I just applied the five-part framework to a prompt I needed, and here's how I structured it: Identity: You are my email triage assistant. I lead IT procurement at a large company. Task: Read this morning's unread emails and build one review page for the whole batch. Don't ask me anything item by item. Context: My standing rules and the review-page fields are in REFERENCES​.md; where the last run stopped is in CONTEXT​.md. Constraints: - Draft only, never send. - Never answer a supplier's commercial question for me; summarise it and recommend. - Newsletters and notifications go in one line at the bottom. - If a rule already answers a question, follow the rule instead of asking. Output Format: One table, one row per email: from, two-line gist, the ask, deadline, recommendation, draft reply, link, and an Approve / Change / Skip column for me. I added all five because each one fixes something that kept going wrong. Identity stops it writing like a generic assistant. Context points to files instead of me pasting background every morning. Each constraint blocks a mistake I've corrected more than once. The output format is the big one: one table means I decide everything in one pass instead of answering questions one at a time. The structured version gets me a page I can clear in one sitting, instead of a stream of "how do you want to reply to X?" questions with no context. My old approach was "help me with my inbox". The framework made me think through what I actually needed to decide.

I created my first three-file folder system for my email triage. My CLAUDE​.md included my role (IT procurement lead) and the run rules: gather first, ask once, every item carries its own context, draft only. My CONTEXT​.md explained what I'm building (email admin from hours to 30 minutes), the problem today, and a handover section. My REFERENCES​.md provided my standing rules, the review-page fields, a made-up example item and my voice. When I asked Claude "I have 40 unread emails and one supplier wants a renewal price confirmed by Friday. How would you help?", the response was different because it went straight to my routine: one review page for the batch, a two-line gist and draft reply per item, newsletters folded into one line, threads already settled marked closed, and chasers listed. Without the folder it asked me to paste the emails and offered generic buckets. Before, I had to explain every time how I want the batch laid out and what context I need to decide. Now Claude already has that information. Key takeaway: the rules section is the part that compounds. Every ruling goes in once, and that's one less question tomorrow.

Quick reference in case you run into any issues getting set up or using the Clief Notes Agent + Roadmap. Go to Start Here → How to Access your Clief Notes Agent & Roadmap and follow the steps. **Browser Compatibility** The Clief Notes extension only works on Chromium-based browsers that support Chrome extensions. You can use: - Google Chrome - Microsoft Edge - Brave - Opera - Opera GX - Vivaldi - Arc Safari and Firefox are not currently supported. For the smoothest experience, we recommend using Google Chrome. **Troubleshooting Guide** - Don’t know your Skool @username? Go to your Skool profile → under your name, copy and paste your @handle. See the image for reference. - If you lost access in the extension: Most of the time, all you need to do is sign out and sign back in using the same email you originally used. - If you see the Creator Dashboard: If you ever get logged out or land on the Creator login, simply switch over to Student and log back in using the same email. - If the extension is gone: Log out and log back in with the same account. - Important: Use the same email every time. That is what keeps your profile, roadmap progress, quiz data, and AI personalization connected. If something goes wrong, you get stuck, or your question isn’t covered here, comment below and tag me or @Nazar Tymiv . We'll help you get sorted.

I just watched the Jake Van Clief breakdown, and the biggest thing that stood out to me was that every chat is a chain of decisions: goals, constraints and assumptions, from both you and the model. It clicked because that's exactly where my time goes. Every morning my agents ask me the same kinds of questions about my emails, and my answers disappear into the chat. My main takeaway was that the dialogue is the raw material. Pull the decisions out of it into markdown files, and the next request drops back to layer 1: one short prompt that calls the whole tree. His "tighten this paragraph" example made it obvious. Once the voice and constraints live in a file, the back-and-forth goes away. The experiment at the end helped me understand that the same applies to meetings: a transcript can turn into actions before the call is even over. Going forward, I'm going to have my email triage agent write every ruling I give it into one rules file, with one home per rule so they can't conflict, and read that file before it asks me anything.

Behind the behind, episode 4: The night the AI got cut off The fourth episode of the mini-series about what I learned building Behind the paper plane, made by the same workspace. 4 minutes 41 seconds. Watch it: https://youtu.be/RpSAC11pX7M Episode 3, Ghosts in the machine: https://youtu.be/aOPweB4hEuI Episode 2, The AI checks its own art: https://youtu.be/Bnm71iXm8QA Episode 1, Thirteen messages: https://youtu.be/C2POk9nGRxE Behind the paper plane: https://youtu.be/fXEyIoVjOeM THE IDEA An AI session can stop at any moment, and a new one remembers nothing. Not the night, not the plan, not one word I wrote. Everything it knows, it reads. So the build carried on for one reason: everything it needed was written down in files. The build's log, the review log and the saved steps. THE NIGHT - At 23:59 I asked what happens if the AI runs out of usage while I sleep: "do you just stop and wait for me to start you again or do you start your self?" The answer: it stops, and it cannot start itself again. - It could set a timer and check on the build every hour. I asked how a check would carry on the work. - Each check reads three things: the build's log, where every step is written down; the review log, where every decision is; and the last saved steps. Then it finds the next unfinished step in the plan and goes on. - I said yes, asked for every 30 minutes, and went to bed. - The timer rang 11 times through the night, and the helpers kept building. Just after five, the film's first part was finished. - At 05:49 the usage ran out, in the middle of the work. 63 steps had been saved since my last message. 4 pieces had not: 2 scenes for the second part, half written, and 2 new styles for the fourth, one nearly done and one barely begun. They were on the disk, and nowhere else. - The timer rang at 06:13, 06:43, 07:13 and 07:43. The only answer each time: limit reached, back at ten to eight. It rang inside the same session that had run out. It could knock. It could not open the door.