Four answers before you spend
The skill will not plan a campaign until four questions have answers. What exact action counts as a conversion? What can you afford to pay for one, worked back from deal value and close rate? How well does the landing page already convert, since a weak page can outweigh any targeting decision? And is the budget large enough for the platform’s algorithm to learn?
Google: structure and bidding
For search, the skill weighs two structures. Single keyword ad groups give tight control but take time to maintain. Broad match with automated bidding can find searches you did not anticipate, but only once an account records about 30 conversions a month. Below that threshold it starts with manual bids or Maximize Conversions, and moves to a target cost or return goal once the data supports it.
Negative keywords get the same attention as positive ones. The search terms report is reviewed every week, and every irrelevant query becomes a negative.
Meta: objectives, audiences, and fresh creative
Choosing the wrong objective is the mistake the skill flags first. A traffic objective buys clicks, not conversions, so direct response campaigns use the conversions objective with the pixel verified. It favors campaign level budgets, keeps a separate budget where a small retargeting audience needs spend of its own, and tests broad targeting next to lookalikes.
Each ad set runs three to five creative variants. New creative goes in when frequency reaches 4 to 6 within 30 days, and every two weeks the weakest performers come out.
Retargeting and the cost ceiling
Retargeting is split by intent and kept apart from prospecting, so its results never inflate cold campaigns. Visitors who reached pricing or checkout, visitors who read product pages, and general visitors each get their own message, with Meta windows of 7, 30, and 90 days.
The cost ceiling comes from lifetime value, gross margin, payback period, and close rate. The skill treats that number as a ceiling rather than a target, and starts at 50% to 60% of it.
Platform playbooks in the ads file
Campaign types, match types, bidding options, audience types, attribution models, and the reporting template are in the ads skill file on GitHub. It is released under the MIT License, “Copyright (c) 2026 ScayverGraphix.” The skills follow the Agent Skills spec published by vercel-labs/skills.
Ask it about one account.
Phrases such as “Google Ads campaign,” “retargeting,” or “CPA optimization” load the skill. Give it your conversion event, your average deal value, and your current landing page rate, and it can size a first campaign. Install it with npx skills add scayver/marketing-skills --skill ads.
Treat its benchmarks as starting points.
The file quotes typical platform costs and minimum budgets. Those figures change and vary by market, so check them against your own account before you plan around them. It also holds 10% to 20% of spend for tests and judges new campaigns only after two to four weeks and at least 30 conversions. Pausing a campaign, it warns, resets its learning.
A destination true to the ad
Our PPC Management service begins with the offer and the path behind it: the search intent, the ad copy, the landing page, the conversion definition, and who follows up. Changing bids alone rarely fixes a broken path.