There are 30 skills that plug into Claude and handle operational ad work — the stuff that eats your week but doesn't require creative thinking. CPA diagnostics, wasted spend, budget modeling, creative fatigue, reporting, anomaly detection, and 24 more.
They work on Google Ads, Meta Ads, or both. No separate platform. No dashboard. No login. They run inside Claude, connected to your actual ad account data.
This is the GitHub repo with all 30:
What these skills actually are
Each skill is a prompt-based workflow that takes your ad data and returns something useful. Not generic advice — specific analysis based on your actual numbers.
You feed it a data export or connect it via MCP, and it does the work. Some of them replace 30-60 minutes of manual analysis. Some catch things you wouldn't find until Monday morning.
They cover three categories:
Diagnostics
Figuring out why something changed (CPA spikes, anomalies, wasted spend)
Analysis
Breaking down what's happening (geo splits, device performance, audience overlap, attribution)
Operations
Doing the recurring work (pacing, reporting, keyword mining, creative fatigue checks)
How to set them up
Clone the repo
Browse the skills
Each skill is a standalone markdown file with the full prompt and instructions
Add to Claude
Add them as project knowledge or use them as individual prompts
Connect your data
Paste exports directly or set up MCP for live account access
No API keys to manage. No third-party tools. If you can get your data into Claude, the skills work.
Setup help →Skip the setup — Ryze AI connects your accounts directly.
The 6 that come up the most
Out of 30, these are the ones that get used daily.
1/ CPA Diagnostics
CPA spikes on a Tuesday. Instead of opening 15 tabs and cross-referencing campaign, ad set, keyword, and audience data manually, you feed Claude the current period vs previous period data.
It isolates what caused the change — audience fatigue, bid landscape shifts, creative decay, landing page drops, budget distribution changes — and ranks them by impact.
Real Example
CPA went from $42 to $67. Claude identifies that 73% of the increase came from one ad set where frequency hit 4.2 and CTR dropped 31%. Another 18% was a search campaign with a broad match keyword matching irrelevant queries. The remaining 9% was a landing page with slower load time after a site update.
You get a ranked list of what to fix, in order, with the numbers behind each one. What used to take 45 minutes takes about 30 seconds.
2/ Wasted Spend Finder
This one scans your accounts for money going to search terms, placements, audiences, and ads that produce zero conversions. It groups the waste into categories and hands you exclusion lists ready to upload.
Not just a list of bad keywords — it groups them by theme (competitor names, informational queries, wrong intent, geographic mismatches), attaches dollar amounts to each group, and recommends match types for your negatives.
Real Example
One run across 12 campaigns over 60 days found $6,800 in search term waste and $1,200 in dead Meta placements. That's $8,000/month recoverable spend sitting there, found in under a minute.
Should be a standard part of monthly account hygiene. Most people do it quarterly if they're lucky.
3/ Budget Scenario Planner
Client wants to go from $30K to $50K/month. What happens to CPA?
Most people guess. This skill uses your actual account data to model diminishing returns at different spend levels. It accounts for auction dynamics — scaling 50% doesn't mean 50% more conversions.
Real Output
The first $10K increase brings CPA from $38 to $43 (within target). The last $10K pushes CPA to $58 because you exhaust the high-intent audience. Recommendation: scale to $40K first, test new audiences, then push to $50K.
Useful for any budget conversation with clients or leadership where you need data behind the recommendation instead of "I think it'll be fine."
4/ Creative Fatigue Detection
Most people catch creative fatigue after performance has already collapsed. This skill monitors CTR trajectory, frequency accumulation, and engagement decay across all your active ads and categorizes them as urgent (replace now), warning (1-2 weeks left), or healthy.
Real Example
Running 14 ads across 5 ad sets. Claude flags 3 as urgent — 40%+ CTR loss over 8 days with frequency above 3.5. Four more in warning status. Seven still healthy. Your top performer from last month is dead — 6 weeks running, audience has seen it 5.8 times on average.
Run it weekly. Run it before creative review meetings. Run it after scaling campaigns because higher budgets burn through creatives faster.
5/ Client Report Narratives
You paste in the numbers. Claude writes the executive summary paragraph — the one at the top of the report that explains what happened and why. The paragraph every client reads and nobody wants to write.
Feed it
$45K spend, 1,180 leads at $38 CPA (down from $44), ROAS 4.2x.
It returns
"March delivered 1,180 leads at $38 CPA, a 14% improvement over February. The primary driver was the new UGC creative set launched mid-month which outperformed static images by 2.3x on Meta prospecting. Google Search held steady at $29 CPA with branded terms contributing 34% of volume. One area to watch — Meta retargeting CPA increased 18% as the audience pool shrinks, recommending a lookalike refresh for April."
Not ad platform jargon. Plain English that a non-technical stakeholder reads without follow-up questions.
6/ Anomaly Detection
Catches unusual performance changes and flags them with context. CPC spikes, CVR drops, spend surges, impression collapses — compared against your 7-14 day rolling average.
Real Example
Thursday at 2pm, branded search CPC jumps 47%. Claude flags it: a competitor started bidding on your brand terms (impression share dropped from 94% to 71%). It also catches a Meta campaign CVR dropping 60% — the landing page started returning 404 errors after a site deployment at 1:30pm.
Each anomaly comes with severity rating, likely cause, and recommended action. The goal is catching problems in hours instead of discovering them during your next routine check.
All 30 skills
Here's the full list. Each one is a separate file in the repo with complete instructions, examples, and setup details.
Want this managed for you? Ryze AI runs these automatically.
Getting started
Clone the repo, pick the skill that matches your biggest time sink this week, and run it with your data. That's it.
Full repo — 30 skills, free
Each skill is a standalone markdown file. Clone it, add to Claude, and start running analyses.
View on GitHubSetup help
Email hello@get-ryze.ai
