thedavidquan01/claude-paid-ad-skills

Eleven free Claude skills that put an AI layer on the CreativeOS from Part 1. They run the weekly loop of build, measure, and learn in your brand voice.
They work in Claude.ai, Claude Code, and Cursor. Each skill is a folder with a SKILL.md file in the Agent Skills format. Claude reads the description, then loads the skill when your request fits.
This repo is Part 2 of the Blackbird Signal Session on paid ads. Part 1 covered creative systems. Part 2 runs that system with Claude.
## Where Part 1 left off
The CreativeOS has two jobs and one loop.
Job 1 is to find a winner. Does Meta work for us?
Job 2 is to scale it. Can we spend more at the same cost per customer?
Both jobs run the same weekly loop: BUILD a hypothesis, MEASURE by counting winners, LEARN by asking why and changing one thing.
Part 1 ended on slide 24, "Part 2: an AI layer on the OS". It promised three things. This repo ships them.
| Part 1 idea | Skill that runs it |
| Your customers already wrote the grid | customer-grid |
| Grid: who down the side, why they buy across, every cell one concept | customer-grid |
| Start where the money already is | customer-grid picks the week-one cell |
| Competitors: their ads, and the angle nobody runs yet | competitor-ad-teardown |
| Awareness stages, and the page repeats what the ad promised | creative-brief |
| One brief, four versions that each change one thing | creative-brief |
| First line in their words | graded-ad-copy |
| Make the statics | higgsfield-static-ads, then visual-qa |
| Count what you learn, not what you launch | ad-performance-readout |
| Judge on CAC, not clicks | ad-performance-readout |
| Fund only what you can judge | ad-performance-readout budget calculator |
| Under 50 a week, Meta is guessing | ad-performance-readout event check |
| Scale in steps | ad-performance-readout and next-cell |
| When one wins, change one thing | next-cell |
| Pick the next cell | next-cell |
| Naming convention | references/naming-convention.md in creative-brief, ad-performance-readout, and next-cell |
## The AI layer: build, measure, learn
Run the two setup skills once. Then run the loop every week.
| Step | Skill | What it does |
| Setup | brand-voice-guide | Builds a one-page voice guide from your real ads, pages, calls, and reviews. Ends with a blind test and a voice score. |
| Setup | visual-guide | Turns screenshots or a Figma export into a DESIGN.md plus a prompt block for image tools. |
| BUILD | customer-grid | Mines what buyers actually say into a who x why grid with verbatim quotes and sources. Picks the week-one cell. |
| BUILD | competitor-ad-teardown | Maps competitor ads to grid cells and awareness stages. Finds the open cells and the longest-running ads. |
| BUILD | creative-brief | Turns one cell into one brief with V1 original, V2 new hook, V3 new creator, and V4 new setting. Names every ad. |
| BUILD | graded-ad-copy | Writes 8 variants for the cell, scores them, checks claims, and gives you the V1 and V2 first lines. |
| BUILD | ad-remix | Optional. Rebuilds one competitor ad's structure in your brand, then checks it is not a copy. |
| BUILD | higgsfield-static-ads | Optional. Makes one static per version in Higgsfield Ads Studio. Quotes the credit cost and waits for your yes. Needs a paid Higgsfield plan. |