiries, receipts, financial records, and important correspondence. 7. I use ChatGPT with QuickBooks to prepare estimates, review receivables, flag potential payment discrepancies, and coordinate customer and subcontractor communications. 8. A daily 4 AM operational summary is shared between ChatGPT and Grok to keep both systems aligned on business priorities. 9. I receive one consolidated report each morning covering today's priorities, new requests, aging commitments, missing appointments, financial follow-ups, completed work, and inbox cleanup results. The core instruction: "Review yesterday's conversations and connected business records. Carry forward every unresolved commitment until verified completed, cancelled, or superseded. Cross-check email, calendar, accounting, and project records. Identify scheduling conflicts, missing appointments, overdue payments, and customer follow-ups. Clean low-value newsletters without touching important correspondence. Produce one concise daily operational briefing." I review important decisions and authorize external actions, such as sending customer emails, changing invoices, or booking appointments. Instead of hiring an administrator, dispatcher, estimator, and operations coordinator, I use AI to handle much of that work while I'm in the field. My goal is a service company that can eventually run without my daily involvement.

My preteen daughter has spent months debating whether or not to get bangs, mostly worried about the "total humiliation" if they went wrong. So we uploaded her photo to ChatGPT and asked for four versions of bangs: blunt, curtain, wispy, side-swept. She loved it and finally narrowed down her choice, and the appointment is booked for this weekend. It’s hardly a groundbreaking use case, but it’s a fun way to try on a decision before a stylist makes it permanent.

I use Convertlyft to see how people actually use my website and to find out what to change first. It covers funnels, session replays, rage and dead clicks, form friction, JavaScript errors, and page speed, and it also audits the site's SEO. It's aimed at anyone who has dashboards full of numbers but no clear idea what to fix. What sets it apart is that it doesn't leave you to interpret the data alone. Instead of handing you charts to decode, it tells you what to fix first and shows the evidence behind each recommendation, so you can check the reasoning before acting on it. What it produces: a set of website fixes, ordered by priority, with the visitor data behind each one. https://convertlyft.com/Vercel

I run a one-man home inspection business, and this workflow produces the first draft of my field report while I'm still on site. It's built around a voice-to-report pipeline, and I also use similar automation for the rest of the business. How the field report works: 1. On site, I open a simple web app on my own website from my iPhone and record one voice clip per area of the house (roof, electrical, plumbing, etc.), using AirPods as the mic. 2. The clips upload to my web server, and OpenAI's Whisper turns them into text. 3. Claude reads the transcript with my saved prompt, which tells it to write each finding in my report format, by section, in a professional inspector's voice. It writes the first draft of the report. 4. A Python script on my office PC drops the draft into that job's folder and emails it to me, usually before I'm back in the truck. The rest of the business runs the same way, with Claude-built Python scripts on Windows Task Scheduler. When a job is booked, a script creates the job folder, builds my inspection checklist, and pulls property research (parcel, aerial, sketch, permits). Photos tagged with a single tap on an iPhone Shortcut get sorted into the right job folders each night. Finished jobs feed one master Excel tracker that drives my invoices on a set schedule. A morning summary email and weekly and monthly financial reports round it out. I'm not a programmer. I built all of it by describing what I needed to Claude in plain English. Together we designed a system where Claude is the brain and hands off tasks to ChatGPT, Codex and Claude Code when they're the better fit. Result: I run 3 to 4 inspections a day on my own and get about 3x the work done.

I use AI to turn a single product into ads, social media posts, and short promotional videos for my business. This is for anyone who needs to promote a product across several channels without spending hours on each one. 1. I give the AI the name or a picture of the product I want to advertise. 2. The AI studies the product and creates an attractive advertisement. 3. It writes catchy promotional captions and sales descriptions. 4. It generates creative designs and promotional images for social media. 5. It creates short video scripts and video-generation prompts. 6. I review all of the content and make any changes that are necessary. 7. I publish the finished content on WhatsApp, Facebook, Instagram, and TikTok. This workflow saves me time, improves my marketing, and helps me attract more customers.

Every year when we pull out the Halloween decorations, my wife and I end up standing in the garage asking each other whether we forgot a bin or left something packed away. This year I stopped guessing. This workflow turns one photo of everything we own into a spreadsheet inventory showing what each piece is and where it belongs. 1. Once everything is spread out, take a picture of all of it, including every item no matter how big or how small. 2. Give the photo to Claude and have it turn the items into a simple spreadsheet. 3. Review the list. Each row shows what the piece is, a short description, whether it goes inside or outside, and exactly where it belongs, such as the porch railing, front steps, living room mantel, or side yard. 4. Use the list instead of digging through bins when you want to know what you actually have. Now we can pull up the list and know what we have without digging through bins or wondering whether something is still in the attic. We plan to do the same for our Christmas decorations, since we have a lot.

I'm an executive coach and leadership consultant, and most of my business comes through inbound and referrals. That makes outreach easy to skip. Every time I actually do it, something good comes of it, but I was only doing it in the quieter weeks. So I built a simple system in Claude to keep it consistent. 1. Name what gets in the way. For me, referrals make outreach feel optional, and client work always feels more urgent. 2. Set up an "interview me" skill. Before Claude starts a task, it asks me questions about context, audience, and what I'm really after. For outreach, that means pressure-testing my list: are these the right 30 people, and why? 3. Create a project with those 30 people, the ones I want to reconnect with or get to know. 4. Get a weekly reminder. Each week, Claude tells me who I can connect with. 5. Track it live. Every outreach and every response gets logged, so I can see what's working. So far, of the last 5 people I reached out to, 4 are on my calendar with potential opportunities. More important, I'm doing outreach every week now, not just when things slow down. Claude handles the questions, the remembering, and the tracking. The relationships stay mine, and I write every email myself. Bio: Kelly Botto is an executive coach and leadership development consultant who teaches leaders how to work in an AI enabled world with four principles: Delegate with intention, Clarify the ask, Trust your intuition, Own the outcome.

I wanted to turn two characters I had already developed, Fant the elephant and Viper the nurse snake, into a children's book about illness, sadness, and friendship. I had story ideas and character art, and I wanted to know whether AI could help me build a real book without losing their identity. This workflow is for anyone who wants to use an AI book generator as a storyboard and voice test, then develop the final art separately. 1. I used my conversation-archive plugin, Archiefvis, to revisit character and series concepts from a year ago. That context helped us choose Fant's fear that his friends only like him when he's entertaining. 2. I ran Kids Book Maker with my original drawings and a page-by-page brief. It produced nine illustrated screens, including the cover, plus Dutch narration and dialogue I liked. The pacing and compositions were useful, but the art drifted toward soft pastel baby-animal illustrations instead of our bold retro cartoons. 3. I kept the page order, camera angles, warm settings, and some dialogue, which I also edited. My original artwork became the character and style reference, and the generator's pages served only as composition references. 4. I rebuilt one illustration at a time and approved useful versions as I went. Fant needed wide-open, expressive eyes because he was performing exaggerated cheerfulness. We corrected his gray skin tone and removed an unwanted tattoo. This is the prompt I adapted for those edits: «Keep the scene composition, camera angle, props, and atmosphere from image A. Redraw the characters to match the approved designs in image B: fixed proportions, expressive faces, bold dark outlines, saturated warm colors, and subtle retro-print texture. Preserve approved elements; change only [specific detail]. Add no unrequested lettering, tattoos, or limbs.» ! An extra elephant leg kept surviving my edits. Precise erasures may need a local masked edit rather than another full regeneration. 6. I made a clean, text-free cover illustration (rthag was No good) , so used an old Version and then added the title over the messed up text in Canva, because AI-generated lettering looked generic and then overly glossy. The result is several approved scenes, a much clearer visual style, and a story that holds together. The full print-ready book was finished by hand in Canva.

I run N Scheme alone, so I set up one Claude project as my company HQ. A coordinator chat takes my requests and hands each one to its own workstream thread: sales, social content, our free lead magnet, a competition entry, and even my own job and PhD search. Each thread is connected to the real tools. Apollo finds and scores prospects, Attio holds the pipeline and my daily tasks, n8n runs the automations, Webflow hosts the site, and Google Drive keeps the files. This is for solo founders who want the admin handled without giving up control. 1. Every week, the sales thread researches 15 to 20 new prospects and scores them against 12 trigger points. 2. For each prospect, it drafts a LinkedIn message and saves it as a dated task in Attio. 3. Every other Thursday, it drafts my social posts. 4. It keeps a shared memory of my rules and decisions, so nothing has to be re-explained. 5. When I ask, it updates my website. 6. For our free AI Opportunity Snapshot, an n8n workflow researches the company and writes a cited report. 7. I approve the report by email before it goes out. The rule I never break is: AI drafts, a human approves. Nothing is posted or sent without my say-so. The result: I spend my time on judgement and conversations, not admin.

My friend is a competitive runner who wears a Garmin watch for every session. Each run produces a spreadsheet's worth of data: pace, heart rate, cadence, ground contact time, vertical oscillation, and power, broken down lap by lap. Garmin Connect shows the numbers but can't explain them. It can't tell him whether a late pace drop came from fatigue, wind, or a hill, or whether his fitness is really improving. I built Stride to add that meaning for runners who already have the data. 1. Export a CSV from Garmin and drop it into Stride. 2. Stride runs sports-science style analysis. Aerobic fitness tracks heart rate within each pace band week over week, so a lower heart rate at the same pace signals improving fitness. 3. It monitors training load ratio for signs that volume is rising too quickly. 4. It builds a running economy score from form metrics. 5. It watches left/right ground contact balance, since a steady drift away from 50/50 can be an early warning sign. 6. The Environment layer adds weather and elevation, so a slow lap into a headwind is judged fairly. 7. AI Coach sends a structured 30-day summary to Claude, which returns a plain-English brief on trends, environmental factors, and things to watch. 8. What-If answers questions grounded in his own data, such as his likely 5K time on a flat, windless course or what pace to target if race day hits 25 degrees C. The result is less time staring at numbers and more time running with purpose. Stride stops short of medical advice or prescribed training plans. It delivers the insight, and the runner and his coach make the decisions. Note: the screenshots use a 60-day synthetic demo profile, so all numbers are illustrative.

I run a newsroom of one, and AI now handles most of the production work. Over the last two months I built five automated news websites using Claude and Gemini. I also automated the newsroom itself. Claude has built more than 40 automations, and it writes everything from simple stories like weather reports and sports recaps to more complicated city council meetings. Here is the workflow for news releases: 1. Claude checks my inbox every half hour for news releases. 2. Claude writes or edits the story based on the submissions. 3. Claude taps Gemini to create graphics for stories that have no art. 4. Claude loads the story into the backend. 5. I edit and approve it. I used to struggle to post eight stories a day, five days a week. Now I publish seven days a week and deliver an E-edition every morning. My daily posts have gone as high as 27 stories in one day so far. The workflow saves me hours a day.

I use this workflow to manage prescriptions for at least two people, so refills are tracked and nobody is caught off guard. It works for anyone who needs one place to see what is due, what things cost, and where the records don't match. 1. Maintain prescriptions for at least two individuals. 2. Keep each prescription organized by the date it was refilled and the quantity. 3. Track each prescription for a refill 5 days before its refill date. 4. Notify the individual when a prescription is 5 days before its refill date. 5. Track the cost of each prescription, the refills remaining, and prescription usage. 6. Highlight any discrepancies in the records. What it produces: advance refill notices five days ahead of each refill date, plus running records of cost, refills remaining, and usage, with discrepancies flagged.