Stop Re-Explaining Your Project to Claude Code

Added: 2026-05-28

[00:00:00]Every codec code session starts with five to 15 minutes of context establishment. It requires manual digging through all conversations finding where things left off and my most common prompt was hey what I was working on. You close a clo code window and the context is lost. So you need to reestablish it and you do the same work again next time. Okay, how do I find those two files? Which files did I create? What decisions were made? and all of that you need to explain every session and this really wastes so much energy. So you have a a bunch of cloudcut code sessions and there is no end to them. There is no way to capture a state of of the work on this project and the next day you just close everything. It gets too crowded and you start everything from scratch. In this video I'm going to show you a three things. A session file that survives across your clo conversations. Obsidian dashboard that gives you a live project state instead of stale cloud.mmd and a handoff prompt that transfers exactly what matters to your agent. Okay, a cloud code session is just a single chat conversation which you have here in a terminal and once the context gets filled up as in my case I can check my live context with a context command. I spent more than 200k tokens in the session. So the context window is full but my work is still not over. And a session file helps you to keep track of across multiple cloud code conversations. Here's example of how the session file can look. The session file is just a markdown file which contains a goal which wanted to achieve the background context on this project. The definition of done and here the core idea is to capture the progress as we work. Here on the right we have those cloud code sessions. Okay. So once the context window filled up we update our progress status in this node and then we transfer the context to another agent to another cloud code session to start fresh. So this way you can capture your work across multiple working sessions multiple days. A very simple example I have this project. I'm developing a course right now and I have those working sessions inside of this uh dashboard and it displays me my sessions where I was working and uh here we are capturing the progress on what I've done in that session and those sessions file can really span multiple cloud code conversations multiple days of work even multiple weeks of work for each of cloud code sessions you have its own session file Well, with the ability to track the state of the project which you're working on. So that's very cool. A session file for your project is still not enough to capture the complete state of your project. And this is where we introduce this concept of uh dashboards which enable dynamic memory. Those dashboards they can contain the project state uh work in progress current tasks sessions context uh which already shown your working plans and this something that changes all the time depending on the current state of the project and I want to make a clear distinction between static memory which is claw.md which captures points which don't change much for example your working style who you are what are your preferences how do you like to interact with AI that's clcmd and here I shown you session file example of a session file and you can see that it's being linked to this dashboard so and this dashboard contains overview of the project active sessions uh plans open tasks this is example of of a simple dashboard and here what I really love that uh what we are doing here is we're using the data view plugin to look for the files which contain link to this dashboard and you can aggregate all of the sessions, all of the plans.

[00:04:07]Here is the plans and you can do it for any type of projects for any type of your files. You could also have here a reference documents for example and this is supported all within the single dashboard. So to demonstrate how this is working, what I can do is uh I can provide this u path to this dashboard and tell to read it. I'm telling that okay so let's load the state of this project it uh reads a smartman file it sees that it has those queries and then it resolve them through this data core plug-in data core actually in data view are u similar plugins um we have similar capabilities now the most important piece that we dynamically resolve those queries we have plans we have all attached session files and now agent can see okay what we were doing there And this way we just loaded u everything into our memory and we can continue working from where we stopped and that's been a dynamic memory enabled by the dashboards. Yeah. To sum it all up we have a markdown node and we have embedded uh views. It could be bases. It could be data view with all of our sessions with our tasks with our plans.

[00:05:20]Then we can ask clot to resolve those queries and in the end we just get a table loaded into the clo context. You can ask it to read the files or not to read the files. It's our decision. But the main idea that we are here aligned with clot. Now clot exactly the same picture as I do and I can continue working where I left off. It's very very convenient. and I use it all the times and I recommend you to audit your quad autom to see what's uh stale there and update it. Move out all of your project so many changes dynamically into those dashboards to make sure that you always have up to-date memory. Now we have those sessions and dashboards for tracking. That's very good. But actually what happens if you hit uh your context window limit? Right now I'm here at 200k tokens and there's this concept of a dam zone. Here is a diagram. Okay, you can imagine that this is 1 million context windows. This whole box here you have loaded your systems uh plus memory instructions and you keep executing you keep uh chatting with a clot until you hit uh this this boundary. I define it as uh 200k tokens. That's my rule. If you're using software engineering, the rule might be a bit more strict. Some people say, "Okay, it's high key tokens." It's somewhere in in that zone where clo starts cutting corners. It just becomes lazy. It becomes um forgetful and continue working beyond 200k tokens is just uh harmful. So, and what people do here, there are a few ways. Um a standard way is to do a compaction. If you're familiar with the compaction, so there is this compact command uh free up context by summarizing the conver conversation so far. So this command is great and it solves this issue but I have a few problems with that command and one of the problems is that um I can't capture my learnings right now with that command. So how do I transfer the context to another agent and I I want to write this information in my obsidian into a session file. That's the first piece which I'm missing. Second piece is that this command is not flexible. I can't change the prompt for compaction because it's um a custom slash command which I have no control over. and say what if I want to have a to start a new session to present a plan what was done before and what is the proposed plan for the next steps let's say you've been working uh with a cloud code session you hit your 200k tokens then what I do from there is I run this handoff command which uh captures what what I've done in that session what are the next steps and what's the intent for the next session so let me just demonstrate that and before closing my session I also on uh retrospective skill which I showed in one of the previous videos to capture the learnings. uh in this session I was uh doing the outline for my video preparing the presentation and uh there are a few moments which u kind of went wrong and you can see that okay cloth captured 12 corrections and I'm saying okay this is bad um this is AI slope and we are analyzing what is the work which we've redone and then we are proposing the additional changes to our skills and here is the proposed diffs which I agree with uh and I tell let's apply those divs and this is extremely valuable so that next time whenever I go through this uh workflow I don't have to explain again that it needs to be done in this particular way it's been extremely helpful now we added uh few pieces to our memory and we are at 23% of our context window and right now I'm going to run this end of command and I'll tell continue um and by continue means if I want to work in on that project I type continue. If you want to park this, I type park. But let's try uh continue. And that's my most common scenario. What it's going to do is it reads this workflow on how to continue work. Now it analyzes if there are any uh comments left, if there's any feedback. It also grabs my CMAX workspace ID. I use this terminal called CMAX. It's uh very useful and you you can see how useful it is right now uh in the in the way that agent can actually programmatically interact with this terminal. Uh it can type to any terminal window which is open here to any surface and I use this to restart my work. Right now agent is generating um a handoff. It wrote this handoff file. Right now it schedules this auto continue. So I'm not uh touching anything. Right now agent from me just uh types um the next prompt and starts a new session and it tells another agent to okay here's a handoff follow the instructions and this is continuation from the previous session and this is what was done here is the current state and here are the proposed next steps that's the power so now we are that 4% uh context window we are just restarted everything and we can continue our work this way. Well, if I want to park this, I would type hand off park and then we would update our progress into a session file in Obsidian. And this way I can capture uh the current state of the project and then whenever I want to continue on in this project, this is a video. I can just u read it from the session file in Obsidian directly. And now that you know how to never lose your context ever again, you might also find it useful on how to build a dashboard for your life.

[00:11:10]Here's example of mine dashboard. That's the next video.