LiftFoil Dashboard Design + Agentic Workflows

Microphone (4:35 PM - 5:51 PM PDT)

I'm gonna see if it's fucking me. You never know. Bro, I think it's me. Holy shit. Whoops. Damn, let me see if this worked. Wait, uh... check, check. One, two, check, check, check. It's me. Oh no. God damn, I'm so sorry man. I'm like, well it's your problem. That's so good dude. I was like, oh man, it's so strange though — my Google Chrome is picking up the audio. So yeah, no no, at least we figured it out.

Where the fuck did I put that shit? I'm getting in. XY, I do like your shirt. Oh thanks man. I am definitely fucking allergic to drugs, give me a fucking happy pen dude. Yeah, for real.

All right, so I just wanted to crunch out — we're gonna be meeting with Connor on Friday, right? Sounds like it. It's gonna be during the night. Okay, but that's good. And I assume that like in a worst case scenario, let's say that you hadn't had a chance to try doing the daemon stuff with the Cloudflare Tunnels. But yeah, so if we want to walk through that really quick just to make sure that it works on your end and that you've got like a LLM backend working for you. Yeah, I put — and then just copy this. So I'm going to save this and then I'll point it to this session.

It was just the deliverables that I had at the moment, the reviews that I was doing on what we had and what was missing. This one were particular things that I found — like at least all headers in the tables need to be toggleable for filter view, so you hit one of the headers and it auto-filters to that, or like ascend or descend. Totally.

It needs a dashboard. And I was wondering what KPIs we could represent for LiftFoil. The Q is good, but in terms of analytical, what they want to see, I don't know which KPIs they are looking into. I would need to read the handoff. I think they are there. I'm thinking that we can probably just throw up whatever, and then we can talk through that with Connor. Beautiful. Beautiful.

And then the ops agent pane needs to have an adjustable side if the — and you can see the orders being scanned. Instead of seeing like 0 to 37, I might check on that again just to make sure it's not a bug. But you see one of 37, like 5 of 37, 10 of 37 being scanned. It just gives like trustability. I find this really good.

Held, autoclear. Weird. I love the pagination here, seven-day clicking on these. I was going to work on it. I know we talked about it, but clicking on them so we have a detailed view — here we have a lot of space that we can work with. The barrier badges, they opened well. And then compliance — so we do have a few things like talk to OFAC and the AEDAs and everything. I wanted to revamp this. When you go to Compliances, you see each and every one and how they are being completed.

Then I see the connectors. This is the NetSuite TalkRest. And these are the external systems that are going to be wired to LiftFoil's Dispatch. So we have the Cloudflare, Anthropic Claude because that's going to be the agents. We have the recent transaction email, the ShipEngine, the EC port rate, and the NetSuite, which is the API that they have. Obviously, it would be good that these were in like toggles, so if I click on them I can have them or they paginate to somewhere else.

What else — sources. I think this Open Design is a reverse-engineered open source of Anthropic's Claude Design. So it's open source, completely open source. I was going over here what I wanted. So I stopped because I said, let me do it with Cam. So I was like, what's the primary use of the dashboard? It's going to be dispatch operators, compliance officers, operations, and leadership. I think it's mostly this. Just to see what it brought up, I'm asking for three mocks — high fidelity, real feeling, pixel type — with tech and warm, just to have some contrast in both views. Pick a direction because I don't want any brand just yet. What do you think?

Yeah, I mean, I'm definitely on the same page there as far as like, I don't think we need to — like LiftFoil's branding is what it is and how it goes and what's there. Because the success here, the success story about LiftFoil's is having that place where they can monitor the orders in and out, like where it's going, if it's following the compliance.

It's really cool. I have it right over here. This is the Open Design. Yeah, I might steal that shit. Yeah, bro. And it's a really juicy problem. I'm sure you got your finger on the pulse on this stuff since you're probably doing a lot of very — anything interesting in this repo, so I'm glad you showed me that. Yeah bro, just give it a hack. And to be honest, it's fucking annoying at the beginning, then you'll get it sorted out.

Because I think one of the issues is that everybody ends up having their own workflows, so everyone ends up having different modifications to skills or different agents. And then keeping quality in check — it's a problem because if someone's using a degraded version of a skill or an agent, then the output is going to be different. And how's their workflow? So how are they tested? How are they iterating over the problem? How are they making sure that the models are not hallucinating, that they're adhering to rules? And it maps the template of rules to the CLAUDE.md. So nobody pushes the CLAUDE.md directly. They push the template. The template is the same for everybody. Anybody has a new rule that Claude needs because of something that happened, right? They push it to this template. They update their CLAUDE.md personal. They do a push to GitHub, and then everybody pulls the latest, and then they use the skills, and it reads the template, and it passes over the update to the CLAUDE.md without any issues. That's one of the many, many, many other issues that comes with working together with agentic environments.

Yeah. And yeah, it's just finding a — yeah, it's hard. I kind of already know we already kind of solved it, but I feel it's different for each team.

Because it depends a lot on how you work and where you go to present the data. We used ClickUp, we used to use it, but then we started finding better ways to communicate documents or parts that we needed. It depends a lot — that's one of the problems that we had with my team. I don't know where you're at, but there are many ways to bake a cake and not all of them are right. Amen, literally bro, literally. It's a pain in the ass at the beginning, but once you guys set it up, it should be good.

One thing I'm thankful about is that this project in particular is about enabling non-technical publishing. An Anthropic plugin soon. Oh, what's the plugin? So let me see, I do have it in CRM. This is work stuff, but I pulled it over. It should be in — it's not here. Let me see if I pulled it over. And I do have it, right? Pardon. Yeah, it's right here because I did add it. Let me see if I added it here. No, it has to be global. And if it's global, it has to be in skills.

I don't have it. Well, no, if it's global, that means it's in your root directory's `.claude`, yeah? Well yeah, but the way that I — oh, I open my root with all my projects inside, I like to do this. So this is my root, technically "projects," so it would be right over here, like on top of CRM, it would be `.claude` — and it's not here. And this is the CRM's one, I want to check the CRM because this is where I pulled it from.

Let me do a quick — no evals, no clean-all, no requirements. I definitely see it in your `.gitignore` there. Yeah, it is in my `.gitignore`, but I'm looking for the file. Skill name: general hardening. Okay, and then copy path and let's see where that's at.

Project CRM `.claude`, bro. I was just — you were looking at it. Project CRM `.claude` and it says skills. No no no, this one I think — nope. I would love Code, yeah, there's a lot of accents, that's the thing. Graph-powered code intelligence — and do we use this a lot at work?

I found it, and after that I'm looking for the dashboard. Oh, there we go. We haven't stopped using it. So it provides a knowledge graph of your code. Agentic blows most of the time — they just hallucinate and they don't know where the what they are. Or let's say that you have a page and there's already an element created, and it duplicates it because it didn't see that it had that element created prior to, right? That it could use it. Well, Axon helps with that.

So what do I do with this? I plug it into the skill and then it looks at my tasks. It analyzes the task, it indexes with Axon, it looks at the coverage map that it provides — so where that issue or plan is. If I have security, it will pick up security.

So it's a finite state machine. It comes to mind when you said the skill degradation, because one of the degradations is you want to implode them with shit, always, and then quality lowers as well. So it goes into the finite state machine and it says, what stage am I in from the pipeline? What are the tools, and what am I going to invoke depending on the material that I have? Am I substantiated? Do I have evidence? No — then I need to gather it. Do I need to do research? It iterates, iterates, iterates, iterates. Takes all the deliverables and starts testing them one by one, taking screenshots.

That subdirectory has two subdirectories: success or fail. To success, it works. Question for you — let's think screenshot for every single validation, to throw away anything. I like to have context because then it forgets. So there's a ledger and there's a review and there's a markdown progress harness.

It's housekeeping, and then you have clean-all. So I wanted to do housekeeping first, but then housekeeping was too lengthy, so I broke down housekeeping into four keeps: GitHub issues, enterprise architectures for modularity and scalability — and it implements it for you — and then any stale process that you have in the backend, it starts closing it out. So down into sister skills, and then I don't want it to clean something that it shouldn't be cleaned. It's not explosive or destructive, but still I don't want to use it here. But if I do "hard" and "doctor" — I should do "harden" then "doctor" with general hardening. I still need to read this.

And then the three directions — do we need to choose that? I mean, I'm not a fan. Large serif headlines, mono kickers, one warm accent, confident leadership deck feel — great for exec review slice. But it's tough, you know. A web application we're building, so I think that's safe for our mocks. Ooh, let's do gold accents. Yeah. Okay.

All right, I'm going to let me see if I can have a go at it. Yeah, absolutely. Pretty sweet, and I'll show you the documents for it. I don't have them right here — I wanted to show you the diagrams and how it worked. I have it in the actual skills folder, but what works best is skills. At the end of the day, take a big skill, you break it down into smaller skills or sister skills, you create a family of skills, and then they communicate between each other. And the best way to make it interoperable with themselves is like having an FSM harness. And that's how I've seen it work.

I'm thinking like, if I were to spin up some sort of janky gate for certain tasks in the workflow — if steps one and two pass this way, `settings.json` where parameters for this stage, and it would discern which stage it's at, you know? So the classes and everything, the documents that it will start creating — this is what it is. So it reads the hardened intake, it substantiates, right? It designs. Evidence, local test results, PR open, and then ready to merge.

It researches if need be, understanding why it's not working, and most of the time it gives you a solid answer. So it brainstorms, investigates, finds itself why it's not working, and then it jumps to — the name is "Class D Bug Fix with TDD Red Green" that Claude gave it. Gets class D bug with the reproducer, presses the stage two light, stage five in both, super five. Okay, so this is the part where it starts developing and starts substantiating, and then it creates and updates the hardening intake, which is the ledger, right? You keep iterating through this stage until you go into four. All these documents are getting read. All these documents are getting funneled. General hardening. This is not the full FSM — it's actually much, much longer. You would basically break it down into these, and it adheres pretty good to this.

Yeah, I mean, and to be honest, what you start doing is putting in hooks there for commits. So every time Claude is doing a commit, do this — except from doing the fucking job, you know? It loves it. Because sometimes I do want him to go off, right? And I'm like, she's bro, she's the best practices, she's the best answers.

An ephemeral chat, which I really like as well. Code has that. Absolutely. Codex has it too. Codex, if you put like, slash side, yeah. I love that shit. Chat — like it's so annoying that you can't fucking follow up with it. I can't do over here is fork. Fuck, bro, it's so clunky. Come on, Amelie. Come on. Yeah, like, dude, use Mythos. Quite actually, bro, it's fucking insane. Fucking insane.

But to the plugin — it's helped us out to push the CMT. We did a complete CRM in three weeks, bro. It's fucking insane. They put me as the lead of the project. It was app.cmt.com.

Yeah, should be this basic shit. I think I... Matt Burr. No, it's a good one. But I do...

There's a website that I created for work. I hate it, delete it from this website. So it's basically — they need compliance for attorney-client privilege. So there's a law firm with paralegals because they need to maintain that attorney-client privilege, which is protected by — it can't be on a cloud or anything. So I needed to deal with extraction. That hasn't been used in a minute. I don't like it out. This was high peak when I was actually working on it, you know?

You can render it right here. I can zoom in or whatever, and then I could see the correct — just approving or doing anything. And I also added this history, so this is like, no, re-extract. And let's say that I approve it. I see a document right over here. And then you see the history down here, which is pretty nifty. And then if you want to export it, you can export it — it will already have my email prepared with the data extracted. I don't know why it's other. Yeah, deep blanking on opposition. Whatever, I really love this. Did this alone, only a month. The back end spinning up, local model being trained and everything.

And then I also have more — there's logs for everything that will appear here. They're not coming up because I have an idea for it. Any tenant I receive, I receive pictures — century, but more as an admin. I can see it right over here and I can see the test or the 555 and then I can get back to them.

Really, that's me. That's my boy, Gabby. Yeah. But I don't care if you see my password, I just want to make sure that I have it right. I mean... Maybe you broke off earlier. Maybe, man, I'm just kidding. Is this Cloudflare? No, it's not Cloudflare Access, I was gonna say. And he was like, yo, Gabby, this is fucked. Thank man. You're supposed to know more, right?

These pages or something like this — see that it's fitted or can paginate inline in the card. Something like this for lift foils, right? Graph, love the animations that you're at it. Them too, bro. They're so nice. Out and then the line going, I love it.

As well. And then — move this over here. I fucking hate you, bro. Huh? Let me get you out of the way.

As a test yesterday, I created this beautiful, beautiful PowerPoint presentation. I did. I think they're really going into it, they're really grinding out — like when somebody used the image, or when Nano Banana came out and it was really, really good. But hard to tell — is it AI? You don't know. No, I mean, I had some prototypes mocked up by GPT Image 2. It was like, it's an image, or source, not as preview. Okay, that's good. Full screen with it. There we go. Okay, better.

Okay, this is the index. Could be a good — let's say, yeah, most definitely. Most definitely we can add this to like a slide. You never know. From this, for compliance, better tracking. I like how it presents it right over here, color-coded bars, and then it shows the top. Okay, exception — fewer, more top-age ones.

Yeah, I think of... overwhelming them. Yeah, the second one is just a little too cluttered — a lot of decimal places and numbers that might not necessarily have any bearing on what the system is doing. So yeah, first initial thoughts, I'm definitely feeling that first one over the second.

Comments, yes. And then that's just for us whenever we move on, by the way. Oh, that sounds good. I was — the one thing I'm not sure of is like, yeah, it's just like too much here, but I like the clean — exactly, the design is really nice. And there might be some use for some sort of like summary heading at the top where it's just, this many packages were shipped, just thinking out loud of how we might use that space. But that doesn't necessarily have to be because it could be a pending process of an action that was made. Right, actions — but actions is the overall, and then you have active, and then pending. And then yeah.

All right. Yeah, it's definitely sexy, that's for sure. Really nice.

Is the calendar being accessible, easy to add all these comments? There's those are — this call is being recorded on my computer by a local model, good. It's great. Please, this comment. Okay, I'm just trying to get a little bit more familiarized with it.

So, Operator 2. Two of chat bar, it's like, you know, in the second dashboard. The one that was the carrier mix, which was cool. Worry about you going too far down one path and then Connor's like, "I'll throw it all away and do this." Not that I agree, not that we'd have to listen to him or anything, but just getting Connor's sense, like, "Oh, that's important, that's not important," and like, "Oh, if we could bring that to top of mind."

No, but I completely agree. Sometimes you can hyperfixate in the rabbit hole and it's being more costly than just doing it simple. At the end of the day, simple is better.

Foreseeable future. Types that we whipped up as we were merging from these three designs. I like this part, it's more minimalistic. The one — okay, this is the one that I wanted to talk about. Yeah, this is the one that's missing. You can hover over and see a flyover of what it is. I preferred this one over — color coded, I can understand which one it is. Yeah, I like it better like this. I'm sort of on the same page. It should be more of a comparison chart. With carrier mix from the third, I like — do the other ones have it?

Uh-huh. Made it look like a live stream. So worth building some fucking mock functionality that only a couple people will see? I don't know. The constraint will be when do you need to present it and is there actually value in it? Overall idea — receive one or two orders every 30 minutes, I don't really know. We would need to know exactly that. Volume, I guess, right? It comes with recurrence, like what is the — it's super valuable. Yeah. It has no bad — one person. So see, it's going to be an accident. Yo, this auto refreshes on XY time. So we could say that it's not necessarily seen.

Tech shop, so I'd imagine nobody's gonna be complaining if we throw a refresh button up. No, yeah, and maybe that's something where we could be — availability to reload page, you know? So it sounds like these guys are pretty like, oh, animated fashion, just like three new jobs and a toast notification, three new rows added. Make it look and feel sexy before we did a demo or anything, you know? Interesting and valuable.

Oh, can you click on the side items in that index? Like, sorry, queue and ships — like what displays, is it? Ah, fuck me. Yeah, I was just curious. I figured it out. Oh, okay. Aha. It's the same, right? Okay. I think there's the other one, yeah. Have to click one over here and then click here. Oh my god, yeah, that's like the last one, the last last last one. Well, if they are old people, like old type school — okay, Boomer. Have fun with your congested dashboard. Literally, bro. Oh, fuck me.

The design, the other design that we're sort of slating, I guess it was the third design. Here is like why — exactly, just like empty negative space for no reason. That's where my head's at, at least. I'm not sure if you've got any strong opinions here. Yeah, no, what I think is we can take the — I think that's right, of number two.

The more I see this, the more I like it because it gives me the preview of the day. I don't know how much is needed. And to be honest, this can be passed on to the public bots, so it's not needed. But that's exactly it. Like, I think it's like a card with like, here's like top of your page. Somehow good, but I'm sorry. Yeah. I'm going to go hero cards. Won't ponder anymore on that.

Like, I like that it's held for review, but I don't need that much information. Yeah, I agree on that. So now — right. And then there's what I'm going to do. Come over here. Read the index code base that we have. Sean was able to — I always forget, let me try it with you.

Yeah, Pip — yeah, I already did this. Where was the beginning? Accent UI. Oh, I did. Were you missing the A? Yeah, I'm not sure. Where do I have Axon actually? Bro, are you retarded? Yeah, for real. No, me too. Melissa Lee. CDA. City thought a — bro, I'm inside Axon, how do you say it's not?

Yeah, that's definitely where my mind would go. I'd also wonder, it's not an NPM package there, right? Huh. Yeah, I was gonna say, like normally with a CLI package, that's Node, you'd be like maybe I gotta link the package so that it's available globally, but that doesn't appear to be the case. Excellent IQ.

I've already got an indexer running in the background, but I'm downloading it right now. Bro, this dude's tripping. That should probably be "axon index." UI — set up Cloud Cursor MCP server, I think. Yeah, it's actually an initialize and then full. But it's inside there. Let me see if I can call it. Hooray!

Stop him. Stop him. So shit from Obsidian, though. And then it says the signature, the callers, who's calling, and impact. Yeah. And then the processes. I can only see the functions or everything. Beautiful. This is the tree. There's your code base tree.

Closing Segment (5:58 PM - 6:09 PM PDT)

Let's just put that. It'll know what you mean. Just a bad bat. Yeah, it has to be this one. That's not these ones — that's from my janky shit show.

Open. Let's download. No, let's do this. Okay, wait. I'm going to say that should be good. And there are second mark and third mark. I just sent you the transcript of our conversation. Love it, that impulse.

And I don't know if it's like — which is honestly a little bit less all over the place, but still — what do I need to change in the path here? Gemma. I'll put open, you remove dash opus, it'll be correct. I like it.

No, this one was just Gemma. I don't know why the fuck it's called — oh, why didn't you fucking tell me, dude? That's a — yeah, fucking idiot.

It's going to only go correct. This one working, this one working, this one has whatever, so — oh hey, go wrong. What the hell did he tell me here? All this shit. Like I'm going to risk an operator that will not abandon the CRP. Rolling 30 days. Medium minutes, this is a good one to have.

What the actual — I need to add a copy and paste button. I literally just saw that on this person's call. Why is my — you know what, to be honest, yeah. Parse.md. Bro, come on. You can paste. Thank you. Oh my God. Okay, that's good.

Oh shit, I'm trying to distill the transcript into like any sort of designer fruits. We'll deploy it at that URL I sent.

Yeah, and really quick — oh sorry, go ahead. WhatsApp with like the whole instructions shit. That's honestly like, after seeing what you had already set up, I'm beginning to think that as a cost savings measure, you already have the — connected to Claude Code. Very, very cheap so we can mock something. I just wanted to see — yeah, 100%.

Media. So actually yeah, I'm going to add that upload for agent tab. Agent. Wonderful. Aesthetic was one. Use the mock three. Use the carrier mix from Muck. Silmarine Cart, let's see how that looks.

Do them in both. The open source alternative — I've been watching you use it and it has been very enlightening, right? Yeah, you don't need to pay or anything, or if you use your Claude and you're off of limits. Serenus.

System Audio (4:35 PM - 5:51 PM PDT)

The other thing, I'm gonna see if it's fucking me. You never know. Let me see. Can you imagine? Could be. Bro, I think it's me. Holy shit. Damn, let me see if this works. Wait, wait. Oh, I think it's me. God damn, I'm so sorry, man. I'm like, well, it's your problem. Yeah. No, no, no, at least we figured it out. Where the fuck did I put that shit? Oh, thanks, man. I am definitely fucking allergic to mornings. Yeah, for real.

Alright, so I just wanted to crunch out. We're going to be meeting with Connor on Friday, right? It's going to be during the night. Okay, but that's good. Alright, so I'll just take this, I'll do this. Yeah, I put... should be it right. No, it's CFAT. It's CFAT, so you're very fat. Okay. And that's the token, that's good. And then I'll just copy this. So I'm going to save this and then I'll point it to this session.

This session was just a... the deliverables that I had at the moment, the reviews that I was doing on what we had and what was missing. This one were particular things that I found, like at least all headers in the tables need to be toggled for filter view. So you hit one of the headers and it auto-filters to ascending or descending. It needs a dashboard. And I was wondering what KPIs we could represent for LiftFoils. The queue is good, but in terms of the analytical side of what they want to see, I don't know which KPIs they are looking into. I would need to read the handoff again. I think they are there. Yeah. Beautiful.

And then the ops agent pane needs to have an adjustable side if the stack gets long or anything. Yeah, it already cleared it, but now it actually goes through each order on the first run and you can see the orders being scanned, instead of seeing 0 to 37. I might check on that again just to make sure it's not a bug. But you see 1 of 37, 5 of 37, 10 of 37 being scanned. It just gives trustability. I find this really good. Held, auto-cleared. I love the pagination here.

Seven day... clicking on these, I was going to work on it. I know we talked about it, but clicking on them so we can have a detailed view. Here we have a lot of space that we can work with. The barrier batches opened well. And then compliance — so we do have how they are being completed. And then I see the connectors. This is the NetSuite TalkRest and the... these are the external systems that are going to be wired to LiftFoils dispatch. So we have Cloudflare, Anthropic Cloud because that's going to be the agents, we have the transactional email, the ShipEngine, the EC port rate, and the NetSuite, which is the API that they have. Obviously it would be good if these were toggled so if I click on them I can have them or they paginate to somewhere else.

What else was I thinking? Open Design. Open Design is the reverse-engineered open source of Anthropic's Cloud design. It's completely open source. I was going over here what I wanted. So I stopped because I said, let me do it with Cam. So I was like, what's the primary use of the dashboard? It's going to be dispatch operators, compliance officers, operations and leadership. I think it's mostly this, but I wanted to see a mix just to see what it brings. I'm asking for three mocks — high fidelity, real feeling, pixel tight. I'm going with tech and warm just to have some contrast in both views. Pick a direction because I don't want any brand just yet. What do you think?

The success story about LiftFoils is having that place where they can monitor the orders in and out, like where is it going, if it's following the compliances. The most important is the threshold. And then a lot of fucking skills and shit that already comes in, some shiny skills as well. It's crazy. It's really cool. I have it right over here. This is the Open Design. Yeah, bro, it's really good. Let me see if I can show you the Google systems.

Yeah, bro, just give it a hack. And to be honest, it's fucking annoying at the beginning. Then you'll get it sorted out. Because I think one of the issues is that everybody ends up having their own workflows, ends up having different modifications to skills or different agents, and then keeping quality in check is a problem. Because if someone's using a degraded version of a skill or an agent, then the output is going to be different. And how's their workflow? How are they testing? How are they iterating over the process?

The MD — so nobody pushes the MD. They push the template. The template is the same for everybody, but if anybody has a new rule that Claude needs because of something that happened, right? They push it to the template, they update their Claude MD personal, they do a push to GitHub, and then everybody pulls the latest. Then they use the skill sync skills, and it reads the template and passes over the update to the Claude MD without any issues. That's one of the many, many, many other issues that comes with working together in agentic environments.

Yeah, it's just finding... yeah, it's hard. We already kind of solved it, but I feel it's different for each team because it depends a lot on how you work and where you go to present the data. Like we use ClickUp. We used to use it, but then we started finding better ways to communicate documents or parts that we needed. It depends a lot, man. It's really hard. It's one of the problems that we had with my team. And yeah, I know where I'm at. Amen, literally bro, literally. It's a pain in the ass at the beginning but once you guys see... so let me see, I do have it in CRM. This is work stuff but I pulled it over. It should be in... it's not here.

Let me see if I pulled it over — and I do have it right here, yeah, it's right here because I did add it. Let me see if I added it here. No, it has to be global. And if it's global, it has to be in skills. And I don't have it. Well, yeah, but the way that I open my root with all my projects inside, I like to do this. So this is my root, technically, projects. So it would be right over here on top of CRM. It would be like, it would be `.claude` and it's not here. And this is the CRM's one — I wanted to check the CRM because this is where I pulled it from.

But let me do a quick... prompt, prompt, prompt. There's a template, create PR, no no no no. Evals, no. Clean all, no. Requirements. Yeah, it is in my `.gitignore`, but I'm looking for the file and... Oh, skill? Name: general hardening. Okay, and then copy path and let's see where that's it — project CRM.cloud. Bro, I was just there. Project CRM.cloud and it says skills. Oh my God. No, no, no. This one, I think. Nope. Codebase index. Yeah, there's a lot of Axons, that's the thing. This one, graph-powered code intelligence. And we use this a lot at work.

I found it, and after that we — I'm looking for the dashboard. Oh, there we go. We haven't stopped using it. So it provides a knowledge graph of your codebase. So, you know, a general — most of the time they just hallucinate and they don't know what the fuck they are. Or let's say that you have a page and then there's already an element created and it duplicates it because it didn't see that it had that element created prior to, right? That it could use it. Well, Axon helps with that.

So what do I do with this? I plug it into the skill and then it looks at my tasks. It analyzes the task. It indexes with Axon. It looks at the coverage map that it provides — so where that issue or that task is going to target. And then it says what I need to modify. And then an agent will review the security part. So it's basically a harness. What I created is an FSM.

So it goes into the finite state machine and it says, what stage am I in from the pipeline? What are the tools that I have? And what am I going to invoke depending on the material? And then you want to implode them with shit. And that's error number one because adherence lowers. And when adherence to the skill lowers, then quality lowers as well. So it's a finite state machine. So it comes to mind when you said the skill degradation, because one of the parts of skill degradation is that they become lengthy.

So peer review, hard in the plan. So it summons agentic agents to review the parts that are pertinent. It creates a subdirectory, right? Let's say success or fail. So all the screenshots that are success go into success and all the screenshots that are fail go to fail. Then I have to do research. It iterates, iterates, iterates, iterates. And then at the end, it does a battle test. So it spins up a local environment that I have. Am I substantiated? Do I have evidence? No, then I need to gather it. Do I know what I'm doing? Or is my training knowledge — the screenshots and the deliverables and evidence of if it actually got done.

Everybody can have it. It housekeeps your codebase architecture. And if you don't have an architecture, it researches top-tier enterprise architectures for modularity and scalability. So I wanted to do housekeeping first, but then housekeeping was too lengthy. So I broke down housekeeping into four. It housekeeps GitHub issues, it housekeeps sessions. So it can keep tabs on what's been doing. And if an agent did something, it can corroborate it. So I do push it to our Git to have context because then it forgets. So there's a ledger and there's a review and there's a markdown progress of what's going on. I don't like to throw away anything.

It implements, starts closing it out, running and cleaning out shit. Because it's not explosive or destructive. But I've used it like four times and I've really, really liked it.

But still, I don't want to — let me see. Where the fuck did he put it? Or no, vary. Let's vary, right? Yeah, yeah. It's a fan. Fuck it.

It won't do something else because it has no task. Open the skill. There's no need to harden this. Let me see if — once I publish it and I make sure that everything's done, but it's been the actual skills folder.

Or sister skills — you create a finite state machine harness. It doesn't need to — it takes very complex ideas or skills and actually makes them work, you know? No, no, no, no. But yeah, you're in the right thing because you would need — thank you.

You can declare what are the parameters for the next stage and how it would do hardening intake. What it is — what I'm going to call class A — and then it keeps proceeding on the nine stages to produce the named artifact, right? It does either serial cases, it does a plan. And then skills invoked: brainstorming plus investigate. This is for a problem or issue or something that you need with external research if need be, or understanding why it's not working. And then it jumps to the IDE too. It gets the — it still classifies the class D bug with the reproducer, compresses the stage 2 light, stage 5 in bug supervisor.

The name is "Class D Bug Fix with TDD — Red, Green, Whatever." That's the name Claude gave it, I guess. But basically, it goes to ID 3. I agree with this, but it's a subprocess that needs to be completed. And then it adheres pretty good to this, you know, off-base eval. But all these documents are getting read, all these documents are getting funneled, and there are other parts because this is just one. So this is the part where it starts developing and starts substantiating, and then it creates and updates the hardening intake, which becomes the ledger, right?

And then you orchestrate them so that they call each other. To be honest, when you start working with the local settings JSON, like putting in hooks there — Claude adheres to hooks, and hooks are very good because they fire on certain conditionals, right? So you can put hooks for commits, put a hook for practices, choose codebase. And then once I have what it came out with, I'll study it and I'll be like, "Ah, right." Or sometimes I'll do fork on Claude. It's so fire, bro. But fork — it could be, man. And it says, "Oh, you can resume the other conversation by using this string." Bro, I'll do it, fuck it, I'll fork, right? Fuck it. I'm in a new conversation, but I still see the old conversation. I don't wanna see the old conversation. So clunky. Come on. Come on, Emily. Come on, bro.

Kind of back to it — it's super over the top, that skill, the app, CMT. Tiger teams, team at work. Ah, boom. I changed this — change it again? I think I forgot a password. Not going to do that. No.

Look, I want to show you this one. There we go. So it's basically machine learning, AI extraction for paralegals. So there's a law that's regulating AI — well, let's say IPA or whatever — but like, spinning up a GPU server so that I can download a model, then a minute, let me move this out. And yeah, 30 days or anything. There we go.

Right over here — the correct or approved document with all the fields that got approved, with their percentage of confidence. What were the approvals or what were the corrections? Blah blah blah. And then you see over here, I see it approved and I see a document right over here. And this is the docs — approving or doing anything, right. And I also added this like history. So this is like, "No, remember that I told you this," or whatever. Then if you want to export it, you can export it to, let's say, directly. And then boom. Exactly, the permissions and everything. It's really, really nice.

This was my project manager. Took me a — I did this shit. Yeah, that's for this project. And then there's things that appear here. They're the parameters on the models and then making it more cost effective for any tenant that wants to use the app. And then the feedback that I receive — I receive pictures — the training signals is what's coming next. It's because I have an idea for training different models and then lowering the feedback, right? With new or received, it's like a type of Sentry but more in the UI. Admin, yeah, which is really nifty.

To be honest, I really — it is me. Yeah. But it was intense. It was intense. I made it through and then they gave me the fucking CMT, which I hate with my life, and I'm dying, bro. I swear to God. Amen.

Can you make it work? See that it's fitted over here. Something like this for lift foils, right? Lines, right? I love them too, bro. They're so nice. Like taking them out and then the line going beep, beep, beep, beep. It's like, fuck you, man, you're supposed to know more. But yeah, I was thinking about that.

Okay, so let's open. Let's see, open full screen. But did you like it? Because Claude is not very good at being creative with frontend. AI to create a frontend — and with the ChatGPT 2.0 now, the image, like, it's really hard to tell, is it AI? It's getting insane. But I can do — that's good. And I'm just still really mad that I can't fucking — okay.

Yeah, we can add this to like a slide-over and then we can ponder on this style for — pass rate, I like pass rate. Here, color-coded bars, and then it shows the top exception queue, needs, shipments, and flight DG compliance 93. Okay, exception at DG rules. I like this for compliance better — tracking the platform connectors and notifications, that's good. Audit chain and attestation, not bad. A landing page that they wanted, which I doubt, but you never know. From this I like the dashboard, queue, shipped, calendar, dispatch. You know, I am clean.

Okay, this could be a good — let's say, yeah. You fucking asshole, stop. Line, let me move you out of the way. You don't know. Near grayscale, blah blah. These are $42 average — the fuck up. But this is more like mono, you know, it's more like code. Yeah, I agree. Normally I'm not going to battle with it. Whatever. I'm not going to — yeah. Fuck this.

Okay, and then the third one. This one is very clean. Too much here. I like the clean. Actions to take, and that would be — mist, pending. Process of an action that was made. Okay, those three could be.

Right, yeah. Well, there's the index. Let's see the dashboard. Huh, this one. And I want to add them. I think that should be it. Don't know if this is needed. Yeah.

Jesus, this stuff. Like them. Oh, DG compliance. Let me see — from operator two, two of the three holds are a lot, 24A of the — beautiful, beautiful. That's good.

Describe this design you want. There's a comment. No, boom. Oh, cool. Thank you. I agree. So they can see the intentions, platform, and application, which would be good. Hero cards, fine. This is more minimalistic. I believe nowadays that like the sidebar — I would add the sidebar over here, a button, right? I would like this as well. Just click them all. Yeah.

Carrier mix from third. I like — yeah, so let's do this. I'm going to pick the first one. No, no, no, no, no. And I get it color-coded.

I can understand which one it is, but yeah, I like it better like this. I can see the live stream, a trip.

What's the overall idea? What's the constraint? When do you... Thank you.

Yeah, so we would need to know exactly that variable. The variable is what is the rate, but volume comes with the rate. So if they have a mid to low rate, there should be a reload. Yeah, it doesn't give him — yeah, it only does three mocks, it's currently static.

Ah, fuck me. To be honest, I really like the other ones. Oh my God. Yeah, it's like, whoa. This is awesome.

Okay, boomer. Have fun with your congested dashboards.

Literally bro, oh fuck me. Well, if they are old people, like old school type, they might like this. Yeah, why? Aesthetic of number one. Number two, with carrier mix from third, and the more I see this, the more I like it.

Yeah, no, what I think is we can take the — yeah, yeah. Thank you. To be honest, this can be passed on to the bot, so it's not needed. I'm just not going to ponder anymore on that. I don't like this.

Yeah, exactly. What I want to see is — I don't want to ruin the mux here. Let's do a prompt. Come over here and then I'm going to go over here, call Axon, give me the number of pages and services. And I should be able to — I always forget their command. I actually never tried it. So I'm going to try and wait. And there we have it.

Also, Axon should be on. View sample flows I need: refactor user class, knowledge graph model, fine, fine, fine, this is good. Where was the command? You're prone? Axon UI. Okay. Fuck you.

Ah, yeah, it's not global. I have it out. Axon UI. Bro, are you retarded? Axon UI. Bro, I'm inside Axon. How do you say it's not? Maybe I'm having PowerShell issues. Yeah, for real. No, me too.

LSLA. It is here, right? I am looking at it. I didn't talk to you. Blockages. Pip. Yeah, I think it was pip install. IQ. But it did use PowerShell.

That's probably it. Next: impact, dead code, cipher, watch host, UI, setup, cloud, cursor, MCP serve. I think so it should be now. CD back. But it's inside there. Let me see if I can call it. Right, pip install just in case. Excellent. Suspicious detected.

A database, shit. Parse output and handle streaming, and then it says the signature. Yeah, so this is Axon. Yeah, it's some pretty nifty stuff. Is that a bass? It doesn't like anything. I was doing an SQL migration and it pinged me, yo, someone's modifying — what's code right now? Let's see, you have functions, class. I love it.

Closing Segment (5:58 PM - 6:09 PM PDT)

It has to be this one. There we go. Go fuck yourself. Let's do this. No information. Okay, site's been good. And the third one. Okay.

With Platypus. Fucking genius. Transcript. What do I need to change in the path? Here, Gemma, I'll put Opus. Thank you. I like it.

Then I have this one. Let me just revert. Okay so, yeah, it's actually — OK, I'm going to do it simple. Yeah, I'm going to do it. Bug it.

Bro, I didn't paste anything. What the actual fuck? Oh. To be honest, yeah. Parse.md. Copy pad. You can paste. Thank you. Oh my God. Jesus.

Yeah. Okay. That's good. Oh, shit. I just gave it a path, and then I moved it out of the path. Thank you.

No, no, no, go ahead. Yeah. To be honest, it's going to depend on which environment we're going to run in. We're probably going to run it on yours. We're not going to push the environment. Yeah. Thank you. Or anything, so if you already have a Gemma model there, but responses need to be formatted.

MMS — it would be good to process MMS. It would be nice to add an upload. I'll be right back.

Over here. Let's do one more mock. Aesthetic wise. Mock 1. Carrier mix from Mock to Summary card, let's see how that looks. The graphs like Mock 1. I wish I could do comments inline, man. Now, use the compatibility from Mock 3. Right? Yeah.

You don't need to pay or anything. Or if you — Claude, and you're out of limits, shut the fuck up. Settings, usage, because I like that they put it right over here.

Combined Microphone + System Audio transcript. Use the tabs above to view separately.

LiftFoil Dashboard Design + Agentic Workflows

LiftFoil Dashboard Design Decisions

Mock Comparison & Selection

Design #1

Design #2

Design #3

Final Direction: Hybrid Approach

Primary Users

UI Gaps Identified

External Connectors

Compliance View

Live Update Strategy

Friday Meeting with Connor

Design Tool: Open Design

Executive Summary (Opus)

Executive Summary (Gemma)

LiftFoil Dashboard - Design & UX

Connor Meeting (Friday)

Infrastructure & Tooling

Transcript Tool & Agent Tab (from Closing)

Microphone (4:35 PM - 5:51 PM PDT)

Closing Segment (5:58 PM - 6:09 PM PDT)

System Audio (4:35 PM - 5:51 PM PDT)

Closing Segment (5:58 PM - 6:09 PM PDT)