Skip to content
AI stack, skill

user-acquisition-skills: agent skills for paid UA and web2app

Last verified

user-acquisition-skills is an MIT-licensed pack of 15 agent skills by Vlad Yasinski for paid UA and subscription growth. It covers Meta launch planning, learning-phase audits, creative testing, kill and scale rules, paywall audits and attribution checks. Install it with npx skills add ua5er/user-acquisition-skills.

Facts

Publisher
First party: published by the owner of this site
Licence
MIT (class A). File: LICENSE. Read on 15 November 2026. Commit 2699bcf3ef819d2a3fbf1a52228bd87b05522435.
Skills
15
Last push
15 November 2026
Archived
No
Read depth
Full read
Last verified
15 November 2026
Install
npx skills add ua5er/user-acquisition-skills

Safety

Writes to
Ad account
Approval step before a write
No
Data leaves the machine
Nothing
Scripts in the repository
5
scan-skill.py run
15 November 2026, commit 2699bcf3ef819d2a3fbf1a52228bd87b05522435
Notes
Turns a media buyer's weekly judgement calls into written rules and audits an agent can apply to exports: when to kill or scale an ad, how much a Meta launch needs to exit learning, why an account is stuck in Learning limited, whether CAC pays back, and which attribution number to trust.

Skills inside

The 16 skills in ua5er/user-acquisition-skills
SkillWhat it doesJob
ua-kill-scale-rulesWrites kill, scale and budget pacing rules for Meta and TikTok from your CAC target, then applies them to a campaign export.Decide which ad sets to stop, scale or leave alone.
paywall-auditAudits a subscription paywall or checkout for message match, proof, price display, disclosure and CTA.Get a ranked change list and a test queue for the paywall.
onboarding-web2app-funnelProduces distinct onboarding or web-to-app concepts with hypotheses, a measurement plan and a test order.Choose what the flow shows between the ad click and the paywall.
unit-economics-checkChecks CAC, cohort LTV and payback from your own data and sets a safe CAC ceiling.Know what CAC you can afford to pay.
attribution-sanity-checkReconciles ad platform, MMP, SKAN and backend payment data and lists the fixes that matter.Find out which acquisition numbers you can trust.
creative-testing-researchPlans a creative testing cycle and researches competitor ads from public ad libraries only.Plan the next round of creative tests.
weekly-ua-reportBuilds a one-page weekly acquisition report that ends in decisions.Turn the week's export into next week's actions.
ai-marketing-system-auditMaps a marketing system as a loop (acquire, convert, measure, automate, intelligence, decide) and ranks the gaps.Find the weakest link in the system and what to build first.
meta-ads-learning-auditAudits a Meta ad account stuck in Learning limited: event volume against the budget learning needs, split signal, test CPMs, funnel and CAPI, plus a post-mortem of past campaigns.Find out why the account never leaves learning and what past tests really cost.
meta-ads-launch-plannerPlans a Meta sales launch that exits learning: budget from target CPA x 50 events a week, one broad ad set on CBO, day-by-day rules, automated stop and restart rules and a scaling schedule.Launch and scale without resetting learning, with a plan management can sign off.
meta-creative-testingDesigns creative testing for Meta after Andromeda: which past ads to relaunch, three test structures with the cost of an answer, winner and kill criteria, and briefs for designers and UGC creators.Test creatives so each answer is worth what it costs.
jev-marketing-gatesTurns one marketing judgment into a typed decision gate with Jev, TypeSafe AI's model that answers yes/no, choice and score questions with probabilities: narrow questions, thresholds in code, a batch runner and a calibration report.Check ad copy, comments or creative tags at scale, with a person approving anything that spends money.
competitor-creative-parserClaude + Jev, step one. Claude describes each competitor ad from a public ad library, Jev tags it with a taxonomy drafted for that market, and code builds a snapshot of what the category runs, split into static and video. Run on 54 real ads: 1,371 typed answers for under one cent.See what a category really runs and where nobody stands, before writing a brief.
market-to-creative-briefClaude + Jev, step two. Turns the market snapshot into production decisions and two brief packs: one for static designers and one for the motion team with three hooks per concept. Code re-checks every number a decision cites against the snapshot.Hand designers and motion designers briefs that are tied to market numbers.
creative-preflight-panelClaude + Jev, step three. Shows every concept to twelve or more buyer profiles through Jev before production, builds a concept by profile heat map and returns go, rework or drop with reasons. A calibration script compares the panel with real results after launch.Cut weak concepts before the team spends hours on them. The live test still decides.
synthetic-audience-panelClaude + Jev. Shows one ad or two to about 1,000 virtual people in a few requests: every combination of seven situations and attitudes, with a control group. Returns a funnel per 1,000 virtual impressions, the main objections and a verdict for two ads on the same panel.Compare two ads on one panel before the live test. Panel answers, not a forecast of CTR.

Operator note

This pack reads like a media buyer's rule sheet written down. The kill and scale skill makes you state a spend gate, a hard cap and a kill line from your own target CAC before any ad gets judged, and it refuses verdicts below the gate. The Meta pair covers the two questions that come up after a failed launch: was there enough event volume to learn, and how much budget does the next launch need. Use it on exports you already have; it needs no API token. Two weak spots: some numbers in the Meta skills come from one account and read as general rules, and two creative testing skills recommend different budget structures without saying which applies when. Rule precedence is also missing, so an ad that trips both a kill rule and a landing page check gets no single answer. Fix those before you lean on it for client work. It is first party, written by the site owner, and marked as such.

Test note

Draft, not yet reviewed by the site owner. Run on 2026-11-15 in Claude Code with the ua-kill-scale-rules skill.

Input: Sample data, invented for this test: a 7-day Meta export for a subscription app with six ads. Target cost per payment 40 EUR. Per ad: spend, payments, trials, last-3-day spend and payments, CTR and frequency trend, click to trial rate. Account median click to trial 2.5 percent. Platform payments 26, backend payments 24. Monthly budget 6,000 EUR, day 15 of 30, 2,700 EUR spent.

Result: Following the method gave a gate of 80 EUR (2x CAC), a hard cap of 160 (4x) and a kill line of 52 (1.3x) over the last 3 days. Action table: ad A1 (420 spend, 12 payments, last 3 days 190 / 6 = 31.7) scale one 20 percent step; A2 (175 spend, 0 payments) hard stop; A3 (65 spend) below the gate, not enough data; A4 (last 3 days 120 / 2 = 60) soft kill, medium confidence on 2 payments; A5 (44.3 over 7 days, frequency 2.1 to 3.4, CTR 1.6 to 1.0 percent, CPM flat) fatigue rule, swap the creative and keep the ad set; A6 (CTR normal, click to trial 0.4 percent) quality rule, check the store page and tracking. Pacing: 3,300 left over 15 days is 220 a day; the week ran at 197 a day, about 90 percent of plan, so no flag. Tracking rule did not fire.

What worked: The spend gate stops verdicts on thin data, every threshold is tied to target CAC, and the confidence column made the A4 call honest. The blended cost per payment (1,380 / 26 = 53.1) sat above target while A1 qualified to scale, and the marginal CAC rule in Step 4 caught that tension.

What broke or needed fixing: A6 fired both the soft kill (75 per payment) and the quality rule. The skill says to run rules in order but not whether the first match wins, so it gives no answer on pausing an ad whose problem is the landing page. Nothing covers the band between target and the kill line beyond an implicit keep. 'Several days' in the fatigue rule needs a number.

Use cases

  • Paid UA agents: Paste a 7-day ad-level export and your target cost per payment; get a rule sheet plus an action per ad (kill, scale, keep, not enough data) with a confidence level.
  • Analytics: Give platform purchases, MMP numbers and Stripe or RevenueCat payers for one period; get a reconciliation table and which number drives budget moves.

What to check before you install

  • Open meta-ads-launch-planner section 7. With ad account access, the agent can publish its own drafts, and needsApproval is false. Add a preview and an owner yes before 'Review and publish'.
  • Remove the 'antidetect profile' option from meta-ads-learning-audit, since antidetect browsers on ad accounts can conflict with platform terms.
  • Replace the unsourced 1.5 to 2.5x CPM claim and the single-account '11 creatives at $150 each' figures with numbers from your own account.
  • Decide which budget structure applies, because creative-testing-research uses equal budget per creative and meta-creative-testing defaults to one broad CBO ad set.
  • Read budget.py and validate.py yourself, since neither was executed.

Alternatives

  • AgriciDaniel/claude-ads has 34 skills in a read-only audit system across twelve ad platforms, while this pack writes kill and scale rules and Meta launch checks.
  • autonnel/autonnel-skills has 6 skills for work after the ad click, such as landing page audits and server-side conversion tracking, and self-hosted-funnel-launch needs an Autonnel instance.
  • adkit/ads-skills has 11 ads skills built around a brief.md flow, which is a different structure from the rule and audit approach here, and its access was not reviewed.

Sources

Each source shows the date it was last checked.

  1. GitHub ua5er/user-acquisition-skills
  2. GitHub ua5er/user-acquisition-skills: LICENSE

Need creative testing and AI tooling tied to your app's cost per install? Start with a short audit.

See the service