Build a Custom Proposal Skill in Claude | Workflow Tutorial

Full video transcript

Let’s do a building out a custom proposal generator skill. I’m here inside of my co-work inside of new tasks and I’m going to ask it the question to get me started. I’m building out a custom proposal skill that I’m going to just make here on the fly. Kick things off with just a quick start instruction. “Hey, I’m looking to build a custom proposal generator skill that I can use for proposing live comedy show experiences at entrepreneur organizations here in the United States.” So, we see my initial instruction to open up the conversation and it’s coming back with, before I start building, a small hinge here: deciding whether this forks into the existing proposal skill, which is a bit more of a generic sales doc, or a standalone side hustle live experience generator pack activation options. Its recommendation is building it standalone. So it’s just pointing out what is at its disposal and then what we could do here that’s going to be a little bit more customized to our specific outcome that we’re going for.

So we can see: what should the skill take to generate a proposal? This is a point where, if you’re using an automatic transcription tool — in my case I use Granola, which will just record the transcript from calls. I have Granola connected to my Claude account via MCP, so it will reference the transcript from any calls after it loads the call after our session. Then I could bring that information in if I’m creating a custom proposal that’s based on a client discovery call that I’m having. For me, that would be useful to have the ability to pull that transcript in automatically and then initiate the process. Not everyone is going to have a transcript to feed to it. You may use tools like Fathom or Fireflies that may or may not have direct MCP integrations with Claude, but it could be a matter of pasting the transcript in and then turning it into a custom proposal. Just for an idea of where pulling in something like Granola and maybe even a HubSpot fits in, I’m going to describe what my workflow is right now to customize this specifically for this use case. I want to have a discovery brief based on our client intake call, which we have the recordings pulled from Granola, and then we want to do any sort of web recon in order to optimize the proposal for that specific client that we’re looking to engage with in one of these experiences. What I just said and relayed there is going to have a big impact on the direction Claude is going to continue to go.

Then it asks me what format I want it in — a Word doc, Google doc? Oh, look, Gamma. I do use Gamma. I have an MCP with Gamma connected to my account and I can have it go through the proposal aspect and then generate an initial draft on Gamma. I want to use Gamma and I want to use our Marketsauce theme to keep it on brand. I’ve set up a Gamma theme already, so I don’t need to spend time resetting that up. Let’s customize this per proposal buyers — conference producers — so we want to customize it to be dynamic for all of these different buyer personas that we’re working with.

Now that I’ve filled out the more specific instructions for what I need it to encompass, what tools do I need this specific skill to be referencing? I just gave it that information — I told it to use my Granola and my Gamma, two tools it’ll have on that front. Moving on to the next phase: locked in. Here’s the architecture I’m building with the three things I need to finish. It’s going to be creating this side hustle proposal generator input flow: discovery call via Granola, community organization. I have Firecrawl as well in my account, which I use regularly and helps with web scraping — so input company organization supplemental brief, and then the process would go through extracting, running Firecrawl to enhance and find more information about them online, detect the persona, scope the experience size and price using guardrails, draft a narrative. If you have any sort of storytelling framework or a certain type of narrative flow you want to follow, you can provide more information about that or add it in after your initial draft. Once this is done, it’s going to push it to Gamma via the MCP using the Marketsauce theme and then return a deck that’s editable for me to review for the final.

Question from the audience: on Firecrawl versus Claude, which from my understanding has access to the internet and can scrape on its own — have you noticed a significant difference using Firecrawl and paying for that additional service? I personally have. I find it’s just much more robust in terms of what it will go and do and scrape. I know other folks here are using it as well and have their own experiences with it. I think of adding Firecrawl as an extension the same way I do adding Gamma as a presentation extension. Gamma’s the best at generating these presentations and using their specific tools to do that — while I can generate presentations in Claude now, it’s Gamma’s specialty. That’s how I view Firecrawl as well. There are a lot of other web scrapers out there too, many of which you can get started with for free and are worth experimenting with.

Now we get down to the last bit of customization: answering these questions. What are the pricing guardrails, audience, out of state, custom branding? If you don’t have this codified, let’s codify it now, so we don’t want it hallucinating on pricing. We’ll start at $1,500 for high school, $3K for college, $5K for corporate, and it goes up from there. This is our first time making this proposal draft and scope document, so we want to create this as the first iteration that we can come back and edit later down the road. The Marketsauce theme is the specific theme name in Gamma, so I just referenced it under our custom themes. Claude’s been really good at referencing that theme name in the past, so I told it to do the same thing for this round using natural language. If Claude has a hard time finding that exact theme, I may need to go find what it was asking for to make sure it always runs this theme, but the name/ID is Marketsauce.

It’s highlighting some stuff related to what it’s pulling out and identifying — what these types of companies will pay routinely for a keynote speaker who does talks. It’s calling this out as something to be aware of in terms of pricing, whether we’re potentially pricing it too low for some of these corporate engagements. Take what you’re finding on the corporate side into account, and if there’s a reason to increase the pricing and have it structured and make sense inside the proposal document, set it up accordingly.

Now we can see it’s building out the skill. You can check its progress on the right-hand side — building the skill, creating that skill.md inside my workspace so I have a copy of it there. When we’re done with this skill, I’ll be able to save it directly to my Claude account so I have a fully customized proposal deck.

Question: I’m noticing you use Opus 4.6 for all this building and construction — can you tell us briefly why you’ve chosen Opus 4.6 over the others? For me, everything I know about Opus 4.6 is it’s the best model right now. Yes, it does use more tokens and has the highest thinking power and is currently the best model from Anthropic, so that’s why I’m using it. It does use more tokens as mentioned, so using a Sonnet or a Haiku will not have quite as much power as Opus, but they will use fewer tokens. For certain tasks, it’s probably fine to use Sonnet. I don’t really use Haiku at all these days, but yeah, it’s just the best model right now, and that’s the one I’ve been using, based on the benchmarks as well, from Anthropic as they release these. As they release future models that become available, the newest best models usually cost the most, and then as they introduce new models, models like this will eventually become cheaper to use and more people are going to be using the latest best models, and they’ll continue upgrading from there.

Since you’re on the Max plan, how long does Opus run for you before you’re getting a timeout on all those tokens? I have not yet hit the max on the $200/month plan, which is 20x. I was hitting it every now and then on the $100/month plan, which is the 5x version of Max, and once I started hitting that — I was doing much more plugin building and running other projects at the same time and building out a few different things — I went ahead and upgraded from the 100 to the 200. I really haven’t had any issues with it. I was using it pretty much 8 hours straight on Sunday and I never hit my token max. So you can get quite a bit out of it.

Now we can see: read-only file, pivot. It built it inside of my vault. I can view my skill — so now this particular skill has all of the guidelines built inside it for this proposal that we’re customizing for this particular project. Granola meeting reference — we would want to give it whatever call we want it to use as a reference document for this particular use case. I’m just going to have it create a mockup — a script scenario between us and the president of EO, or bringing this experience to Entrepreneurs’ Organization worldwide. This is not a thing right now, as much as I am working on this being a thing — it’s just an example of a mockup for how we would run this. EO is an entrepreneur organization, so I’m going to pull their website — I just searched it online and found EO Austin, so we’ll start with the Austin chapter for now.

Going to save this skill — so we have it saved and can always update it as we continue refining it. But once I have the initial starting point of a skill I want to continue to build on top of, obviously you want to test it. If you have your own client discovery call you’d want to feed through here, you could just do that. I’m just having it create a mock transcript scenario to run through this process, and hopefully finish this next run with a Gamma link I’ll be able to use and modify from the proposal standpoint.

Question: in your experience so far with Gamma, or for anybody here in terms of design creatives — what percentage of the way does it get to actual deliverability to a client? Is it 100% you’d send out without worrying, or more like 90% needing refinement? I’ve had struggles with design across Claude and Canva, which is what I use, but I’m curious about Gamma and overall design capability. For me, if I’m making something, I’m spending a lot more time making sure the content is solid before sending it to Gamma. I tend to not use a ton of images if I’m putting together a more branded Gamma presentation, so I’m more concerned with making sure it’s using the right theme so it’s on the color palette we want, but it’s usually getting about 80% of the way there for that draft. Then I’ll go through and make formatting adjustments, and if there’s an image I want to add into any section, I can add that in. Gamma’s got additional creative capabilities now with their graphic design feature — if I’m looking to really dial in one particular piece of graphic design, that’s where I’ll spend more time, versus trying to get it to create five or six different images inside a proposal document straight from Claude over to Gamma. I haven’t played around too much with Canva’s integration — I’d still say it’s probably not getting to 100% anytime soon, but it gets me close enough that I can pick it up and run it to the finish line.

That makes sense — the Canva route worked pretty well for me too, maybe 70-80% of the way there, then refining some pictures and text. I was just curious if I was leaving anything on the table by not using Gamma. Claude is also getting much better at creating popup presentations. I’m sure Gamma and other presentation platforms are keeping an eye on what tools like Claude are doing, but right now there’s a lot of collaboration you can do inside Gamma beyond the initial design concept, and more use cases beyond just getting a V1 version.

So it built out this kind of interesting transcript — basically a full fake conversation between me and the president of, I guess, the made-up president, or maybe the real president since it was researching these folks. That’s pretty wild — going from one of your teammates pretending to be that person to a full fake transcript. It’s interesting because it’s actually pulling in some names and information from my own knowledge bases here, a warm intro that I forgot I had, to potentially get in touch with this person. That’s an insight for me. It’s running through this transcript scenario, which in itself may be practice for simulating how this offer could be refined for this specific type of audience.

Now it’s at the bottom of the file, ready to test the full skill. Now that we have this transcript, we’ve mocked our way to giving it this starter point and said, “Run the proposal generator, use the side hustle proposal generator to make sure it does the right one.” Because it’s taken in the transcript, it’s going to move into the next phase of what that skill I built is supposed to do — additional research, then package the findings into a proposal aligned to this particular client example, and create the Gamma presentation draft for me to review.

Question: does this give anybody ideas for skills that would be useful to build, or how to go about building skills in general? The thing you mentioned about pulling persistent memory there with the warm intro makes me think — can I plug this into my phone contacts, my LinkedIn, my personal notes, and have it be like a network connector? I always forget — I’ll reach out to somebody and realize I know three people who know that person, and I have no way of keeping up with that. I’m wondering if it could help match that for warm intros. I haven’t tried fully syncing all my phone contacts with it yet. I do know you can connect Apple Notes, iMessage — what I’ve built out is more like a Notion CRM database where I’ve updated a lot of contacts already. That’s more of where it’s pulling that in, but I’m sure there are ways to figure out the LinkedIn API side of things, or maybe not — LinkedIn has some weird restrictions. But yeah, plugging it in with your existing contacts and knowledge base can surface potentially useful connections.

So it’s running this pipeline — discovery, pain extracted, EO audience, budget signals, decision makers. It’s running the organization recon, going in and firing off scoping and pricing. Typically, if I’m running a skill I know is going to take some time to run and automate, I have this open on one screen and focus on something else that doesn’t need my direct attention, kind of in the background. I’ve found that as I’m running workflows in Claude, it will sometimes still require me to have some level of awareness that it’s going through a process and needing certain approvals. Not every workflow requires me to approve or reject things, but how I’m batching running some of these skills has changed a little as I’ve built out more of them, running multiple sequentially just to avoid overloaded context switching between too many different projects at once, scattering too much throughout the day. If anyone else is running into context-switching fatigue working with these tools — I know I can get really excited about new updates and new things that can be done, but splitting too much attention across too many projects at once isn’t the most optimal way to batch these tasks. I’m still refining how my days flow as I add more of these things in, making sure I’m running it effectively with how it works with my schedule.

Now we can see it’s gone through this whole process — flags recon work from the transcript alone, two dates held, sponsor activation. For the sake of time, since we’re coming up near the top of our time, I’m going to go ahead and say yes, let’s approve it as is. “Go ahead and fire off the Gamma MCP render for me and give me the link.” We can see it’s using the Gamma connector, firing it off, creating the Gamma, and in just a matter of seconds I’ll have a Gamma I should be able to open up and continue refining before this gets sent off to a client.

For anybody else with their own custom proposals they’d want to generate — one of the questions it asked me is whether I have any reference points to build the proposal generator from, which is always something that, if you have them at your disposal, providing two or three different examples of what good looks like will help your skill and the content it creates get more dialed in for the outputs you’re actually looking for.

Question: is it as easy as just throwing in old proposals, old emails, from a context perspective, into the chat and saying “ingest this,” or is there a better way to deliver it? That’s probably the fastest way to get started with it — not overloading it on too much context and making sure it doesn’t have bad examples mixed in. If you have a hundred different proposal templates but thirty of them weren’t good examples, clear out the stuff that’s not a good reference point. You probably don’t need thirty either at that point — three to five, you’ll be solid. Then have that be a frame of reference for creating a framework or structured theme.

It looks like it’s done. I’ve noticed that as Claude continues to improve, a lot of times you can actually see the initial generations, like right here. So I did that — you can open in Gamma, edit it in Gamma. Let me open this document. Can you all see my Gamma app right now? Now we have a full presentation basically built out from the Claude thing we just ran through — different slides tell the story following that format it came up with, getting much better with text and words. You can see how the color scheme in the theme here matches the Marketsauce theme I’d given it. Creates the whole timeline, investment, and the more refined numbers and context I would have given it, that would have been translated over here as well. If you want to edit any images, you can go through and edit it with Gamma’s AI agent, and there are plenty of training resources on how to get started with Gamma. If you guys want to get started building your own theme, it’s pretty quick to get started in there.

But yeah, that was basically us going through the process start to finish that second time, customizing the proposal generator, modifying it to your workflow. In this case, it was taking a Granola transcript, doing additional research to enhance the conversation, and then structuring a custom proposal based on who the buyer was, who we talked about. You can run that again and again for different clients if you’re running this on an ongoing recurring basis as a skill in your account. I hope that was a useful second demo to run through today. Thank you for all the questions — I’m sure it’s useful for others going through this process as well as we’re all building out our own custom skills, personalizing them for our AI staff. Thanks for tuning in this week — we’ll be here same time, same place next week, digging into more skills.