Claude Mastery Sprint - January 20th 2026 Session

Full video transcript

Hello everybody. Welcome in to the Claude Mastery Sprint. We’re kicking off week three of our four-week sprint series. We’re digging into Claude, learning about different ways to use skills, capabilities, connectors, code, co-work — all the great things that Claude can do and continues to come out with. There have been some brand new advancements into what Claude is capable of in just the last few weeks. It’s incredible how fast this technology is changing, and I’m excited to tune in with everybody here today to dig into it a little bit more, with the caveat that I still have not gotten really super deep into Claude co-work because I’ve been spending way more time on my PC and it’s only usable on a Mac. But I know we do have some folks here on today’s call that have been using co-work, and a big difference maker that co-work brings to the table, in alignment with what we’re talking about learning Claude Code, is that co-work has much of the same agentic capabilities that Claude Code has and is capable of in a much more user-friendly way, which means that it can access files on your computer, on the cloud, as well as MCPs and different connectors inside of your Claude account to give you superpowers. We’re going to cover some of that as well.

I’d love to say, first and foremost, welcome in everybody here tuning in with us live, and of course those watching the replay — thanks for tuning in as well. How’s everyone doing today? Any big wins in the world of how we are actually implementing Claude in our business, big or small? The learning that we’re going to get from this type of workshop is going to come from putting it into practice, testing it out, getting our hands dirty, figuring out what’s working, what’s not. And then if you’re stuck, we have these calls to help support getting unstuck and sharing new ways that we’re all using Claude. So any wins that anyone has to kick things off? Assad, I see your hand went up. How you doing, Assad?

Assad: Fine. Actually not wins yet because I’m still working on it — catching up. But one question about connectors: when we have to configure the permissions, what should we allow and what should we not allow? Should we always allow, or should it be permission-based?

Darby: I think this is a question based on how the connector is going to be used and working inside of your process, and your personal preference for how you want it to ask questions about updating files. Let me give an example for those watching who want a visual. I’m in my connectors in Claude — I’m on Firecrawl MCP, so I go to configure and you can see I have “always allow” set, and the settings are right here. All of these are different tools that within the Firecrawl MCP the agent has the capability to use. If I wanted it to always allow, like I do right now, it’s just going to run — it’s never going to ask me for permission to stop or approve before continuing. But if I wanted it to always ask before it scraped a specific web page, I’d select “needs approval.” Then anytime it was actually going to activate the scrape tool, it would say, “Hey, I need your approval. Do you approve me doing this?”

Assad: Yeah, I was looking at it for ClickUp. I wanted to add ClickUp to my series because I use ClickUp, and I wasn’t sure whether to always allow it. Let’s say Notion, for example.

Darby: Yeah, I was going to use Notion as the example — I think it’s a better one for write or delete tools. Notion naturally defaults to needs approval. For me, it’s like if I had Claude just go update my Notion with “always allow” set, and I said “go delete this” or “go update this,” it just completely cleared out a page. I was like, “Whoa, I didn’t want you to do that.” I didn’t have a marker in place that would confirm the action before it executed. I’d say if you’re working with something sensitive, like your whole project management software in ClickUp, where you’re saying “go and do these tasks,” you might want some barriers in place just to make sure that you, as the human in the loop, confirm Claude isn’t going to delete an entire project or do something crazy. To simplify it: if it’s not changing something in your documentation or your agenda, needs approval isn’t as critical; if it’s going out to scrape and write and won’t change what you have, that’s lower risk. I think the answer is yes, with the caveat that you may also have it updating things you don’t mind it updating because you know how it will interact. But I’d start with needs approval first to make sure it’s doing what you want and has the right guardrails, in that case. In general it’s not going to take much time to click “yes, I approve this,” but be aware you don’t want it running rampant through your entire CRM and deleting a contact without needing approval first. Thanks for the question, Assad.

Big Jason, I know you’ve got a win to share about how you’ve set up co-work to work with your Mac files and enhance an existing skill. Would you be open to sharing that?

Jason: Yeah, for sure. I had no idea if co-work would do this — this is Claude Desktop and I’ve been creating, for my team, an “ask quiz survey method” skill. But the issue was the master class is so big that I was running into usage limits, and you can only upload a certain amount of files, so I kept having to restart and ask it to remember everything. I thought, huh, co-work actually has access to my desktop via a connector I’d set up. So I went to co-work and said, “I want to make sure my ask Claude skill covers everything found here — can you help?” Through the plus sign in the chat, I went to the directory, and it first said to upload files; I said, “I gave you access, genius,” and it said okay. It wasn’t just on my desktop — it was on my cloud server too, which was cool, so it can access your cloud server as well. It went through and analyzed my skills, said “okay, I have all this,” then went through the entire subdirectory hierarchy and said “I found gaps, you’re missing these elements.” I immediately recognized this really needed to be included — it found missing worksheets and templates and gave me recommendations, saying “we can enhance existing files, you should consider adding advanced segmentation. Would you like me to help: create new reference files for the gaps, enhance existing files with missing content, pull specific details?” I said do them all, starting with number one. It said perfect, went through, created the new files, enhanced the existing ones, and I just kept talking with it, and then all of a sudden it was done — 3,712 lines of comprehensive reference material. It gave me the file to download, and it was all done in probably about 10 minutes.

Darby: Wow. So you used it to create the skills, or you used it to put the repository inside Claude projects?

Jason: I used regular Claude Desktop to create this, but I was running into issues because there were so many files with usage limits and you could only upload a certain amount at a time before restarting. So I used the connection to my desktop to view all the files in my cloud directory, and it reviewed my existing skill and the directories of files and said my skill was missing X, Y, and Z, and let it fix or enhance my skill.

Darby: Okay, so you used it to update your skills, and then you took those and put them into Claude, or kept them in co-work?

Jason: No, it’s in my Claude now. In the future I would use co-work, Darby, to actually possibly start the skill creation from scratch, because it would have to make sure it had access to everything.

Darby: Thanks. Question from Bob here about how you’re reviewing and validating all the content it created — were you reviewing it as it was producing it, or did you go back through and double check afterward, Jason?

Jason: Yeah, I reviewed all the files, and then I started using it, and I’m very familiar with the method, so it was legit. I was using Opus too.

Darby: Awesome, thank you for sharing, Jason — cool to see how you’re tapping into co-work. Co-work just came out a week ago, so we’re still understanding where it fits into this process. Much of what co-work is doing, as Jason demonstrated, is similar to how Claude Code operates in terms of agentically going and researching and checking out files.

Bo Peter shared a win in the last week: used co-work to identify SEO gaps in HFS to see what linkable assets to build, then built out an ROI calculator into a deployable format, intended to be a link-building web asset.

Darby: Awesome win, thanks for sharing. Feel free to drop your questions, because co-work is still relatively new, and I think each of us will find our own uses for it. John asked for clarity on how co-work and Claude Desktop are parsing his “ask method” files — how and where it was finding those files.

Jason: If you go into settings, it’s a desktop app extension, and you can give it access to your file system to control your Mac — desktop commander. You can give it access to Excel, PowerPoint, Word, all that. You just say “go to this directory,” or you don’t even have to give a directory — you can say “look, I need to find everything on my desktop about surveys,” and it’ll look through your desktop and find it. It’s really that simple, which is very cool.

Bob (paraphrasing John’s question): Is the difference between just asking Claude to review something and asking Claude co-work to do something the fact that it can search an entire drive instead of uploading an individual file — is it the scope at which it can do the analysis?

Jason: Yeah, I think the big difference is, let’s say you had eight files, you could just upload those into chat, not a big deal. But I had this huge comprehensive course causing issues, and it was a thousand times easier and faster to let co-work find it all and make sure I had every single file, framework, and strategy.

Bob: So is it fair to say that when you click the plus icon in a regular Claude chat, desktop or web, you have to give it a file, maybe select multiple files, but you can’t just point it to a folder?

Jason: Yes, you can — in desktop.

Bob: Oh, you can? That’s why I was confused — I didn’t see anything co-work does that you couldn’t already do in desktop. That’s why I was trying to figure out what co-work is adding, though of course it’s difficult when you can’t see or use it.

Darby: Marcel just dropped a tip that co-work is now also available on the $20 account, still Mac only, though hopefully they’ll open it to PC sooner than later.

Bob: That’s interesting — I just inherited an old Mac, so maybe I’ll try it myself. Thanks for trying to clarify, Jason.

Darby: Any other wins or roadblocks — maybe you’re digging into co-work, or building stuff with Claude Code, or still diving into projects and skills and connectors and leveling up your Claude abilities? I still find myself working mostly inside of projects and using skills paired with a few connectors, and it works fine for a lot of things, especially more code-specific stuff. I am jumping into Claude Code and building landing pages and things like that. I’ve been really deep into research this past week, using Claude plus my Firecrawl MCP for a lot of research. I created a skill I shared with everybody in the community as a demonstration of how to build these skills — for anyone doing local events or outreach to press or media channels wanting to get word out about an event. I’m doing a comedy show in February, so I’ve been reaching out to local influencers, Instagram accounts, Austin-based media channels, and made a skill so everyone in the community and this sprint has access to it.

If you download this “press chief” skill file, then go into your Claude account, Settings, Capabilities, scroll down, click Add, you can upload a skill right there — you can drag that skill file and it’ll replace the skill. Now you have the entire skill inside your Claude account, and I know Jason is using these with his team, so the whole team gets access — you do it once and the entire team has access individually. Once it’s in your account, you can edit the skill with Claude and give it more context related to your business.

To use a skill: go into a project and say, for example, “execute the press chief skill but focus on the San Antonio market to find media outlets there we can connect with,” and when you click enter it will begin executing the skill. It’s important to know what connectors your Claude has, in alignment with these skills, because that will influence what tools Claude uses when tapping into a particular skill. I’ve been using Firecrawl MCP as an added research layer, and I found that it found much more specific, targeted lists of media outlets and channels than using Claude alone. To add a connector: go to the plus button for files and connectors, click connector, and you can see the different MCPs added, and manage connectors to add something like Firecrawl MCP there.

One thing I want to flag: keep an eye out for new connectors, because every day or two there seem to be more popping up natively inside Claude. One recent example — Jason pointed out a Gamma connector appeared inside Claude after we’d been finding a workaround for it the prior week. With the Gamma connector, once you log into your Gamma account, once you produce a researched press list you could say “take this list and create a presentation in my Market Sauce theme in Gamma,” and it transfers over and generates a presentation or document in that theme.

To find connectors: click “browse connectors” in the connectors panel — you’ll see options like Active Campaign, and new ones like Klaviyo appear over time. If a connector you want isn’t listed yet, you can add your own custom connector and set up a remote MCP server, or use Zapier as a workaround, though a direct connection inside Claude will be smoother and won’t burn Zapier credits.

Bruce asked: Are all skills created in Claude Desktop automatically available in Claude cloud, or do skills need to be put in GitHub to connect via Claude and GitHub?

Darby: My earlier disconnect with Claude Code and skills was having to upload them into a repository. But with Claude co-work, if you’re on a Mac, it connects the skills directly to you using Claude on your desktop, which solves that problem without needing GitHub. Jason, you’re not using a GitHub repository and it’s just tapping into all your Claude skills, correct?

Jason: Unfortunately I’m still on PC, so all my skills are stored locally on my machine. I was wondering, though, if we put skills in GitHub, could they be shared with other people that way — so a team could each connect to that same GitHub repository for the skills?

Darby: I think so — I haven’t done that because I’ve been keeping it mostly to myself, but if you have your repository and share it with your team, I don’t see why not, at least the way skills currently work.

Jason: It seems like it would be a smart way to do things, so you’re not ending up with different copies floating around, different versions.

Darby: Yeah, Claude Code came out much earlier, and co-work just came out a week ago at the time of this recording, but the trajectory right now is going to consolidate so that your skills are just accessible inside your Claude account, which makes the most sense to me. I don’t love the idea of rebuilding all my skills into GitHub, so I’m not doing it right now. Most of what I need to get done can get done in projects with my MCP connectors, and I’m reserving the time I spend in Claude Code for more production-ready website-type projects and things I’ll be working on at the hackathon this weekend, all internal, not the kind of thing needing heavy security protocols since I’m not selling those apps — at least not yet.

Omar: Any tips on setting up Claude Code while running a VPN?

Darby: I don’t have a tip for that, actually — that would be a good question to ask Claude directly.

Bo Peter: I did a clone of my skills into my computer folder — hopefully you have more.

Darby: Yeah, that’s what I was getting at with Bob’s question too — hopefully you don’t have to create skills once and then have them minimally available wherever you’re logged in. That’s why I haven’t been replicating my skills into repositories — I’ve just been cranking on my PC, using the desktop version, but Claude will presumably come out with a PC version of co-work sooner or later, at least that’s my hope.

A comment came through: “co-work doesn’t seem to know the skills set up in chat” — interesting, maybe it doesn’t pick up skills set up that way.

Darby: Any other roadblocks on setting up the infrastructure to work more effectively in Claude? Omar, what’s going on?

Omar: Not good — first, it’s really good to see you, happy new year. I’ve tried multiple times to set up Claude Code on my desktop. I thought it might be my VPN, switched that off, that didn’t seem to help — it always fails at the basic install. Is there some intermediate step I’m missing? Do I need to set something up on GitHub first, or on my MacBook first?

Darby: Have you tried the web app version, or was this new information? For getting Claude Code set up to where you can log in anywhere and connect with a GitHub account, going through Claude Code on the web app is the fast track. On the left-hand side of your screen, top left, you’ll see the code icon — click that and it opens claude.ai/code, where you can work with Claude Code connected to a GitHub repository to make edits and set up a project. We covered this on call number three, which I may have uploaded the wrong recording for in the classroom — I’ll double check that.

For a quick recap: to get Claude Code set up connected to GitHub without installing anything locally — open Claude Code, have a GitHub account (it’s free), create a new repository (add a README, can be made private). There’s also a git-ignore setup for keeping environment variables safe, covered in a document shared on another call. Once the repository exists — for example, “test card game” — refresh and search for it in Claude Code; it shows a single main branch. From there you can prompt Claude Code directly, for example: “create a simple HTML game that has two cards, one funny, one fundable — clicking the funny card shows a joke, clicking the fundable card shows a business idea, create 10 examples to test, then build a super simple app I can deploy to Netlify to play, set it up inside this repository ready to test right away.” It’s worth having more complete instructions ready before starting, refined in a Claude project with context first, then pasted into Claude Code. For non-technical users who find Claude Code too dev-heavy, getting a Mac and using Claude co-work (or waiting for the PC version) is another path — co-work adds the capabilities of Claude Code in a more non-technical-friendly way.

After prompting, Claude Code created a branch and updated the repository within about 39 seconds, and the app was ready to deploy. Claude Code can also help troubleshoot itself, read screenshots you give it, and run multiple agents at once on different branches for working on several things in parallel.

Darby: I’ve been using Netlify to get apps up and running — this entire web page for the upcoming show was built entirely inside Claude Code and deployed on Netlify.

Darby: At the very least, being good at making new skills, getting new connectors set up at the right time, and knowing when and where to use them inside projects — those three things alone will take you light years beyond where the average person is using Claude. As you get more comfortable working inside Claude, and the tool itself gets better, your outputs will tend to get better too. Is anyone feeling less confident using Claude than in week one? I hope not. Is anyone feeling noticeably more comfortable and confident?

One member: Code still scares me.

Darby: Are you on a Mac?

Member: Yeah.

Darby: I’d check out Claude co-work and probably just run with that — I think that will do everything you need.

Member: I’m having a blast with skills, but code — for some reason there’s resistance to even opening it.

Darby: I’d say lock in with what you’re doing with skills, and if you’re working with developers already, let them handle more of the code-heavy stuff and stick in your zone where you’re crushing it with skills.

Member: Is it safe to hand a skills file over to an outside developer? Is there real legal recourse if something went wrong?

Darby: I’m not a lawyer and don’t dabble in international copyright or IP law. If a contractor is going to take your work and run, litigating that can be difficult and expensive depending on where they’re located. You could have them sign something, but it may not carry much enforceability. If you’re genuinely concerned at that level, it’s worth talking to a lawyer, or considering hiring someone stateside for more protection, though that may or may not be worth the hassle.

Member: NDAs with contractors abroad aren’t always enforceable either, which makes it a real question of how much to trust the process.

Darby: That came up during Austin Tech Week at Capital Factory too — lawyers were saying once you ship your code to a developer in another country, it’s going to be a lot harder to enforce anything.

A question came in: Are there places on the web to find free or paid skill markdown files useful for a business?

Darby: Probably — some people put out skills online for others to use. During the first kickoff we shared a skill document for Market Sauce with a set of bite-sized skills and recipes to get started. In the Mastery Workshop classroom, under the cookbook and implementation resources, there’s a Google Drive link with a bunch of skills to reference — good starting points to build from and iterate on, not meant to be used rigidly as-is.

Omar: A quick update — I was able to install Claude Code through the web app, and surprisingly, when I tried again with the desktop version, it worked perfectly this time, so maybe there’s been some improvement, or maybe I was missing something before. Anybody on a Mac, it works.

A question came in: Do you have a clear way to build LinkedIn or Instagram carousels between Claude and Gamma or another tool?

Darby: There is a way to build carousel-style social posts inside Gamma, and Gamma has an integration to push directly to LinkedIn. Sabrina Romang did a workshop with us during the Claude Mastery series (part six) about using Make.com plus Blotato to automate scheduling and posting carousels with Gamma — that’s in the foundation training if you want to go deep, with a Make template available, especially if you’re using Blotato. Beyond that, once you have the Gamma connector set up in Claude — go to connectors, manage, confirm Gamma is connected — you can log in and connect it, and if you’ve customized a theme in Gamma’s theme library it’ll show up when generating from Claude.

Darby demonstrated this live: prompting Claude to write an outline and then use the Market Sauce template in Gamma to create a social media carousel for a comedy show event (Persing, doors 6:30, show at 8, live DJ until 11), including hook copy about needing a break from ChatGPT. Claude generated the outline correctly but then hit an error attempting to generate the actual Gamma presentation (“Gamma presentation not found”), which appeared to stem from a Gamma connector session/authorization issue needing to be reconnected. After reconnecting and setting the tool’s permission from “needs approval” to “always allow,” the error persisted, and Claude reported it couldn’t find a “generate-presentation” tool even though the connector showed as connected and in use. Switching to Sonnet without extended thinking and explicitly instructing it to use the “generate” tool name (rather than “generate-presentation”) also didn’t resolve it in the moment, and the troubleshooting was left unresolved live on the call, illustrating that this is the kind of connector hiccup that can occur and sometimes requires trying the same request multiple ways to work through it.

Closing out, Tim shared that he ran the press chief skill and found it excellent. Vladimir flagged that a wrong recording had been uploaded to the classroom for call three, which Darby confirmed and said would be corrected after the call. One attendee shared that a fellow community member had used a Claude skill built on B2B direct-response copywriting principles taught by John Benson to analyze and rewrite her business health-check offer copy from just two prompts (“evaluate this copy using your B2B direct response writing skills”), producing what she felt was a meaningfully improved, more sellable version. A participant also shared using the Claude Chrome extension to pull the transcript from an eight-minute Zoom recording, strip timestamps, and export it to a Word file in about two minutes, with Claude also offering to pull relevant screenshots from the video. The session wrapped noting the group would reconvene Thursday, same time, as part of the ongoing four-week Claude Mastery Sprint series.