Disclosure: I am the founder of Optmyzr. I’ll use one of our open source capabilities as an example below, but the frameworks here apply to anything you install or build.
If you’ve used Claude, ChatGPT, or Gemini for marketing in the past six months, you’ve probably hit the same wall I have. Conversation is great until you need the same thing to be done the same way every week. Then go back to copying the information template in a new window, I hope you did not forget the step, you wonder why this powerful tool still feels this manual.
Skills are what break down that wall.
I’ve written before about capabilities like scalable PPC systems and why agents are useless without access to your marketing data. This clip zooms in.
- What is skill, really?
- Where do you find them?
- How do they work across the major AI platforms?
- But perhaps most important for agency owners: how do you take existing talent and brand it as your own?
What are ‘skills’ in AI
A skill is a bunch of files that teach an AI assistant to do a certain task correctly, every time.
For Claude, a skill is actually a folder that contains a SKILL.md file with instructions, alongside optional code and reference files that the AI can process. Enter the folder once, and from then on, when the activity matches the skill, the assistant loads the playbook and follows it.
It’s the difference between telling a new hire to “check this account” and giving them your agency’s written screening process. The output gets more consistent.
Although the concept is universal, implementation varies by platform.
- Claude offers a very seamless experience, allowing you to install and use capabilities directly within the interface.
- ChatGPT makes similar capabilities available although usually only in premium Business or enterprise plans.
- Gemini remains very developer-oriented, often requiring the Gemini CLI or specialized environments, making it less accessible to the average marketer.
Because it’s easy to use, I mainly use skills with Claude, and that’s where we’ll focus.
Where to find pre-built PPC skills
Many account managers prefer to copy documents rather than write their own. They will also love the catching skills someone else has already developed. But finding them can be difficult. There is no single App Store for skills, and many good ones are on GitHub.
For Claude, the Anthropic team sends the official skills to work with things like PDFs, and Microsoft Office programs. Additionally, you’ll find growing collections on GitHub from individual developers and software vendors.
Many companies publish their own. Ours resides at github.com/optmyzr-skills.
A rule of thumb I’ve come up with: talent is only as reliable as the team that built it. A skill from a well-known software vendor and its methods is different from a single piece of information repackaged as a skill by someone you’ve never heard of.
Ensuring that the use of skills is consistent throughout your organization
This is where it gets interesting for agencies and internal teams.
In a solo plan, you add the skill to your account and you’re the only one who uses it. Perfect for a freelancer. It’s a pain in the team, because everyone has to enter everything separately, and the versions go around when one person updates and the other doesn’t.
In Team and Business plans, an administrator can apply skills across an organization. Claude has talent management at the org level at Claude for Work and Enterprise.
The practical advantage is that, with a PPC team of five people, you enter the joint research capability at the organizational level once, and every account manager gets the same version on day one. When you upgrade it, everyone gets the upgrade automatically. No more “what version are you using” in group calls.
I think of it as the moment when ad scripts stopped living in each ad account and moved to Advanced Scripts from Optmyzr, which allows advertisers to send common script code to all accounts and change script interpretation and settings in a centralized management system maintained by Optmyzr. Simple, maintainable, and scalable; all the most important things in the account groups that promise a level of quality to their participants.
The hidden white-labeling engine: Why talent acquisition is an agency’s best friend
Here is the part that should excite agency owners. Most of the capabilities are well-built folders, and the open source ones live on GitHub. Which means you can fork them, edit them, brand them, and use your modified version with your customers.
Let me walk you through the example of a 15-person agency. I’m getting the ability to test Google Ads with open source (we’ll get to that one I think in a second). The default report it generates has “Google Ads Audit” at the top in plain text. It’s useful, but generic and almost impossible to assign to a client.
What I can do in about an hour:
- Copy or download the skills folder.
- Open SKILL.md and edit the report generation instructions to change to the name of my institution, reference my logo, and use my product colors.
- Drop my logo into the skills folder so the assistant can use it when generating PDFs or HTML reports.
- Add or remove checks based on what I really care about – if I only use ecommerce accounts, I can tell the ability to save Performance Max, Seller Center, and feed life more.
- Pack and install for my team.
What comes out the other end is a branded, agency-specific screening tool that produces client-friendly PDFs with my name on them. I didn’t have to build a basic workflow. The first author has already done that. I just added the last 10% which makes the output sound like mine.
Texts were powerful because you could edit them. Abilities have the same power. Agencies no longer know which software software vendors choose to white label. If the skill is open source, you can white label it yourself in an afternoon.
A working example: the Google Ads Audit capability
As I always refer to: we just released a free, open source Google Ads audit skill at github.com/optmyzr-skills/google-ads-audit. Apache 2.0 license, no Optmyzr login required.
In short, what it does. It works in 14 categories and about 42 best checks:
- account settings
- Conversion tracking
- Campaign structure
- High performance
- Budgets
- Bidding
- Pointing
- The audience
- Key words
- School Quality
- Search terms
- RSAs
- Extensions
- Landing pages
- Industry standards
- Analyzing competitors
It asks three benchmarking questions at the start (prime goal, target CPA or ROAS, account maturity) to match the score to the type of account you’re looking at.
If you don’t want to connect anything, there’s a four-paste flow: pull four CSVs from the Google Ads UI, paste them into Claude, and the skill runs the diagnostics. The output is a list of top-5 findings with monthly dollar impact, an A/B/C grade with a breakdown of each category, a 7-day action plan, and an estimate of wasted spending.
All the principles I explained above apply to it. You can install it for free. You can use it throughout your agency. You can forge it, brand it, and have it generate client PDFs with your logo and framing style. The Apache license expressly allows that.
If you also want it to pull in live account data instead of CSVs, use multi-account portfolio rollups, and trigger auto-adjustments, that’s where the Optmyzr MCP server comes in – and that’s the paid tier. But the research logic itself is yours to use, change, and product.
What is being done about this
Choose one workflow that your team is currently doing manually. Testing, search term updates, ad copy creation, weekly report editing — anything that has the same pattern every time is a candidate. Find it or build a skill for it.
Then move it from individual installation to team use. That one change kills a surprising amount of version drift in a group.
Brand at least one skill as your own, even if you never ship the branded version to customers. Going through the process of fork and reform once changes the way you think about what is important as a “tool” for your agency. It’s easier than you might expect.
Capabilities are how a standard chatbot begins to behave like a scripted operating system for your team. Agencies that know how to install, remove, and fork them in the next 12 months will work very differently than those who are still copying information from chat windows.
The Optmyzr test is one example. There will be hundreds.
How to apply research skills
There are two ways; choose the right one for your group:
If you have a technical person on the team – an internal dev, a power analyst, anyone comfortable with a GitHub URL – install it as a Claude plugin directly from the repo. One command.
The ability is always synchronized. When we post a new check, tighten a benchmark, or add a category, your entry is automatically downloaded. For an agency that runs this across the board, this is the right approach. Everyone uses the same version.
If you’re a developer who just wants something to work – no GitHub, no command line – download the zip from the repo’s release page and upload it using Settings > Skills > Skills within Claude Desktop or claude.ai. It’s up and running in less than a minute.
The tradeoff: summary. If we upgrade the ability, you will need to grab the new zip and reload. For a solo practitioner who uses monthly testing, that’s usually fine.
In any case, the repo is at github.com/optmyzr-skills/google-ads-audit. Once installed, type / search in any Claude chat, answer a few guide questions, and get audited.
