Meeting Title: Cursor Skill Feedback Date: 2026-03-13 Meeting participants: Brylle Girang, Amber Lin
Amber Lin: Hey, hello. Brylle Girang: Hello, Amber. Good morning. Amber Lin: Good morning. Sorry, I’m a little bit late. Brylle Girang: Okay. Amber Lin: I think we have about 30 minutes. I… what I had on mine is we can talk about the… EP skills, and I also want to talk a bit more about. how skills are created, so I can see, Because I’m creating skills for analysis, and I would love to know How you do your skill creation. And learn from you on that a bit. Brylle Girang: Okay, gotcha. Before I share, would you mind, like, sharing your idea? Like, what analysis are you doing? What skills do you want to build for that, so that maybe we can reverse engineer? Amber Lin: Yeah, so, let’s… let’s see. So, the analysis usually happens like this. There’s… say, there’s a BigQuery database that has the client’s data at different tables, and… Essentially, the analysis I would want to do, then, is discovery. Brylle Girang: So that would include, say. Amber Lin: Okay, let’s look at revenue, let’s look at sales. What are the trends, over time, by different categories? Is there anything interesting we can find? Then, another category would be on marketing, so… The spend, people’s LTV, sorry, so… So that would be looking at customer value, and then how much we’re spending on them, so different… Channels of spend. And then… I think that’s the main parts, and then if there’s, like, operational data of… Profit, margin, like, that’s something we’ll look at as well. Brylle Girang: Okay, gotcha. So you’re primarily using BigQuery, right? Okay, gotcha. It might be… well, when we build a skill, the most ideal part of that specific skill is, like, the connection. With external tool. So we need to figure out if we can connect BigQuery with Cursor, and let me just try to see if that’s possible. BigQuery. Amber Lin: I can have that part figured out, I think. Brylle Girang: Yeah. Amber Lin: My question is of, right now, the skill I created kind of just runs on its own. I remember when I used the EP skill, like, sometimes they will pop up a new chat window, or sometimes they do show all the execution logs. Like, how do you do that? Brylle Girang: Okay, perfect. Let me just share my screen here. And I’m also going to share with you some skills that I’m currently working on, so that you can also see it. flow Is this… this might not be the right window, just a moment. Is this it? No, it’s not this… just a moment, I have 3 cursor windows right now, and it’s… There. Okay, there. Okay. So, this one is a bit easier, because what we did here is just we collected the problems that we have per role, and then I just asked Cursor to, like, check this, there are already recommendations. I want you to build the skills and workflows. For them, and as well as check if there are already existing skills or workflows that we can… that we can build, right? I think your best friend here would be the plan mode. Are you using the plan mode to, like, build your skills? Amber Lin: I am using the agent mode. Brylle Girang: Okay, use the plan mode, that’s good. Better, because you mentioned that sometimes the skill doesn’t really function like how you want it to be, and that’s because you don’t know the thought process of the agent when it was creating the skill. The plan mode can give you, like, a little insight on what a thought process is first. So that you can just give the go signal. So every time I’m creating a skill or a workflow, it always starts with the plan mode. So you see here, it tried to, like, collect all the information that it has. before doing something. So it has, like, a planning phase first, before actually doing anything. Amber Lin: I see. Okay. I… I remember you mentioned there’s, like, a… like a… what is it called? Quads? Anthropic Skill Creator Skill? Do you use that, or how do you… How do you use that? Brylle Girang: Yeah, I use it every time. Every time I’m going to create a skill, I use the skill creator. You can just say. When you’re building the skill, let’s say, hey, I need this skill, or etc. Use Skill Creator. Amber Lin: That’s it. Mmm. Brylle Girang: Should work. And then it will use the skill creator, twice, once during the planning, and then the second time is when it’s actually creating the skill. So, the… the good thing with Skill Creator is that it forces, like, the agent or the AI to actually look at the resources available, like. Person push it to look at the resources first. We have also built, like, guardrails within our repo. around us… around skills. Since we’re already giving, like, the people here the power to build skills, we need to be able to, like, create guardrails around the actual creation, because we don’t want a repo which has 100 skills. And, like, 50% of them are redundant, right? So, if ever… Cursor tells you, hey, this might not be a skill, this might be a playbook, this might be, like, a linear map or a map instead, then that’s totally fine. It might, it might tell you to, hey, do you want to use a command instead? Do you want to, like, create a rule instead? Etc. And that’s going to be totally fine. That’s already embedded into our repo. Amber Lin: Oh, I see. Could you also use the skill creator command to go back and fix your previous skills? Brylle Girang: Yeah, definitely. Have you, have you committed it, or not yet? Amber Lin: Oh no, I’m doing it on my personal… Brylle Girang: Okay. Repo just to experiment what skills look like, so I’m trying to see, like. Amber Lin: what else I need to learn before I start. Using our repo and stuff. Brylle Girang: Yeah, yeah, you can do… you can definitely do that. You can just say, hey. review the skill, etc. This time, use Skill Creator, and then improve upon it, and then. Amber Lin: And it. Brylle Girang: It would be really helpful if you could just give Cursor, a list of the things that didn’t work, and then the things that you want to improve, and then it will proactively search for other areas to improve using the Skill Creator. Amber Lin: I see. Okay. Then I remember you said, like, we made guardrails. Where in the repo can I read about that? I find that very interesting. Brylle Girang: Yeah, sure, let me, So it’s going to be in cursor rules, And then… Skill to workflow… skill versus playbook. This one. So, just a bit of a background on why this is formatted this way. So, the main reason why skills are really helpful, and rules are really helpful, is that these are divided into, like, two parts. You can see here this one colored in violet. and then the rest of the dock, right? This is called. Amber Lin: up front. Brylle Girang: matter. Front matter, it is the main thing that Cursor shows the agent at first. And then if Cursor looks at this rule, it reads this, and then it determines that, hey, this isn’t relevant to what I’m doing, it doesn’t read the other… the other stuff. What does that mean? It’s faster, because it doesn’t try to read everything that we have. And it will be able to scan through all of the rules, all of the skills. In a much faster way, by just reading this description. Amber Lin: Oh, I see. Is there, like, a… similar to Skill Creator, like, a role creator? Brylle Girang: The skill creator should be able to create amazing. Amber Lin: Oh, okay, okay. Brylle Girang: I can use that one. And then, it tells here, you can explore this, but the main gist here is that if people are creating skills, tell them to… provide them suggestions. Is this a skill? Is this a playbook? Or is this… A rule, instead. Amber Lin: Cool, okay. Brylle Girang: And then, when you… you mentioned earlier that as much as possible, you want, like, confirmation steps before proceeding, you can definitely embed that in the plan. You need to be specific about it, though. You can say that, hey, after you’re done with the initial discovery. I will be QAing, I will tell you to proceed, etc. Amber Lin: Hmm, I see. Okay. Sounds good. Any other sh… like, as you’ve created so many skills, any other, like, good-to-know or advice that you have, or mostly you would just say use Skill Creator? Brylle Girang: just… just… just… just let yourself try, let yourself explore. I’m pretty sure that that skill won’t be… 100% the first time you create that, but as long as you test after it’s created, give feedback. Fail, try again, etc, then it should be in a place where it’s 100 or 90%, and that should be fine. So, it’s okay if you mess up, like, the first try. That’s going to be really important for the feedback loop, and it’ll be able to, like, give you some thought about what you actually want, right? Because… Amber Lin: Sure. Brylle Girang: The plan mode is really helpful, because sometimes you just… you just… you just want the goal, and then you don’t have, like, the bandwidth to think about the actual process. Amber Lin: That is actually what I did here. Brylle Girang: So I just told Cursor, hey. These are the workflows that we want to solve, what can we build? Are there already existing skills? Amber Lin: Prepare a detailed plan. Brylle Girang: And then it gave me this plan that’s super detailed. Amber Lin: Hmm. Very cool. Brylle Girang: Just iterate, you know, fail. Revise, etc. Test and test, so just test the skills, and then that should work. Amber Lin: Cool, okay. Yeah, I’ll let you know how it goes. I’ll… I’ll test my first section on my skills, and… see how it feels, and how I can improve that. Amazing. Thank you for taking… Brylle Girang: Time to try it out, Amber, we really appreciate that. Amber Lin: Yeah, it’s something that I wanted to do for a long time, so thank you for showing me the ropes. I mean, we have 15 minutes left. Do you want to talk about the EP skills? I know, like, I tried them. I also know you guys are creating so many new things, so maybe we can start off with just telling me where we’re at in creating the new stuff, because maybe my comments are no longer… applicable. Brylle Girang: Okay, gotcha. So we have… I think this is also, accessible on your end, because this is shared with all the BrainForge teams, so… We actually created all of the workflows based on the things that we have here per role, and this included the EP, Stuff at first, but since we’re basically, how do you call that? Removing the AP role and then focusing on CSOs and SLs. Our main priority right now are the CSOs. But I do think, like, these EP roles are almost already covered by the existing skills that we have, and primarily the EP audit skill. Closes out most of this. Amber Lin: Right now, focusing on… Brylle Girang: The client success owners. And most of this are actually just, you know. having… or making it easier for the CSOs to communicate with their client, and making it easier for us to discover the risks associated with the clients. So you can see here, there are things about run the client review call. refresh account strategy, and then we added here. Like, how possible is it? What do we suggest? Does our platform support it? Etc. And what evidence should we make? For CSOs, more of… most of this are transcripts, because about client relations, most of them are verbal, most of them are messages through our Slack channel. This is what we’re focusing on right now. And, the skills that I was creating earlier, it solves, like, 6 of this. But the other 3 are not yet possible. I think it’s this… I don’t know, but 3 are not possible yet, especially when it comes to client health, because AI can just, you know. decide on that out of the blue. It needs human judgment. I would say. Amber Lin: Mmm, I see. How do you determine, say, like, column F, G, and H? How do you… I guess, how did you create this document? Brylle Girang: Through the courser, too, so… Since cursors connected to our repo, it can answer, like, the support of our platform. So, are the things in our vault already… Already enough for us to create this skill. And then automation difficulty. Can all of this be automated? These are high, because this needs human judgment. You can run an interview call with just an AI. You can… you can… Refresh the account strategy with just cursor. Cha-da. So, the main thing here is that we gave cursor The fields to look out for. And using the resources that we have in the vault, the transcripts mainly. Amber Lin: linear, etc. Brylle Girang: It was able to, like, infer an initial thought. Amber Lin: Mmm. Okay. Very cool. Also, because we also made… I’m looking at the doc now, so there’s, like… one that’s SL… well, CSL, SL, EP Task Log, and then there’s the… role-based skills. How are they different, or are they kind of the same? Brylle Girang: Yeah, so, they’re kind of the same, but for the CSSLE piece task log, this is… how do you call this? This is a lot messier, because we use… the existing documents instead of the actual stand-ups for this. So, you would notice that most of these are not being done. Because these are the things that are in paper, but the team is actually not doing. For the role-based skills, these are the things that we have… we are actually doing, and we have mentioned a lot of times. So, most of the… actually from meetings, from escalations, from frustrations, those are the stuff that we ask Cursor to, like, look into. maybe… get meetings where Otam or Robert has expressed frustration, etc. Those are the things that are mainly here. Amber Lin: Hmm. Okay, very cool. Brylle Girang: Yeah, that, I’m pretty sure that there are things here, too, about analytics. You can build upon this, you can just, you know. Add the task that you’re doing so that we can park it, or maybe you can… you can have a go at it? There are QAs here, there are, you know, summaries, refactoring, validating, etc. Hmm. also the dashboard metrics, etc. So we want to get you in a place where You can actually focus on, like, the insights specific to the client, and then let go of the things that are how do you call this? That are redundant. Maybe using SQLs, etc, or creating SQLs, creating queries. Those are the things that really want to automate for you. Amber Lin: I see. Alright, that sounds cool. Checking… These… Brylle Girang: You know a good place for you to Amber Lin: Huh. Brylle Girang: Maybe you can start with, like, creating a map of the things that Eden or Ellen. Amber Lin: Right. Brylle Girang: is doing, or wants. Maybe there are terms, terminologies that are specific to the company. And then you can just document that via the default, so that if you ask AI, you don’t need to, like. Oh, say the same things over and over again, like. Amber Lin: I see. Brylle Girang: This is not what Element wants, etc, etc. Or maybe you can create… Amber Lin: I see you. Brylle Girang: like, Their wish list, what are the things that they want to see, etc, so that when you run your own workflows, it will directly get info from that. Amber Lin: Okay, so I think it will be helpful if I create a type 1, a type of task list like this, and another one, which I think we already have, like, the field mappings. Yeah. So, I’ll go… I’ll go see if that improves my skills. Brylle Girang: Yeah, that would really help. Also, when we, you know, get more clients, when we… when you do the same things that you’re doing right now with… with Eden, with Element, for other clients, that should be a lot easier, because you have. Amber Lin: Oh my god. Brylle Girang: reference point. Amber Lin: Cool, okay. That was all my questions for today. Any questions you have? Brylle Girang: No, I’m good. I really love that you’re actually using cursor more now, so that’s amazing. Just try to find stuff to break, just to… just try to find stuff to improve. That’s what we want, okay? Amber Lin: Awesome. Alright, thank you so much. Brylle Girang: Thank you, Amber. Bye-bye. Amber Lin: Right, bye.