ad-launch-checklist

Shows an agent how to plan or hand off "ad launch checklist" so the concrete output is a landing page, homepage, or portfolio. The procedure moves through Ad Launch Checklist, Hard rules, Step 1: preflight. Use it when a new assignment calls for this craft.

ad-launch-checklist とは?

ad-launch-checklist is a Claude Code agent skill that shows an agent how to plan or hand off "ad launch checklist" so the concrete output is a landing page, homepage, or portfolio. The procedure moves through Ad Launch Checklist, Hard rules, Step 1: preflight. Use it when a new assignment calls for this craft.

対応~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/mavrin-xv/ads-creative-lab/tree/HEAD/skills/ad-launch-checklist

Installed? Explore more 生産性&コラボレーション skills: steipete/gemini, steipete/gh-issues, steipete/skill-creator · View all 6 →

お気に入りのAIに質問する

このエージェントスキルを事前に読み込んだ状態で新しいチャットを開きます。

ドキュメント

Ad Launch Checklist

The expensive launch mistakes are boring: an ad set that went live before review, a budget that can never reach a verdict, 4:5 creatives cropped into reels, ads named "Copy of Copy", a pixel that stopped firing two weeks ago. This checklist turns them into gates. Platform facts age; each one in references/meta-facts.md carries a last-verified date and a 90-day TTL. Re-check any fact past its TTL before you rely on it.

Hard rules

  1. Everything is created PAUSED. Campaign, ad set and ad. No exceptions.
  2. Activation needs the owner's explicit yes to the exact list of ads. Activate top-down: ad, then ad set, then campaign.
  3. Read-only by default. If you have API or tool access, read first; any write call (create, edit, budget, status) is listed and approved before it runs.
  4. No token, no problem: without API access, produce the build as a click-by-click sheet the owner follows in Ads Manager.
  5. One change at a time once live. A launch is not the moment to also change the landing page.

Step 1: preflight

Check and write each result down. Any red stops the launch.

CheckGreen means
Ad account statusActive, no unpaid balance, no restriction
Payment methodValid, with headroom for the test budget
Page and Instagram accountConnected to the ad account, allowed to run ads
Pixel or datasetLast fired within the past 2 days; the purchase event is receiving events
DomainVerified, purchase event configured for it
Landing pageLoads on mobile, the price and offer match the ad, the CTA works
Trackingmeta-capi verify steps done if server events are used

Details and how to find each value: references/launch-structure.md.

Step 2: budget maths

python3 scripts/test_budget.py --cells 6 --cpa 25 --daily 60 --days 14

The minimum test budget is cells x 2 x target CPA: each cell needs to be able to spend about two CPAs before you can tell anything. The script also prints how many days the daily budget needs, the kill threshold per cell, and how many live ads the budget can feed. If the days exceed the window, cut cells, do not shrink the bar. Sample: examples/acme-coffee/outputs/test-budget.md. Why 2.5 CPA for the kill: if a cell were truly on target, zero purchases after spending 2.5 CPAs happens only about 8% of the time.

Live ads per ad set by daily budget (a working default, also printed by the script): under 50, 3 to 6 live and 2 to 3 new concepts every two weeks; 50 to 150, 6 to 10 live and 3 to 5 new a week; above 150, weekly launches. Too many ads on a small budget means most never get spend. Purchases can land about 72 hours late, so never judge a day-1 cell.

Step 3: structure

Default build for a small brand's purchase test (adjust only with a reason):

  • Campaign: objective Sales, auction buying, budget at ad set level so each test set keeps its own budget. Attribution 7-day click, 1-day view.
  • Ad set: optimise for the purchase conversion on your pixel, billing on impressions, broad or Advantage audience, location and age only if the product truly limits them. Placements limited to feeds, reels and stories unless you have creatives for the rest.
  • One ad set for the statics test rather than two competing for the same people.
  • Ads: one ad per concept, carrying both formats through placement customisation: 9:16 for reels and stories positions, 4:5 as the default for feeds.

Step 4: naming and UTMs

Name everything by cell ID: campaign AC-2026-05-test, ad set AC-2026-05-statics, ad AC-C01-static. The ad URL carries utm_source=meta&utm_medium=paid_social&utm_campaign=<campaign>&utm_content=<cell ID>. The same ID is in the file name, the ad name and utm_content, so a sale in the shop traces back to one concept. Never rename a live cell.

Step 5: build and verify

  1. Upload media, create campaign, ad set and ads, all PAUSED.
  2. Open the preview of each ad in each placement. Check crop, text, link, CTA button.
  3. Fill templates/test-card.md: cells, budget, start date, kill and winner rules, decision date.
  4. Show the owner the list: ad names, previews, daily budget, total cap, end date.

Step 6: activation

Only on an explicit yes to that list. Then note the time. New ads usually sit in review for a while; do not edit them during review, since an edit restarts it. Copying an ad can create a new creative, which loses its social proof.

Output

The preflight table, the budget block, the build sheet (or the API calls made), the filled test card, and the activation record.

Individual skills in this repo

This repo contains 1 individual skill — each has its own dedicated page.

関連スキル